3.1 Data sources and filtering criteria

This study requires the data on air quality and financial data of bonds and issuers in China. Wind Information Co. Ltd (Wind) provides the requisite financial data for bonds and issuers, including basic issuer-level characteristics, such as tangible asset ratio, age of the issuer, leverage level, sales, profitability, revenue growth and cash holding. Wind also provides bond-level data, including issuance size, credit ratings, maturity, the outstanding debt amount, offering spread, transaction data and yield-to-maturity information.

Our sample of median and long-term corporate bonds spans a 13.5-year period from January 2007 to May 2020.6 These debt instruments include enterprise bonds, non-financial medium-term notes and corporate bonds, which results in 17,826 new issues of corporate bonds. Next, we remove bonds or issuers without information on identifiers or city-level air quality index, as well as those with more than one coupon payment per year (Livingston et al., 2018). Our sample coverage is extensive, whereby the primary bond market comprises 12,794 bonds from 3019 unique issuers. Merging the data in the primary bond market with bonds with valid data of trading information gives rise to a sample of the secondary market. In terms of trading information, we collect monthly closing and weighted average trading prices, as well as yield to maturity (YTM). We only retain a bond-month observation if its maturity is more than 1 month, the average daily trading price is between CNY 5 and CNY 130, and it has valid trading data and financial data to construct control variables. The above procedures give rise to a secondary bond market sample with 152,160 bond-month observations, which are associated with 10,411 bonds by 2787 issuers.

Similar to Dong et al. (2021), we collect the Air Quality Index (AQI) from the Ministry of Ecology and Environment of the People's Republic of China (MEEPC), which is summarised from the daily reports of air pollution by local environmental protection bureaus in every city.7 AQI is estimated as the combination of the values of six atmospheric pollutants, including nitrogen dioxide (NO₂), sulfur dioxide (SO₂), suspended particulates smaller than 2.5 μm in aerodynamic diameter (PM2.5), carbon monoxide (CO), suspended particulates smaller than 10 μm in aerodynamic diameter (PM10) and ozone (O₃). Before 2014, AQI did not exist. The available alternative is the Air Pollution Index (API), which is based on PM10, NO₂, and SO₂.8 The API and AQI are strongly correlated (Zheng et al., 2014). For the period prior to 2014, we use the API to measure air quality.

We compute the natural logarithm of the mean of AQI/API for each city to measure the local air quality of each bond issuer on a monthly basis, denoted as Log_AQI. The MEEPC provides six categories of air quality: I, indicating ‘excellent’ (AQI ≤ 50); II, representing ‘good’ (50 < AQI ≤ 100); III, corresponding with ‘lightly polluted’ (100 < AQI ≤ 150); IV, indicating ‘moderately polluted’ (150 < AQI ≤ 200); V, suggesting ‘heavily polluted’ (200 < AQI ≤ 300); and VI, indicating ‘severely polluted’ (AQI > 300). These categories are employed consistently in the periods before and after 2014.

3.2 Variables

Bond yield spread at issue (Offering spread) is our key measure of cost of debt for the primary bond market, where a bond is issued for the first time. In terms of the secondary bond market, the yield-to-maturity spread (YTM spread) is employed. To estimate the YTM spread, we employ a comparable Treasury note based on maturity as the benchmark. All else equal, the credit risk of the underlying bond security is captured by the YTM spread.

We construct three alternative sets of spreads for sensitivity checks. The benchmarks in constructing Offering spread (CDB) and YTM spread (CDB) rely on similar bond instruments issued by the China Development Bank (CDB). Second, Offering spread (Synthetic) and YTM spread (Synthetic) are estimated with reference to a synthetic yield (Ang et al., 2023). First, the zero-coupon Treasury bond yield curve is calibrated based on the exponential functional form of Svensson (1994). Next, the implied price of the synthetic Treasury bond with identical coupon rate, frequency of coupon payment and maturity is estimated as the present value of each cash flow with the Treasury bond zero-coupon rates as the discount rates. Matched Treasury bond yield is computed based on its implied price. Subsequently, YTM spread (Avg) replaces the YTM spread, using the average trading volume-weighted price in the calculation.

We construct a raft of control variables to capture various features of bonds and issuers following Livingston et al. (2018). The important issue-level variables include the issue size/bond outstanding value measured by the par value of a bond in CNY 1 billion (ISize); credit rating represented by rating dummies, ranging from Below AA−, AA−, AA to AA+; the puttable, callable, or sinking fund feature; and years to maturity (Maturity). To account for the unique regulatory environments, dummy variables are assigned for both corporate and enterprise bonds (Corporate and Enterprise). The cross-listing dummy (Cross) identifies bonds traded in both the interbank and exchange-based markets. To control for firm-level variations, we compute a series of financial variables, including asset tangibility (Tangibility), cash holding to represent asset liquidity (cash and cash equivalents divided by current liability, Cash), earnings (operating income divided by average total assets, ROA), public companies (Public), state-owned enterprises (SOE), Leverage (total liability divided by total assets), the age of the issuer (Age), Sales (log of sales in billions of CNY) and Growth (annual growth rate in operating revenue) to proxy for firm profitability. The estimation approaches are articulated in Appendix 1.

3.3 Summary statistics

Table 1 provides a comprehensive overview of air quality across various cities and provinces in China. In Panel A, we present data on the 30 cities with both the highest and lowest Air Quality Index (AQI), along with their respective provinces, alongside the ranking of provinces based on their AQI during the sample period. Notably, the cities with the most severe air pollution are predominantly situated in Henan, Hebei and Shandong provinces, while provinces like Hainan, Yunnan, Guizhou and Sichuan exhibit comparatively better air quality. In general, the majority of provinces experience mediocre air quality, with AQI levels hovering around 90, although only one province consistently maintains excellent air quality, with an average AQI of 50 or less.

Figure 2 further elucidates the trends in air quality by depicting the average AQI values across all provinces in China and specifically highlights the average AQI values of Beijing, the capital city of China, over time. The data indicate a concerning trend of deteriorating air quality over the observed period. In Panel b, we shift focus to the primary market for bond issuance, highlighting the 10 leading cities in this aspect. Cities such as Beijing, Shanghai, Shenzhen and Guangzhou emerge as prominent players in bond issuance, reflecting their economic significance and financial activity.

Table 2 illustrates the descriptive statistics of the key variables in our empirical analysis. The mean and median offering yields are approximately 200 bps above their maturity-matched Treasury yields in the primary market. Among the 12,794 bonds issued in the primary market, the offering yield spread has a mean (median) of 2.071% (1.830%). Bond offering yield differs from the yield on a comparable bond issued by the CDB based on maturity at 1.519% on average. The approach of Ang et al. (2023) gives rise to a 2.102% distinction between the offering yield and the yield on a synthetic government bond. This is consistent with prior work in China that the yields and spreads are lower than corporate bonds in the US market (Hu et al., 2020; Walker et al., 2021). The majority of the bonds are rated AA or above (99.1%), of which 50.1% are rated AAA, followed by 26.3% with AA+ and 22.7% rated as AA. The average maturity is 4.82 years, with an average issue size of CNY 1.424 billion ($US204.545 million). Issuers in the study have an average age of 18 years. Public issuers account for 18.9% of the bonds, whereas 85.1% are issued by SOEs. The average debt ratio is 61.2%.

Our analysis extends to the secondary bond markets, where we observe comparable distributions to those in the primary market. Specifically, the average monthly YTM spread stands at 2.344%. A significant 89% of our observations boast ratings above AA, indicating a generally high level of creditworthiness within the market. The average maturity of bonds traded in the secondary market is approximately 3.324 years, accompanied by a substantial bond outstanding volume totalling CNY 1.954 billion (equivalent to $US280.675 million).

Interestingly, we find that the average air quality index (AQI) remains consistent across both primary and secondary market samples, hovering around 79.043. However, there exists a notable degree of variation within this metric, reflecting the diverse environmental conditions across different regions where bond trading occurs.

4.1 Baseline results

Table 3 demonstrates our baseline results. Columns (1)–(4) ((5)–(8)) represent the primary (secondary) bond market. ${\beta }_1$ captures how yield spreads relate to air pollution. The coefficient of Log_AQI is 0.214 (0.137) in column (1) ((5)), which is statistically distinguishable from zero. Replacing firm fixed effects with industry fixed effects in columns (2) and (6), the positive relation between air pollution and yield spreads remains intact.

Based on the model estimation in columns (1) and (5), when AQI increases by 100, bond yields increase by 99 and 63 bp in the primary and secondary markets, respectively.9 This represents an additional cost of CNY 12.34–14.10 million ($US1.77–2.02 million) per year to issue an average-sized bond. We further employ a dummy variable denoted as AQI (dummy), which equals one if Log_AQI is above the sample median and zero otherwise. Bonds issued by firms from severely polluted areas are subject to an additional yield spread of 7.4 (4.2) bps in comparison with their counterparts from the less polluted areas in the primary (secondary) bond market (columns (3) and (7)).

Furthermore, as indicated in Panel B of Table 1, firms from Beijing, Shanghai, Shenzhen and Guangzhou dominate corporate bond market issuance (35% of the issuance frequency). As such, we exclude these firms in estimating the regression in columns (4) and (8), to alleviate a potential concern that issuers from the above areas drive our findings. The positive relation between AQI and bond yields continues to hold. Further, the relation between other variables and bond yield spreads are largely in line with the extant literature. To illustrate, yield spreads decrease with bond credit ratings (Livingston et al., 2018). We also document lower yield spreads among publicly listed companies, SOEs, mature firms, larger and more profitable issuers.

4.2 Endogeneity and robustness checks

Our results indicate an air pollution premium in the corporate bond sector in China, whereby bonds issued in areas with poor air quality are associated with higher costs, which are both economically and statistically significant. A raft of robustness checks are implemented to address issues arising from the fundamental differences in cities, endogeneity, unobservable factors, macroeconomic conditions of Chinese cities, differences across industries and alternative measures of bond yields.

5.1 Trading frictions

Previous studies finds that investors become pessimistic with worsened air quality, thus depressing asset value (Ding et al., 2021; Levy & Yagil, 2011), reducing trading intensity (Gervais & Odean, 2001) and market liquidity (Baker & Stein, 2004). Even though the majority of the bond investors are institutions, fund managers are also found to exhibit certain mood-driven behavioural biases (e.g., Itzkowitz & Itzkowitz, 2017; Shiller, 2000). This implies that air pollution may reduce trading intensity in the market, thus increasing illiquidity. As such, investors may command higher yield spreads as compensation for holding bonds from highly polluted areas. To explore this factor empirically, we regress trading volume on Log_AQI, along with control variables at both the primary and secondary levels.

Specifically, Trading volume is the first-month bond transaction volume in billions of CNY for the primary bond market, and the monthly transaction volume in CNY billion for the secondary bond market, respectively. Table 4 presents the relationship between bond trading volume and local air quality. If the air pollution premium is attributable to investor mood swing and the associated reduction in trading intensity, AQI and trading volume are expected to be negatively related. However, the slope coefficients on trading volume are statistically indistinguishable from zero. The evidence suggests that the air pollution premium is not associated with trading frictions in the Chinese bond market.

5.2 Information asymmetry

We further explore the role of information asymmetry of the issuers. Information asymmetry is more prevalent in firms with weaker governance (Masulis & Reza, 2015). Issuers with weaker governance and disclosure regimes may concern investors, who demand a higher cost of debt. It is documented that air pollution affects the mental health and cognitive ability of financial market participants. For instance, air pollution is likely to induce mental breakdowns, immoral activities and increase the likelihood of suicides (Bakian et al., 2015; Lim et al., 2012). The detrimental impact of air pollution on mental health potentially gives rise to errors in making corporate decision and dampens the ability to analyse information (Ackert et al., 2003). Air pollution may exert detrimental influence on the information environment of firms in various ways. First, air pollution directly affects local labour. Wu et al. (2022) find that air pollution is associated with a greater degree of accounting conservatism. Xue et al. (2021) document significant loss of skilled labour driven by air pollution. The adverse impact of air pollution on local employees induces investors' concerns over the accuracy and reliability of firms' disclosures, thus exacerbating information asymmetry. Second, heavily polluted areas are potentially subject to heightened regulatory enforcement risks (Seltzer et al., 2022), which elevates business uncertainty and increases the complexity for investors to analyse firms from polluted areas. Poor air quality potentially has a negative influence on the local information environment, which induces investors to require higher yields to invest in bonds from polluted areas.

The listing status of firms and firm size are employed as proxies for their level of information asymmetry, which are available for both public and private issuers. Idiosyncratic volatility (IVOL) is adopted as the third proxy for information asymmetry, which requires stock return data and is only available for public issuers. If the air pollution premium is associated with worsened information environment, it should be more pronounced in companies that are susceptible to information asymmetry. First, we partition our sample into public versus private firms based on their listing status. We next analyse the role of firm size measured by its sales level, as a widely used proxy for information asymmetry (Freeman, 1987). Publicly listed firms and large firms tend to have more readily available information compared to private firms and small firms. Idiosyncratic volatility is a popular measure of asymmetric information, which captures significant fluctuations in stock price movements due to trading by informed traders based on private information (Moeller et al., 2007). We calculate IVOL for public firms by employing the market model using daily return over a rolling 252-day period and taking the standard deviation of the residuals in the regressions.

In Panel A of Table 5, columns (1) and (5) ((2) and (6)) display the regression output for public (private) companies. Columns (3) and (7) ((4) and (8)) illustrate the results for large (small) companies, of which Sales are higher (lower) than the sample median. Bonds issued by firms with lower information asymmetry (public and large firms) are associated with a lower loading on the air pollution premium. Nonetheless, we observe positive coefficients for all identifications, which is in line with our baseline results.

Taking the primary market as an example, the coefficient of public companies is 0.053, while the number is 0.243 for private firms. Economically, this result indicates that yield spread will increase by CNY 3.476 million for public listed firms and CNY 15.935 million for private firms per year for an average size bond issue, when the air quality index increases by 100. The CNY 12.459 million cost difference is statistically and economically significant. The air pollution premium is more prevalent in small firms. The analysis in Panel B is restricted to public issuers, as the calculation of IVOL requires stock return data. Results are overall in tandem with Panel A, whereby the air pollution premium attenuates among public issuers subject to poor information environment in the secondary market. As such, information asymmetry is likely a relevant factor in understanding the air pollution premium. These results also indicate that more transparent corporate disclosure potentially mitigates the detrimental influence of local air pollution on debt financing costs.

5.3 Awareness and attention to climate change

We also explore the possibility that the higher borrowing costs in Chinese bond markets are associated with growing investor awareness and attention to climate change risk. Krueger et al. (2020) administer a survey to explore the climate risk management strategies among active investment managers. Survey results indicate that investors are aware of the threats brought by climate-related financial risks. As such, they also believe that climate risks should become important determining factors in investment decision making and portfolio management. The evidence echoes Alok et al. (2020), who show that portfolio adjustments are made by fund managers in times of climate disasters.

Three salient events have potentially considerably shifted investors' perception of environmental risk and elevated their awareness in recent years. The first event is the Paris Agreement on 22 April 2016, whereby 196 countries agreed to limit global warming to below 2°C.11 The Paris Agreement represents a significant milestone in the multilateral management of the climate crisis. It is the first time that all nations are united by a binding agreement for a common cause to combat climate change and the associated impact. We anticipate a notable change in investors' perception of climate risk after the Paris Agreement. As such, a more profound positive relationship between local air quality and yield spread is likely to prevail.

Parallel with the Paris Agreement, the second salient event is the 18th National Congress of the Chinese Communist Party (CCP) in November 2012. A key outcome of this event was that the construction of ecological civilisation as a major goal of the party was written into the party constitution. This is a significant extension of the preceding constitution, which included cultural, economic, social and political realms. This purportedly increases the attention and resources dedicated by the CCP to mitigate climate change risk. As such, we expect a more pronounced air pollution premium following the National Congress.

Our third significant shock to investor attention to air pollution is when the average monthly AQI across all cities exceeded 100 for the first time. This occurred in January 2013 and is widely regarded as the milestone when China's struggle with air quality became mainstream in both China and around the world.12

To explore the effect of these exogenous shocks on investors' perception of climate-related financial risk, Equation (1) is augmented with a Post Paris dummy variable for the Paris Agreement, a Post 18th dummy variable to reflect the National Congress and a Post 2013 dummy variable for the period after the first time when extremely severe air pollution was recorded.

Specifically, the dummy variables equal one if the observations are after May 2016, December 2012 or January 2013, respectively. Interaction terms between Log_AQI and the three event dummies are incorporated to ascertain the influence of investors' attention to climate change on the air pollution premium. We use a 2-year window before and after each event.13 The interactions between Log_AQI and Post 18th are positively significant, implying reduction in investor tolerance on climate risk after the reform. A similar pattern is documented in the interaction terms between Log_AQI and Post 2013 in both markets, in columns (3) and (6) of Table 6.

To further explore investors' attention to and perception of air pollution, we collect the daily Baidu search index data on the key word ‘air pollution’ (空气污染) at the city level and the national level.14 Next, we aggregate the daily Baidu indexes within a month, which gives rise to a series of monthly investor attention and sentiment measures concerning air pollution in China. We define Baidu Trend (City) and Baidu Trend (Country) as the total value of the Baidu search index on ‘air pollution’ every month at the city level and country level, respectively. While AQI reflects the severity of air pollution, the Baidu search index on ‘air pollution’ is likely to complement the actual air quality by capturing the perceived air pollution and concerns over air quality. If the air pollution premium is associated with rising investor attention and awareness to air pollution, bond yields are expected to increase with the Baidu Trend measure of air pollution. We regress measures of yield spreads in both markets on both Baidu Trend (City) and Baidu Trend (Country) and present the regression output in Appendix 7. In line with our conjecture, yield spreads load positively on the investor attention measure using the Baidu search index. For instance, the slope coefficient is 0.012 and significant at the 5% level. This indicates that bonds are more costly when investors pay a greater amount of attention to worsened air quality. The results are insensitive to the use of the city level of the national Baidu research index on ‘air pollution’.

5.4 Default risk and air pollution

We further investigate whether local air pollution affects the fundamental operations and default risks of companies headquartered in polluted cities, thus leading to differences in bond pricing. Air quality may exert adverse influence on firm operations in various ways. For instance, worsened air quality presents a hazard for local employees' physical and mental health, which leads to departure of local talent. As a consequence, firm performance and productivity drop significantly (Xue et al., 2021). Alternatively, firms may resort to increase employee compensation to retain human capital. As another example, air pollution may compel local governments to shut down coal-fired thermal power plants, resulting in industrial electricity shortages or increase in electricity prices that directly impact the operating capacity and costs of companies. Taken together, it is likely that air pollution may increase operating costs, reduce operating cash flows and thus increase default risk of companies located in polluted areas.

Table 7 illustrates the regression results where we regress the aforementioned proxies for firm operations on the level of AQI and a dummy variable for AQI when it is higher than the sample median. Across various model identifications, there is no evidence that air pollution worsens the fundamental operations for issuers, including credit risks, operating costs and operating cash flow. As such, the air pollution premium is unlikely to be associated with the explanation associated with default risk.

5.5 CRAs and air pollution

Our findings thus far document that investors demand higher yield spreads to lend to firms from more polluted areas. This is associated with rising investor awareness of climate-related financial risks, who are important participants in financial markets. In the bond markets, credit rating agencies (CRAs) constitute another vital group of participants. CRAs may incorporate the environmental risk of bonds issued in highly polluted area into credit ratings, which may lead to higher costs of debts. The air pollution premium is potentially a reflection of the changes to credit ratings to incorporate climate-related financial risks. To explore this possibility, we employ both the ordered logit model and logit model as shown in Table 8, by regressing Bond credit rating on Log_AQI in columns (1) and (2) and columns (5) and (6) for primary and secondary markets in Panel A of Table 8, respectively. The Low bond rating dummy is utilised as an alternative measure in columns (3), (4), (7) and (8). If the air pollution premium reflects the information content of air quality in credit ratings, there should be a negative relation between credit ratings and air quality.

Panel A of Table 8 shows that with or without controlling for the CRA fixed effects, there is no apparent relationship between bond credit rating and air quality during the bond issuance period in the primary market and subsequent trading in the secondary market. As such, Chinese CRAs do not appear to incorporate air pollution in issuing credit ratings. The air pollution premium is not attributable to updates to credit rating to incorporate environmental risks related to air quality. We also incorporate the three salient events that relate to heightened attention to environmental issues in China and display the regression output in Panel B of Table 8. The interaction terms between the event dummy with a 2-year event window surrounding each event and air quality are statistically indistinguishable from zero. This implies that the findings in Panel A do not vary with the level of attention to environmental concerns.

The preceding findings provide a nuanced perspective that challenges the prevailing anecdotal narrative suggesting CRAs' belated response and tepid attention to climate-related financial risks. Contrary to this notion, our analysis reveals a more nuanced landscape. For instance, Moody's, a prominent global rating agency, only acquired a data vendor specialising in climate-related issues in July 2019, following its integration of environmental, social and governance (ESG) factors into risk assessment methodologies in late 2018. This timing underscores Moody's recognition of the evolving significance of climate-related risks in financial markets. Moreover, warnings from institutions such as the Center for International Environmental Law (2015) highlight the potential systemic ramifications if CRAs fail to incorporate climate change risks into their evaluations. Only very recently, the largest CRA in China, Chengxin_Moody announced that it plans to take ESG factors into consideration when rating corporate bond issuers. Our finding that air pollution does not affect credit rating is in line with our result in Section 5.3, whereby air pollution does not cause a dramatic impact on the default risk of firms. Our results suggest a nuanced picture wherein CRAs are increasingly recognising the importance of environmental factors in their evaluation frameworks, albeit with some lag. However, their assessments seem to prioritise fundamental business performance over short-term environmental fluctuations, as evidenced by the relatively stable credit ratings despite fluctuations in air pollution levels.