Is there a green bond premium and does accounting for COVID-19 epidemic matter?

(1)

1

Is there a green bond premium and does

accounting for COVID-19 epi

demic

matter?

A comparative study of yield and spread of green bonds and

ordinary bonds in the US corporate bonds market during the

normal period and COVID-19 period respectively.

Zimo Li 11839260 Bachelor Thesis

BSc in Business Administration and Specialization in Finance University of Amsterdam

Supervisor: Mr. Sjoerd van den Hauwe Word Count: 8444

(2)

2

Statement of Originality

This document is written by Student Zimo Li who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

Inspired by the rapid development in green bonds in the corporate market and the sudden outbreak of the COVID-19 crisis, the paper investigates the impact of

COVID-19 on corporate yield spread and the presence of green bond premium in normal period and during the COVID-19 crises for the US corporate bonds market. The results suggest a rapid increase in yield spread since the outbreak of COVID-19 and through both the nearest neighbor matching algorithm and regression with fixed effects methods, no green bond premium is found in either normal period or the COVID-19 crisis. However, the non-existence of green bond premium does not mean the absence of benefits of green bonds. The signaling effect has been observed in several studies and there could be long-run benefits for companies issuing green bonds. Since the market for the green bond has not been matured and the COVID-19 crisis is ongoing, more future studies on the corporate green bond premium and the impact of COVID-19 on the green bond are needed.

(4)

4

1. Introduction:

In 2007, the European Investment Bank (EIB) issued the world’s first green bond under the label of Climate Awareness bond to provide funding for renewable energy projects (EIB,2020). That marks the start of green bonds, which primarily fund projects that have positive environmental and climate benefits. According to Climate Bond Initiative (2020), the market has grown rapidly since 2014 and reached a new global USD257.7 billion value at issuance in 2019, up by 51% compared to the value of issuance in 2018. There have been 1788 green bonds from 496 issuers in 2019, while there were only 73 issuers back in 2014. According to Martin and Moser (2016), both investors and managers respond positively to green investment

independent of the return of the green projects. In theory, green bonds are identical to ordinary bonds in all other ways but the proceeds of green bonds are invested in green projects. As investors investing in green bonds would be willing to give up parts of their gains for societal benefits, green bonds should enjoy a green bond premium in comparison with ordinary bonds with the same credit rating and level of risk

(Friedman & Heinle, 2016). However, the theory may not represent reality. The existence of a green bond premium has been studied by several papers by different methods while mixed results are obtained. Hence, in this paper, the analysis will be based on the latest data to review the presence of green bond premium in the US corporate bonds market.

On the other hand, since the start of 2020, the COVID-19 epidemic has been a tremendous crisis for the whole world. According to the World Health Organization (WHO,2020), up to June 22, 2020, there have been more than 8.8 million confirmed cases of COVID-19 globally and more than 450,000 people have been dead. What is even more worrisome is that the number of cases continues to rocket and shows no sign of control. António Guterres, the UN secretary-general has described COVID-19 as the largest challenge for the world since COVID-19 (United Nations, 2020). Along with the largest public health crisis in a generation, the world economy has also been severely hit. According to US Bureau of Labor Statistics (2020), the US

(6)

6

unemployment rate reached to an unprecedentedly record of 14.7 percent in April 2020 and China’s economy shrank 6.8 percent in the first quarter of 2020 compared to that of last year, marking the first quarterly contraction of China since 1992 (National Bureau of Statistics of China, 2020). The World Trade Organization (WTO,2020) expected the world trade to decline by 13% to 32% and International Monetary Fund (IMF,2020) estimated that the contraction of the global economy by 3% on the assumption that the COVID-19’s impact will be alleviated in the second half of the year. The Crude WTI Crude Oil Futures also went down to a historical negative on April 20th because of the weak global demand. Almost every aspect of the daily

activities has been influenced by the COVID-19 crisis both financially and physically. In the past, financial crises have created tremendous volatility on the bond market. According to the research by Schuknecht, Hagen, & Wolswijk (2010), the yield spread for the non-benchmark bond has been largely increased significantly during the subprime crisis because of the investors’ increased risk aversion. This is also supported by the study by Guidolin and Tam (2010) that during the subprime crisis, the yield spread peaked following the default of Lehman’s brothers in September 2008 and only stabilized back to normal level back to 2010. However, the past papers only focus on the impact of financial crises on fixed income instruments in general, there has not been specific research on the impact of the financial crisis on green bond, especially the green bond premium. Hence, it is very interesting to discover how financial crises can influence the green bond premium.

In the following session, the paper first tests about the change in the yield spread in the US corporate bonds market from the normal period (January 1st, 2016 to

January 31st, 2020) to the COVID-19 period (February 1st,2020 to June 15th, 2020) by the regression of yield spread on a crisis dummy and controlled variables. Next, the greedy nearest neighbor matching method suggested by Larcker and Watts (2019) is performed to match each green bond with an ordinary bond with the most similar covariates. Separate paired t-tests on yield and spread on matched pairs issued before COVID-19 and during COVID-19 are conducted to test for the presence of green bond premium before and during the crisis. Although the primary method for

(7)

7

investigating the green bond premium is the greedy matching method, it is still meaningful to reconcile the regression with fixed effects method brought by Baker et al. (2018). Hence, the regression of yield on the green dummy and several covariates with fixed effects is also performed separately for the bonds issued in the normal period and the COVID-19 period in this paper.

In the result session, the significant and large increase in yield spread is identified during the crisis, reflecting the high volatility in the market and risk aversion from the investors. There is no presence of a green bond premium for both bonds issued before the crisis and bonds issued during the crisis via both matching method and regression method with fixed effects. One of the potential reasons for the missing green bond premium could be that investors are not willing to lower their gains solely for the benefits of green projects. However, that does not mean green bonds bring no extra value to the companies. In the discussion session, the signaling effect of green bonds on share price by Flammer’s paper is discussed, which suggests that green projects can lead to an increase in share price and bring long-run benefits.

Since COVID-19 epidemic has profound impacts on almost every aspect of the economy and there has not been any research about the impact of the crisis on green bond premium, the biggest contribution of this paper is to unveil the initial impact of COVID-19 on green bond premium and to inspire more future researches into this topic. Nonetheless, since COVID-19 is still evolving rapidly with no sign of ending and this paper only captured the data on the early months of the outbreak, the comprehensive impact of COVID-19 on the green bond premium would only be known in the future. Besides, since corporate green bond market is in rapid development, this paper can contribute to the field of green bond premium in the corporate green bond market where relatively few studies have been conducted and prompt more future studies by involving more variables such as ESG ratings when the corporate bond market gets matured. Lastly, the absence of green bond premium in the corporate green bond market could provide a clearer expectation for firms that will issue green bonds. The issuance of green bonds could lead to a rise in share price or bring long term value, but it does not lower the cost of borrowing at the moment.

(8)

8

2. Literature Review and Hypotheses:

2.1 Impact of Financial Crises on the bond yield spread

Before zooming into past literature about green bond premium, the general impact of past financial crises on the bond market is reviewed first to provide an expectation of the change in the bonds’ yield spread for COVID-19. Yield Spread normally represents the difference between the yield of debt security over US

Treasury yield of the same maturity and the rise in yield spread during crises has been widely studied. According to Schuknecht, Hagen, & Wolswijk (2010), the yield spread for non-benchmark bonds raised significantly during the subprime crisis. Similarly, the paper by Buckley, Avgouleas, and Arner(2018) suggested that there has been soaring yield spreads for the European bonds during the European debt crisis from the end of 2009 to 2012. In both papers, the increase in yield spread can be largely explained by economic fundamentals such as the increase in debt and deficit, which can in return lead to a higher chance of default. That also accords with the finding from Longstaff, Mithal, and Neis (2005)that default risk accounts for a large portion of the credit spread. Besides, according to Schuknecht, Hagen, and Wolswijk (2010), financial crises can intensify the effect of the determinant variables on the yield spread through increased global risk aversion. For example, the increase in debt ratio by 1 percent according to the reference country led to an increase in the yield spread 0.16 basis point before the collapse of Lehman’s brother’s in September 2008 but 1.25 basis points after its collapse. Hence, a soaring yield spread has been

identified during past financial crises due to the economic fundamentals such as higher default risk and increased risk aversion.

Regarding the COVID-19 crisis, both demand and normal production have been severely suppressed. The majority of companies have faced a lot of financial

difficulties. According to the American Bankruptcy Institute (2020), the total number of commercial chapter 11 filings for bankruptcy in May 2020 has been increased by 48% to 722 compared to May 2019. Therefore, there is an expected rise in default risk. Also, according to OECD(2020), risk aversion has been raised globally in ways

(9)

9

that not have been seen since the subprime crisis. Hence, it is expected that the outbreak of COVID-19 will increase the yield spread significantly due to higher default risk and risk aversion.

Hypothesis 1: There should be a significant increase in yield spread in the COVID-19 crisis

2.2 Measurement of green bond premium

Back to the green bond premium, since in theory investors are willing to give up part of their gains for the benefit of green projects, the green premium is represented as a lower yield and spread than comparable ordinary bonds. Although green bonds represent relatively a new market, several studies investigating the presence of a green bond premium have been published but showed mixed results. According to

Nanayakkara and Colombage (2019), they obtained the conclusion of the presence of the green bond premium of 63 basis points by panel regression of daily observations of 82 green bonds and 43 comparable corporate bonds from the same companies from 2016 to 2017. Also according to the research by Torsten and Frank(2017), by

analyzing the credit spread of a cross-section of 21 green bonds and 21 conventional bonds from the same issuer between 2014-2017, a green bond premium of 18 basis points was observed. Another paper by Baker et al. (2018) used a linear regression model with fixed effects over the US municipal bond issued from 2010 to 2016 to show the presence of 7 basis points premium for normal green bonds and 17 basis points for certified green bonds.

Certain comments could be made for these three pieces of literature. For the first one, it involves the daily credit spread of the conventional and green bonds where the number of green bonds and the number of conventional bonds are not the same. Also, criteria for matching the green bond with the comparable corporate bond were not given. For the second literature, the number of green bonds is relatively small, and the bonds selected are from the early stage of the concepts of green bonds, which may not represent the whole market status. The third literature improved based on the Karpf and Mandel (2017)’s method by considering of bonds’ taxable feature. However, the

(10)

10

fixed effects model may ignore the time-variant variables and other non-linear variables.

On the other hand, the paper by Larcker and Watts (2019) used an exact matching method by selecting 640 matched pair of green security and non-green security with the same issuer on the same day with the same structure and the years to maturity differential within one year. By observing the distribution of the differences between the matched pairs, 88 percent of the pairs showed a zero difference in the yield

spread. The t-test on the matched pairs also did not show the evidence of a green bond premium. Another paper by Flammer (2020) focused on the 225 matched pairs of green bonds and comparable ordinary bonds obtained through the greedy nearest neighbor algorithm with 14 variables into consideration. The t-test on the difference in the yield spread of the matched pairs also did not show the presence of the green bond premium. Larcker and Watts(2019) also conducted the greedy matching method with the same month, same-week, and same-day samples. No evidence of the

presence of the green bond premium was found on all these samples. The exact matching method and greedy matching method may be considered as a better method as they control the matched pairs with the same or very similar controlled variables. The potential impact of omitted variables therefore can be minimized.

There are some potential reasons to explain the non-existent green bond

premium. One rational supported by Larcker and Watts (2019) is that green projects are profitable to bring more competitiveness in the future and therefore higher returns. Hence, the returns of green bonds can be competitive. On the other hand, people may not be willing to give up their returns for environmental benefits. Also, Flammer confirmed the signaling effect of green bonds where the share price rose with the issuance of green bonds.

Overall, the exact matching method by Larcker and Watts seems to be the best in controlling for the covariates. However, it is not possible to find such exact matching in the US corporate bond market so far. Hence the greedy matching method would be used primarily for testing the presence of green bond premium in this paper. Based on

(11)

11

the study of Larcker and Watts(2019) as well as Flammer(2020), the following prediction was made:

Hypothesis 2: there is no presence of green bond premium in the normal period

2.3 Impact of Covid-19 on green bond premium

There was no past research about the impact of the financial crisis on the green bond premium. Torsten and Frank (2017) suggested that green bond was used as a hedge by some investors against environmental risks. However, the financial crisis this time is not about the environmental challenge. Hence, it is difficult to expect the green bond will be used as a hedge for this case. As people become more risk-averse during the period of the financial crisis, the risk and return would primarily the focus. People may care less about the environment during the period. As a result, the

prediction is that:

Hypothesis 3: there is no presence of green bond premium during the COVID-19 crisis

3. Methodology:

3.1 Regression with crisis dummy to investigate the impact of COVID-19 To test the first hypothesis of the presence of increased yield spread during the COVID-19 epidemic, a dummy variable suggesting a crisis is created with a value of 1 representing the bonds issued in the COVID-19 crisis and value of 0 representing the bonds issued in the normal period. By running a regression1 of yield spread on the crisis dummy and controlled variables with fixed effects, the coefficient of the crisis dummy will represent the average difference in the yield spread between the COVID-19 period and the normal period. The use of the dummy of crisis is like the method used by von Hagen, Schuknecht & Wolswijk (2011) to test the difference in yield spread before and after the subprime crisis. However, the explanatory variables such as the debt ratio are not used by the regression model in this paper as the primary

1

In addition, robust standard errors are used in all regression model in this paper in order to minimize the (potential) impact of heteroscedasticity.

(12)

12

focus of the test is to identify the difference in yield spread before and during the crisis instead of the causes of the differences in yield spread.

The control variables included here are bond characteristics that often influence yield spread. Those variables are: 1) Offering size. As the larger the size, the bonds can have greater liquidity and hence a lower yield spread; 2) Years to maturity. In a normal situation, the longer the years to maturity, the higher risk and higher the yield spread 3) Credit ratings. The better the credit ratings, the lower chance of default, and the lower the year spread.4) Currency. Yield spread may be different with different currencies.

3.2 Nearest Neighbor Matching Algorithm

Since the green bond premium is represented as a lower yield and spread in green bonds than comparable ordinary bonds, the primary concern to identify the green bond premium is to exclude the impact of other control variables such as years to maturity and credit ratings. The best solution to ensure the identicalness in control variables brought up by Larcker and Watts (2019) is to find the green and ordinary bond pairs that have the same issuer, the issue date, the same coupon rate, and ideally the same length of maturity. It is possible to identify such pairs in the municipal bonds where in many cases, similar bonds are issued simultaneously. However, given the relatively small number of corporate green bonds temporarily, such pairs do not exist in the US corporate bond market.

The following best alternative is the greedy nearest neighbor matching algorithm. This method is also stated by Larcker and Watts (2019) as well as Flammer(2020). The idea is to match each green bond with an ordinary bond with the most similar covariates. In order to match the pair as similar as possible, the pairs will have exact matching on year-month of issuance, currency, and callability. According to the Municipal Securities Rulemaking Board(MSRB,2018), the yield to call can be significantly different from yield to maturity. Hence, it is important to have the exact match on callability. For the other covariates, they are matched by minimizing the Mahalanobis distance. Since the Mahalanobis distance represents the distance

(13)

13

between two points in multivariate space, by minimizing the Mahalanobis distance, the green bond will be matched with the ordinary bond with the most similar

covariates. Other covariates include other basic bond characteristics: credit rating, ln (offering amount), coupon, years to maturity, and years to worst2. Through the nearest neighbor matching algorithm based on those covariates3, the aim is to obtain matched pairs with very similar covariates and the only difference is “greenness”.

After the matching, the p-value and standardized mean difference of the

covariates among the matched green bonds and ordinary bonds are checked to ensure the matched pairs are similar in terms of all the covariates. After the matching, paired t-tests will be conducted on yield and yield spread to test for the presence of the green bond premium.

Test statistics: t = 𝑋̅̅̅̅−𝜇𝐷 0

𝑆𝐷/√𝑛 where XD represents the yield(or yield spread) of the

green bond – that of the ordinary bond in a matched pair.

The nearest neighbor matching algorithm and the paired t-tests will be performed separately on the bonds issued in the normal period and the bonds issued during the COVID-19 crisis. If a significant negative 𝑋̅̅̅̅ is obtained, it shows the presence of a 𝐷

green bond premium.

3.3 Regress with fixed effects: Reconciliation with Baker et al. (2018) Besides, the linear regression method with fixed effects by Baker et al. (2018) would also be applied in this paper as limited research has investigated the corporate bond market by this method and hence it is meaningful to test whether the bond premium results would be observed in the corporate bond market under this method. A dummy called green is created where the value of 1 represents the bond is a green bond and a value of 0 represents an ordinary bond. The primary model of this method is the regression of yield on the green dummy and other bond characteristics with

2_{Years to worst normally refers to years to maturity for non-callable bonds and years to call for callable bonds}

3

The nearest neighbor matching algorithm is available at different computer software. In this paper, the Matchby function available in R package “Matching” is deployed to perform the nearest matching algorithm to match each green bond with the most similar ordinary bond with described covariates.

(14)

14

fixed effects. The coefficient of the green dummy would represent the difference in yield between the green bonds and ordinary bonds.

The major fixed effects here used here are year-month of issuance and years to maturity and credit ratings. Other controlling variables include the ln (offering amount), the fixed effects of currency, the fixed effects of industry, the fixed effects of the use of proceeds, and the fixed effects of the security types of bonds. The fixed effects of the use of proceeds and the fixed effects of the security types of bonds are included to control more relevant variables. In addition, since some of the controlled variables such as taxable State, taxable AMT, and new money used in the models by Baker et al. (2018) are only relevant to municipal bonds, these variables are not included in this paper. Also, another regression of yield on the green dummy, the interaction effect of month×years to maturity×ratings4, and the same other controlling variables is conducted as a variation of the specifications. Similar to the nearest neighbor matching algorithm, the two regressions are conducted separately on the bonds issued in the normal period and COVID-19 period.

This model requires the fixed effects to be effective controls. However, it may not be the case. There could be omitted time-variant variables and non-linear factors affecting the validity of the method.

4. Data Selection:

4.1 The motivation for choosing US corporate bond market

The paper tests the hypotheses using US corporate bonds. There are primarily four reasons: firstly, although the US corporate green bonds are not the most in terms of quantity, they have the most complete information accessible by Capital IQ; secondly, prior studies mainly focused on the US municipal bonds, but few studies focus on corporate bonds. While the number of green corporate bonds is also rising rapidly, it is meaningful to focus on the corporate bond sector: thirdly, the corporate

4_{Since there are too many levels of factors yearmonth and years to maturity for interaction effect, issue month is}

(15)

15

bonds are deemed with higher default risk than the municipal bonds. Hence, the impact of the COVID-19 crisis on the green bond premium could be more significant on the corporate market; fourthly, although the most desired exact matching method is not yet possible for corporate green bonds, the nearest neighbor matching method is good enough to produce valuable insights for testing green bond premium in the corporate market.

4.2 Sample Period Determination：

From a quick screening of the FactSet database, 7 corporate green bonds were issued in the US market in 2016 and since then the green bond corporate market has kept rising. Hence, the starting point of our sample is January 1st, 2016. From January 2016 to January 2020, the US has been in a stable status of growth. According to Bureau of Economic Analysis (2020), after getting out of the mere 0.1% quarterly GDP growth in the 4th quarter 2015, the US quarterly growth has then ranged from 1.1% (4th quarter, 2018) to 3.5% (4th quarter 2017 & 2nd quarter 2018) from 2016 to 2019. As a result, all corporate bonds issued from January 1st, 2016 to January 31st, 2020 are considered issued during the normal period in which the economy has been growing consistently. There is no assumed major financial crisis for this period.

After a few cases of COVID-19 being identified in China since late December 2019, WHO declared COVID-19 outbreak as a Public Health Emergency of

International Concern (PHEIC) on January 30th_{, 2020 and raised the risk assessment}

of COVID-19 to a very high level for China and high level for the world

(WHO,2020). Therefore, all the bonds issued from February 1st_{, 2020 to June 15}th_,

2020 when this paper stopped collecting data are considered as bonds issued during the crisis.

4.3 Sample Selection

The primary databases used in this analysis are FactSet and Capital IQ to provide comprehensive information about the fixed income securities. In addition, since Bloomberg has been well-used in researches relevant to green bonds and has the most comprehensive green bond collections according to industry professionals, it is also

(16)

16

been used to identify missed green bonds in the FactSet database (Larcker and Watts, 2019).

Through the screening function of FactSet, a total of 12009 fixed coupon

corporate bonds issued in the US market from January 1st, 2016 to June 15th, 2020 are identified. Among those bonds, 11903 are labeled as ordinary corporate bonds and 106 are labeled as green bonds through the presence of a green bond flag. In order to obtain as many green bonds as possible, the Bloomberg database, the most

comprehensive green bond database, is also consulted. A total of 154 corporate green bonds from the same period are identified from the Bloomberg database. Through a cross-checking with FactSet database, 68 green bonds appear in both datasets, and 9 bonds are labeled as green bonds in the Bloomberg database but corporate bonds in the FactSet database. Those 9 bonds are manually changed to green bond labels.5 For the remaining green bonds in the Bloomberg database, they are mainly issued by United States International Development Finance Corp and Citigroup Inc, which do not appear in the FactSet database nor provide enough information such as yield spread at issuance for the analysis. Hence, they are excluded from the analysis.

After the correction for the green bond label, the raw dataset consists of 11894 ordinary bonds and 115 green bonds. For each bond, the following information is collected mainly through Capital IQ as it contains the most information about US securities: the ISIN of the securities, the date of issue, the amount of offering, the offering yield and yield spread, the length of maturity at issuance, the years to worst at issuance, callability, green label, coupon rate, coupon payment frequency, currency, industry, use of proceeds as well as the S&P credit rating

4.4 Data Cleaning:

Further data cleaning steps are carried out to prepare data for the following analysis:

5_{The ISINs of the 9 corrected bonds are US26444HAH49, US22282EAG70, US595620AV77, US26442UAH77,}

(17)

17

1) Remove those bonds whose yield spread cannot be calculated or nor in the Capital IQ database. In total, 2055 bonds are removed (2049 ordinary bonds and 6 green bonds)

2) Since the coupon payment for the green bonds is either annual or semi-annual, 765 ordinary bonds with zero-coupon payment, monthly payment, quarterly payment or perpetual payments are removed

3) 81 ordinary bonds are removed because neither the offering price nor the offering yield can be found

4) 942 bonds (including 9 green bonds) are removed because of the missing of the S&P credit rating data which is a crucial component for bond yield. 5) Lastly, 304 corporate bonds are removed because of the missing of the

offering amount.

4.5 Data Summary and Description

After the cleaning step, now the complete dataset has 7862 corporate bonds in total, of which 100 are green. The categorical variables such as credit ratings are also treated as numeric in this case. For the variable credit ratings, the numeric value is assigned as the order of credit rating where AAA rating has been assigned a value of 1, and D rating has been assigned a value of 22.

---Insert Table 1 about here ---

Table 1 shows the bond characteristics across the normal period and COVID-19 period. In comparison to the bonds issued during the normal period, the bonds issued during the COVID-19 period have a lower yield by 0.27 percentage point and higher yield spread by about 99 basis points. That also meets the expectation of significantly higher yield spread during crises. Both the years to maturity and years to worst have been increased by roughly 1 year during the COVID-19 crisis and the proportion of bonds with call-option also increased. The median of the credit rating has been increased by 1 grade to 8, which corresponds to a BBB+ rating. The proportion of green bonds in the corporate bond market has been increased from 1% to 2%, though

(18)

18

the market share is still tiny. The general offering amount has been also increased by 100 million on average and lastly, the coupon rate has been decreased.

From Table 2, the number of green bonds increases across years with a jump from 2019 onward. Although the data collected for 2020 is only up to June 15th, there have already been 33 green bonds compared to a total number of 37 green bonds in 2019. Across the years, the average yield and spread for the green bond have always been lower than that of ordinary bonds, however, without controlling for the controlled variables, no conclusion can be drawn.

---Insert Table 3(a) and 3(b) about here ---

Table 3(a) and Table 3(b) show the difference between ordinary and green bonds on different bond characteristics in the normal period and the COVID-19 period respectively. Overall 6457 ordinary bonds and 72 green bonds are issued for the normal period and 28 green bonds and 1305 ordinary bonds are issued during the COVID-19 period. In both normal and COVID-19 period, the green bonds and ordinary bonds are quite different in terms of many dimensions such as coupon rate and ratings as the p-values of these dimensions are smaller than 0.05 and standardized mean differences(SMD) are greater than 0.2. Hence, it also suggests that matching is necessary to minimize the impact of differences in controlled variables on the yield and yield spread.

On average, in both normal and COVID-19 periods, green bonds have a higher offering amount6, lower yield to maturity at issuance, lower coupon rate, and higher credit ratings than ordinary bonds. While in terms of the length of maturity and the length of years to worst, the green bonds have longer years to maturity and longer length of years to worst in normal periods but shorter years to maturity and shorter years to worst in the COVID-19 period than ordinary bonds.

In terms of cross-period comparison, for both green bonds and ordinary bonds, coupon rate, yield to maturity, and yield spread follows the same conclusion obtained

(19)

19

from Table 1. The exception is that the years to maturity and the years to worst have been increased for ordinary bonds while decreased for green bonds from the normal period to the COVID-19 period.

5. Hypotheses Testing and Results

5.1 Hypothesis 1 testing and result

Our Hypothesis 1 suggests that there should be a significant increase in yield spread in the COVID-19 crisis

Although Table 1 shows that the mean of yield spread from bonds issued during the COVID-19 period is higher than that from bonds issued in the normal period by around 99 basis points, a proper statistical test is still necessary to eliminate the impact of control variables on yield spread. As mentioned in section 3.1, a regression of yield spread on the dummy of crisis and bond characteristics with fixed effects is conducted. The result of the regression is shown in Table 4.

From the results, the average yield spread during the COVID-19 period has been significantly higher than the normal period by the estimate of 121.2 basis points after controlling the ln(offering amount) and the fixed effects of currency, ratings, and years to maturity. The null hypothesis where there is no difference in the average yield spread during the normal period and the COVID-19 period is rejected by the extremely small p-value (p=0.000). The significant increase in the magnitude of yield spread during the COVID-19 period aligns with the past studies’ results of large increases in bonds’ yield spread during financial crises. The first hypothesis that the yield spread will be increased during the COVID-19 is supported. One point to note is that the adjusted R2 value of 0.265 for this model is not large because other

explanatory variables from economic fundamentals which change during the COVID-19 crisis and cause the change in yield spread are not included. The focus of the paper is to identify the presence of rising yield spread during COVID-19 instead of finding

(20)

20

the reasons for the rising yield spread. Hence, the model in this paper fulfills this purpose despite a relatively weak explanatory power.

5.2 Hypothesis 2 and 3 testings by Nearest Neighbor Algorithm

Hypothesis 2 and 3 are about the presence of a green bond premium in the normal period and COVID-19 period. As mention in section 3.2, the nearest neighbor

algorithm is used to match a green bond with the most similar ordinary bond. Then the paired t-tests on yield and yield spread are conducted to test the presence of the green bond premium.

Hypothesis 2, which predicts that there is no presence of a green bond premium during the normal period, is tested first. According to the choice of covariates explained in Section 3.2, the greedy nearest neighbor matching algorithm matches each green bond issued in the normal period with an ordinary bond with exact matching on ‘year-month of issuance’, ‘currency’ and ‘callability’ and minimal Mahalanobis distance with covariates ‘years to maturity’, ‘years to worst’, ‘credit ratings’, ’coupon’ and ln(offer amount).

The results for the bond characteristics of the matched pairs in the normal period are shown in Table 5(a)

---Insert Table 5(a) about here ---

From Table 5(a), the standardized mean differences in all matching covariates are lower than 0.08 and the p-values are all much higher than 0.05, indicating no evidence of differences between green bonds and matched ordinary bonds in terms of all these covariates. Well-matched pairs of green and ordinary bonds in the normal period are achieved.

Paired t-tests are then conducted to test whether there is a difference in yield and yield spread between green bonds and paired ordinary bonds during the normal period.

(21)

21

For both differences in yield and yield spread, their estimates are a bit higher than 3 basis points but not statistically significant as the P-values are higher than 0.05. The null hypothesis that there is a difference in yield and the yield spread between the matched pairs cannot be rejected. Hence, there is no enough evidence of the presence of a green bond premium during the normal period. Under the nearest neighbor matching method, hypothesis 2 that there is no green bond premium in the normal period is supported.

The same procedure is applied for bonds issued during the COVID-19 period and the matching method remains the same and the covariates do not change:

The results for the bond characteristics of the matched pairs during the COVID-19 period are shown in Table 6(a):

---Insert Table 6(a) about here ---

From Table 6(a), the matching is generally good for the dimensions: “years to maturity”, “years to worst”, “coupon” and “ratings” as they have the mean

standardized difference less than 0.2. The covariate to take note is the ln(offer

amount) since this value has a standardized mean difference of 0.219 which is slightly higher than 0.2. However, there is no noticeable difference in the ln(offer amount) between these two groups when considering its p-value of 0.416 (higher than 0.05). Hence, the matching is still good enough. It can be noted in general the standardized mean differences from the COVID-19 period are higher than that of the bonds from the normal period, indicating relatively a less “reliable” matching. That may be caused by the smaller sample size for the COVID-19 period.

Similarly, paired t-tests are then conducted to test the difference in yield and yield spread between green bonds and paired ordinary bonds during the COVID-19 period.

---Insert Table 6(b) about here ---

From Table 6(b), the differences in yield and yield spread are not statistically significant as the P-values are greater than 0.05. Overall, based on the current sample size of 28 matched pairs, there is no enough evidence of the presence of a green bond premium for bonds issued during the COVID-19 period. Under the nearest neighbor

(22)

22

matching method, hypothesis 3 that there is no green bond premium in the COVID-19 period is supported.

Overall, based on the greedy nearest neighbor matching algorithm, there has not been evidence of the presence of the green bond premium in both the normal period and the COVID-19 period.

5.3 Hypothesis 2 and 3 Testing by Regression with fixed effects

As mentioned in session 3.3, here the presence of green bond premium also is analyzed by regression of offering yield on the green dummy and bond characteristics with fixed effects. The regression is run separately on the bond issued during the normal period and the COVID-19 period. The variations of the fixed effect and interaction effect of the issue month, ratings, and years to maturity are considered. The regression result is shown in Table 7.

Although the coefficients of Green in all these four regressions are negative, none of them are statistically significant as the absolute value of t-statistics is smaller than 1.96. Hence the regression method with fixed effects does not provide enough evidence of the presence of green bond premium in both periods. The regression method with the fixed effects for the corporate bonds in this paper does not reconcile with the previous results by Baker et.al(2018) and Larcker and Watts (2019) where the presence of the green bond premium was identified in the US municipal bonds. Instead, the result matches our greedy matching method results with no green bond premium.

Overall, under the regression method with fixed effects, hypotheses 2 and 3 are supported; there has been no evidence of the presence of the green bond premium in both the normal period and the COVID-19 period.

(23)

23

6. Discussion

6.1 Result Reiteration

First, the COVID-19 pandemic is still evolving and leaves a lot of uncertainties. The large increase in yield spread following the COVID-19 outbreak represents the sharp rise in global risk aversion. The fundamental impact of the crisis is hard to tell yet. Still, the paper aims to make a certain contribution to the impact of COVID-19 on the bond market. The result reconciles with past observations of rising yield spread amid crises.

Through the greedy matching method to match the green bond with the most similar ordinary bond, the desirable matched pairs have been observed in the normal period. This further confirms the prior studies by Larcker and Watts (2019) as well as Flammer (2020). In the normal period where the controlled variables are almost similar, there is no presence of a green bond premium.

On the other hand, since the green bond gets popular from 2016, the major countries issuing green bonds are in general experiencing stable economic growth with no financial crisis. It is the first time to possibly evaluate the impact of a

financial crisis on a green bond premium. The matched pairs in the COVID-19 period are not as ideal as those in the normal period perhaps due to a smaller number of bonds but statistically, they are good enough. There is also no evidence of the presence of a green bond premium in the COVID-19 period.

The linear regression method with fixed effects does not show a green premium in both the normal period and the COVID-19 period may further strengthen our argument for the limitation of the regression method. There could be some omitted variables for explaining the presence of green bond premium in the municipal bonds but not the corporate bonds. Therefore, corporate bonds do not show a green bond premium. Or potentially, the municipal bonds and corporate behave differently with regards to the green bond premium. Such a premium could be possibly present only in the municipal bonds but not corporate bonds. However, one point to note is that municipal bonds have more explanatory variables such as state tax while such

(24)

24

variables are not relevant to corporate bonds or such information is not available to corporate bonds. The overall R2 in models from Baker et al. (2018) is higher than that of the models on corporate bonds in the paper. Hence, in terms of explanatory power, the models on municipal bonds could provide more reliability. Despite that the regression method in this paper is still valid as the R2 is mainly higher than 0.8, it is difficult to further increase the R2 of the model over 0.9 by incorporating other available corporate bond characteristics. This further proves that the greedy matching method is a better option than the regression method with fixed effects temporarily in terms of minimizing the impact of other explanatory variables on corporate bonds’ yield.

Another issue with the control variables is that the regression method with fixed effect does not take into account bonds’ callability, which as explained earlier could have an impact on yield. According to Table 3(a), in the normal period, the green bonds issued are more likely to be callable than ordinary bonds. Hence, green bonds could have a lower yield because of callability rather than green bond premium. However, this is not the case for the bonds issued in the COVID-19 period where they have very similar callability and in fact, table 3(b) shows that the proportion of

callable ordinary bonds is 1 percentage point higher than that of the green bonds. Nonetheless, the failure to embed the callability suggests one weakness of the model from Baker et al (2018).

6.2 Alternative reasons for no premium

There are several potential reasons for the unfound green bond premium. Larker and Watt (2019) suggested that the issuance of green bonds can lead to an increase in the ESG rating, which can bring more benefits and lower risks in the future. Hence the future return on green bonds could be higher and investors will not have a lower yield by investing in green bonds.

Also, the assumption that the investors are willing to accept a lower return for green projects may not hold in real life. According to the report by Chiang (2017), the respondents for a treasury survey indicated that a lower yield is unacceptable for a

(25)

25

green bond. Bonds’ yield was ranked as the second most important factor behind credit rating for examining a bond offer. Similar results were also found in the interviews done by Larcker and Watts (2019) as well as Flammer (2020). According to Preclaw and Bakshi(2015), green bond investors are unwilling to accept lower yield unless they can receive risk-adjustment compensation like ordinary bonds.

Furthermore, even though the risk and returns are primary concerns for investors during the COVID-19 period, the green bonds should not be disadvantaged as green bonds are generally issued by issuers with relatively high credit ratings and they do not bear additional default risks. They should have the same level of risk and returns as the other ordinary bonds with the same credit ratings. Hence, that could also explain the absence of a green bond premium or a green bond discount in the COVID-19 crisis.

The absence of a green bond premium does not mean no benefits are associated with the issuance of green bonds. The signaling effect especially through certification has been identified by Flammer’s paper (2020), which states that the issuance of green bonds can signal the firm’s commitment towards environmental impact and lead to a rise in the share price. Hence, primarily firms may value the issuance of green bonds as a way to increase its share price via the signaling effect rather than obtain a lower cost of capital for green projects. Also, the issuance of green bonds could bring substantial value to the firm in the long run such as better credit ratings.

Also, if a firm has gained a higher ESG rating through the issuance of green bonds, there can be a free-rider effect for all its bonds. Once a firm is considered green, all its bonds could enjoy the “green bond premium”. That has been supported by the Larcker and Watts’s research (2019) that similar green bond premium also presents in placebo bonds, which are ordinary bonds issued by green companies when the regression method with fixed effects is applied. Hence, the bonds issued by the same firm should merely have the same return. However, that does not explain the greedy matching result where bonds from different companies have been matched.

(26)

26 6.3 Limitation and future studies

A potential improvement for the matching and regression is to include more relevant covariates like ESG scores. The primary reason for not including ESG scores in this paper is that the measurement of ESG scores is incomplete and not

standardized. The green bond issuances are generally in larger sizes and better credit ratings. It could mean that companies are competing intensively on areas like

sustainability. Through the matching method by similar credit ratings and issuance size, those matched pairs may come from both companies with high ESG scores where ESG scores itself may have an impact on yield spread. The green bond issuers will not need to have higher ESG scores than ordinary bond issuers. Hence, more relevant indicators such as ESG scores could be used as controls in future studies when the green bond market becomes more developed and relevant information such as ESG scores is more readily available and standardized.

Also, certification could be an important element for a green bond premium. According to Baker et.al(2018), the certified green bonds showed a higher green bond premium than non-certified green bonds. However, the FactSet database does not specify whether the green bonds are self-labeled or certified. Also, Flammer(2020) mentioned in her paper that the certification has not been standardized. Hence, the effect of certification on green bonds could be more researched in the future when certification information is more standardized and readily available.

Overall, despite the increase in the issuance of the corporate green bond, there have been only less than 200 corporate bonds issued in the US market for the past four years in total. The measurements such as ESG score and certification are not standardized yet. The corporate green bond market is still rapidly evolving and the market consensus for green bonds is not reached yet. It is still a far early answer to say whether the investors are paying for a green bond premium

(27)

27

7. Conclusion

In conclusion, this paper focuses on investigating the presence of green bond premium in the US corporate bond market as prior studies mainly focused on green bond premium in the municipal market. The sudden outbreak of the COVID-19 crisis provides the new inspiration to this paper. Firstly, by using regression with controlled variables, the significant rise in yield spread during the COVID-19 period has been identified. That serves a new perspective: whether the green bond premium will behave differently in normal periods and crisis period; whether the increase in risk aversion during the crisis period would eliminate the potential green bond premium. Therefore, the sample used in this study is purposely divided into the normal period and the COVID-19 period.

Through both the greedy matching method and regression with fixed effects, no green bond premium is identified in either the normal period or COVID-19 period. That further proves the temporary absence of a theoretical green bond premium in the corporate green bond market. Probably, the lack of green bond premium is caused by investors’ unwillingness to accept a lower return purely for the benefit of green projects. However, that does not mean green bonds add no value to the firms. Several studies have confirmed the signaling effect of the issuance of green bonds, which can cause a rise in share price and bring long-run benefits. Since companies are competing more and more in the field of sustainability, the burgeoning green bond market has not been matured yet. There is a need for continuous studying the topic of corporate green bond premium when the market is more developed. An exact matching method or more data like ESG scores could be used in future studies to bring more insights. The effect of certification on the green bond premium could also be studied into depth when such certification is more standardized. Also, since the COVID-19 crisis has not ended, the comprehensive impact of COVID-19 on the corporate green bond is not known yet. This paper would merely like to serve an inspiration for more studies in the future on the impact of crises on green bonds.

(28)

28

Bibliography

American Bankruptcy Institute. (2020). May Commercial Chapter 11s

Increase 48 Percent over Last Year, Total Filings Down 42 Percent.

Retrieved from

https://www.abi.org/newsroom/press-releases/may-commercial-chapter-11s-increase-48-percent-over-last-year-total-filings

Baker, M., Bergstresser, D., Serafeim, G., & Wurgler, J. (2018). Financing

the Response to Climate Change: The Pricing and Ownership of U.S.

Green Bonds. SSRN Electronic Journal. doi: 10.2139/ssrn.3275327

Buckley, R., Avgouleas, E., & Arner, D. (2018). Three Major Financial

Crises: What Have We Learned. SSRN Electronic Journal. doi:

10.2139/ssrn.3247455

Bureau of Economic Analysis. (2020). Gross Domestic Product. Retrieved

from https://www.bea.gov/data/gdp/gross-domestic-product

Climate Bonds Initiative. (2020). Green Bond Highlights 2019: Behind the

Headline Numbers: Climate Bonds Market Analysis of a record year.

Retrieved from

https://www.climatebonds.net/2020/02/green-bond-

highlights-2019-behind-headline-numbers-climate-bonds-market-analysis-record-year

European Investment Bank. (2020). Climate Awareness Bonds. Retrieved

from

https://www.eib.org/en/investor_relations/cab/index.htm

Flammer, C. (2020). Corporate Green Bonds. SSRN Electronic Journal. doi:

10.2139/ssrn.3125518

Friedman, H., & Heinle, M. (2016). Taste, information, and asset prices:

implications for the valuation of CSR. Review Of Accounting

Studies, 21(3), 740-767. doi: 10.1007/s11142-016-9359-x

Global financial markets policy responses to COVID-19. (2020). Retrieved

30 June 2020, from

http://www.oecd.org/coronavirus/policy-

responses/global-financial-markets-policy-responses-to-covid-19-2d98c7e0/

Guidolin, M., & Tam, Y. M. (2010). A yield spread perspective on the great

financial crisis: Break-point test evidence. (). Retrieved

from

http://econpapers.repec.org/paper/fipfedlwp/2010-026.htm

International Moneytary Fund. (2020). World Economic Outlook, April

2020: The Great Lockdown. Retrieved from

https://www.imf.org/en/Publications/WEO/Issues/2020/04/14/weo-april-2020

(29)

29

Karpf, A. & Mandel, A. (2017). Does it Pay to Be Green? Available at

SSRN: https://ssrn.com/abstract=2923484

Larcker, D., & Watts, E. (2019). Where's the Greenium?. SSRN Electronic

Journal. doi: 10.2139/ssrn.3333847

LONGSTAFF, F., MITHAL, S., & NEIS, E. (2005). Corporate Yield

Spreads: Default Risk or Liquidity? New Evidence from the Credit

Default Swap Market. The Journal Of Finance, 60(5), 2213-2253. doi:

10.1111/j.1540-6261.2005.00797.x

Martin, P. R., & Moser, D. V. (2016). Managers’ green investment

disclosures and investors’ reaction. Journal of Accounting and

Economics, 61(1), 239-254. doi:10.1016/j.jacceco.2015.08.004

Municipal Securities Rulemaking Board. (2018). Mark-up Disclosure and

Trading in the Municipal Bond Market.

National Bureau of Statistics of China. (2020). Preliminary Accounting

Results of GDP for the First Quarter of 2020. Retrieved from

http://www.stats.gov.cn/english/PressRelease/202004/t20200420_17398

11.html

Nanayakkara, M., & Colombage, S. (2019). Do investors in Green Bond

market pay a premium? Global evidence. Applied Economics, 51(40),

4425-4437. doi: 10.1080/00036846.2019.1591611

OECD. (2020). OECD Interim Economic Assessment Coronavirus: The

world economy at risk.

Preclaw, R., & Bakshi, A. (2015). The Cost of Being Green. Retrieved 30

June 2020, from

https://www.environmental-finance.com/assets/files/US_Credit_Focus_The_Cost_of_Being_Green.

pdf

Schuknecht, L., Hagen, J., & Wolswijk, G. (2010). Government bond risk

premiums in the EU revisited. Frankfurt am Main: European Central

Bank. Retrieved from

http://www.econis.eu/PPNSET?PPN=627443168

Torsten, E., & Frank, P. (2017). Green Bond Finance and Certification

(September 17, 2017). BIS Quarterly Review September 2017. Available

at SSRN:

https://ssrn.com/abstract=3042378

U.S. BUREAU OF LABOR STATISTICS. (2020). Unemployment rate

rises to record high 14.7 percent in April 2020. Retrieved from

https://www.bls.gov/opub/ted/2020/unemployment-rate-rises-to-record-high-14-point-7-percent-in-april-2020.htm

(30)

30

United Nations. (2020). We are all in this Together: Human Rights and

COVID-19 Response and Recovery. Retrieved from

https://www.un.org/en/un-coronavirus-communications-team/we-are-all-together-human-rights-and-covid-19-response-and

von Hagen, J., Schuknecht, L., & Wolswijk, G. (2011). Government bond

risk premiums in the EU revisited: The impact of the financial

crisis. European Journal Of Political Economy, 27(1), 36-43. doi:

10.1016/j.ejpoleco.2010.07.002

World Health Organisation. (2020). Archived: WHO Timeline - COVID-19.

Retrieved from Archived: WHO Timeline - COVID-19. (2020).

Retrieved 30 June 2020, from

https://www.who.int/news-room/detail/27-04-2020-who-timeline---covid-19

World Health Organisation. (2020). WHO Coronavirus Disease

(COVID-19) Dashboard. Retrieved from https://covid19.who.int/

World Trade Organization. (2020). Trade set to plunge as COVID-19

pandemic upends global economy. Retrieved from

(31)

31

Tables

Table 1: The summary of the bond characteristics by normal period and COVID-19 period

This table reports the bond characteristics of corporate bonds by normal and COVID-19 period, where normal period presents the January 1st,_{2016 to January 31}st,₂₀₂₀

and the COVID-19 period represents February 1st, 2020 to June 15th, 2020. The mean of the differences in each bond characteristic and the P-value are also reported.

Normal Period COVID-19 Diff

N 6529 1333

Variables Mean Median Mean Median Mean P-Value

Yield(%) 4.21 3.93 3.94 3.5 0.27 <0.001 Yield Spread (BPS) 203.66 150 302.77 252.20 -99.11 <0.001 Years to Maturity (Years) 10.79 8 11.71 9 -0.92 0.001 Years to Worst(Years) 9.20 6 10.38 8 -1.18 <0.001 Callability (Yes=1) 0.91 1 0.94 1 -0.03 <0.001 Ratings (AAA=1) 9.28 9 8.60 8 0.68 <0.001 Green (Yes=1) 0.01 1 0.02 1 -0.01 0.003 Offering Amount($M) 689.47 500 793.08 600 -103.61 <0.001 Coupon(%) 4.18 3.9 3.91 3.5 0.29 <0.001

(32)

32

Table 2: The summary of green bond and ordinary bond by years.

This table reports the number of ordinary bonds and green bonds by year as well as their average yield to maturity (in %) and average yield spread (in BPS) by year

Ordinary Green

Year N Yield (%) Yield Spread (BPS)

N Yield (%) Yield spread (BPS) 2016 1209 3.86 221.49 7 2.39 119.99 2017 1593 4.19 203.29 10 3.83 166.08 2018 1463 4.66 183.74 13 4.11 108.57 2019 1983 4.22 220.86 37 3.50 158.28 20207 ₁₅₁₄ _3.90 _276.64 ₃₃ _2.64 _190.98 Total 7762 4.18 221.24 100 3.25 160.71 7

(33)

33

Table 3(a) Bond Characteristics stratified by Green in the normal period

This table reports the bond characteristics of ordinary bonds and green bonds respectively in normal period, the value represents the mean value and the number in bracket represents the standard deviation. The mean standardized difference and p-value in each characteristic between ordinary bonds and green bonds are reported as well

Stratified by Green

Ordinary (mean(SD)) Green (mean(SD)) p SMD

N 6457 72 Coupon(%) 4.20 (1.77) 3.40 (1.43) <0.001 0.499 Years to Maturity(Years) 10.77 (8.99) 12.72 (9.14) 0.067 0.215 Years to Worst(Years) 9.17 (9.11) 11.19 (9.92) 0.061 0.212 Callability(Yes=1) 0.90 (0.29) 0.99 (0.12) 0.019 0.364 Yield to Maturity(%) 4.22 (1.78) 3.42 (1.42) <0.001 0.496 Ratings (AAA=1) 9.30 (3.57) 7.40 (2.70) <0.001 0.6 ln(Offer Amount) 5.61 (2.08) 6.22 (0.73) 0.013 0.393

Table 3(b) Bond Characteristics stratified by Green during COVID-19 period

This table reports the bond characteristics of ordinary bonds and green bonds respectively in normal period, the value represents the mean value and the number in bracket represents the standard deviation. The mean standardized difference and p-value in each characteristic between ordinary bonds and green bonds are reported as well

Ordinary (mean(SD)) Green (mean(SD)) p SMD

N 1305 28 Coupon(%) 3.93 (2.06) 2.77 (1.29) 0.003 0.678 Years to Maturity(Years) 11.71 (9.77) 11.54 (9.02) 0.925 0.019 Years to Worst(Years) 10.39 (10.00) 9.86 (9.99) 0.779 0.054 Callability(Yes=1) 0.94 (0.24) 0.93 (0.26) 0.853 0.034 Yield to Maturity(%) 3.96 (2.10) 2.82 (1.27) 0.004 0.659 Ratings (AAA=1) 8.64 (3.29) 6.86 (2.34) 0.004 0.626 ln(Offer Amount) 5.91 (2.08) 6.29 (0.46) 0.33 0.255

(34)

34

Table 4: Regression of yield spread on Crisis dummy, bond characteristic, and fixed effect

This table reports the result of the regression of yield spread on crisis dummy, bond

characteristic, and fixed effect. Here the bond characteristic used is ln (offering amount). The fixed effects are credit ratings, yield to maturity, and currency.

Dependent Variable: Yield Spread

Variables Coefficient Robust Std. Err. T-statistics P-Value

Crisis*** _121.201 _6.170 _19.65 _0.000

ln (offer amount) * _4.040 _1.985 _2.04 _0.042

Adj. R2 _0.265

(35)

35

Table 5(a) Bond characteristic for matched pair during the normal period

This table reports the difference in covariates between ordinary bonds and green bonds in the matched pair during the normal period. Since “Yearmonth”, “callability” and “currency” are matched exactly, they will have a standardized mean of difference off zero and zero distance on these dimensions. Hence, those three covariables are not reported in the table. The value represents mean and the number in bracket represents the standard deviation.

Ordinary Green p-value SMD

N 71 71

Coupon (%) 3.42 (1.29) 3.45 (1.38) 0.92 0.017

Years to maturity (Years) 12.85 (9.07) 12.86 (9.13) 0.993 0.002

Years to worst (Years)" 11.21 (9.75) 11.35 (9.90) 0.932 0.014

Ratings (AAA=1) 7.63 (2.52) 7.44 (2.71) 0.654 0.075

ln(offer amount) 6.23 (0.77) 6.22 (0.73) 0.915 0.018

Table 5(b) Paired t-test for yield difference and yield spread difference for the matched pairs in the normal period

This table reports the paired t-tests for the yield and yield spread difference in the matched pairs where the difference in yield(or yield spread) in each pair is calculated the yield (or yield spread) of the green bond – the yield(or yield spread) of the matched ordinary bond. All the values are represented in basis points.

95% Confidence Interval

Mean Diff P-value Lower Bound Upper Bound

Difference in Yield (BPS) 3.634 0.319 -3.587 10.856

Difference in Yield spread(BPS)

(36)

36

Table 6(a) Bond characteristic for matched pair during COVID-19 period

This table reports the difference in covariates between ordinary bonds and green bonds in the matched pair during the COVID-19 period. Since “Yearmonth”, “callability” and “currency” are matched exactly, they will have a standardized mean difference of zero and zero distance on these dimensions. Hence, those three covariables are not reported in the table. The value represents mean and the number in bracket represents the standard deviation.

Ordinary Green p-value SMD

N 28 28

Coupon (%) 2.85 (1.37) 2.77 (1.29) 0.818 0.062

Years to maturity (Years) 11.18 (9.19) 11.54 (9.02) 0.884 0.039

Years to worst (Years)" 9.93 (9.66) 9.86 (9.99) 0.978 0.007

Ratings (AAA=1) 7.14 (2.37) 6.86 (2.34) 0.651 0.121

ln(offer amount) 6.39 (0.48) 6.29 (0.46) 0.416 0.219

Table 6(b) Paired t-test for yield difference and yield spread difference for the matched pairs during COVID-19 period

This table reports the paired t-tests for the yield and yield spread difference in the matched pairs where the difference in yield(or yield spread) in each pair is calculated the yield (or yield spread) of the green bond – the yield(or yield spread) of the matched ordinary bond. All the values are represented in basis points.

95% Confidence Interval

Mean Diff P-value Lower Bound Upper Bound

Difference in Yield (BPS) -5.801 0.449 -21.301 9.700

Difference in Yield spread (BPS)

(37)

37

Table 7 Regression of yield on the green dummy, bond characteristics, and fixed effects during normal and COVID-19 period.

This table reports the coefficient of the green dummy with the variation in fixed effects and interaction effects when the offering yield is regressed on the green dummy, bond

characteristics, and fixed effects. Here the bond characteristics use is the ln(offering amount), the fixed effects used are fixed effects of currency, the fixed effects of industry, the fixed effects of the use of proceeds, and the fixed effects of the security types of bond. The value is represented as basis point and the value in bracket represents the t-static result.

Dependent Variable Offering Yield (BPS)

Normal Period COVID-19 Period

(1) (2) (3) （4） Green Bond -7.458 (-0.5571) -5.878 (-0.4054) -10.051 (-0.3400) -18.381 (-0.5578)

Ratings FEs Yes No Yes No

Maturity FEs Yes No Yes No

Yearmonth FEs Yes No Yes No

Month × Maturity × Ratings FEs No Yes No Yes

Observations 6529 6529 1333 1333

Is there a green bond premium and does accounting for COVID-19 epi­demic matter?