*4.2. Mean-Variance Analysis*

Since the dominance of banking and finance—as the issuers of green bonds in Asia may affect the bonds' performance, we present a mean-variance analysis of the rate of return of bonds, based on the region of issuance and the type of issuer. The results of this analysis on the overall sample are presented in Figure 3, and numerical values for mean and variance are provided in Appendix A.

**Figure 3.** Mean-variance analysis of the rate of return of green bonds. Source: Authors' compilation.

The overall analysis of the mean and variance of the returns of bonds shows high variation between regions of issuance. The relatively high variance of Asian bonds reflects the diversity of these bonds (confirmed in Section 4.1). In Europe, the bonds issued appear to have higher risks, with relatively low returns, and, in comparison with Asian and North American bonds, do not seem to be appealing to investors.

Figure 4 represents the main focus of our mean-variance analysis, as it provides a sectoral analysis of the risks and returns of green bonds, based on the region of issuance. As in Figure 3, specific numerical values for the mean and variance are reported in Appendix A.2. First, the mean and variance values for the manufacturing sector stand out, as they are twice as large as those of other sectors, especially in the case of European bonds. These extreme values could be explained by the size of this particular subsample, as manufacturers represent around 5% of all issuance on average. With the notable exception of the manufacturing sector, Asian bonds tend to offer higher returns than those issued in Europe and North America, but also come with higher risks. It is interesting to note that bonds issued by companies in banking and finance in Asia do not present a striking difference with those issued by other sectors, contrary to what our hypothesis would suggest. On the other hand, bonds issued by power and utilities stand out due to their high variance, compared with other sectors. This feature could explain the small share of issuance of power and utilities in the Asia-Pacific, especially as their low risk characterizes bonds issued by power and utilities companies in Europe and North America. Indeed, if bonds issued by power and utilities are deemed risky, then it is not surprising that they attract few investors, hence their relatively low share. Generally, European bonds are characterized by low returns but have low associated risks, with both the mean and variance around 1. This could explain the dominance of Europe in the green bond markets, as they could be considered more reliable assets by investors.

#### *4.3. Regression Analysis*

The core of our empirical findings lies in the regression analysis. While summary statistics and mean-variance analysis can highlight the characteristics and features of data on sectoral issuers and the difference in performance depending on the region and type of issuer, it cannot provide a conclusion on the relationship between the issuer and performance, nor can it help elucidate the significance of the difference in performance, depending on the region and issuer.

To answer these questions, the study introduces a regression analysis, estimated based on the equation provided in sub-Section 3.2, whose results are presented in Table 4. Equations are estimated on the full sample (using dummy variables to represent each region), as well as on each of the three regional subsamples, using White robust standard errors to control for model misspecifications, such as heteroskedasticity. The relatively short length of the panel (*t* = 2 for most observations) exempts us from additional time series testing on the data. Therefore, we use traditional panel data analysis estimation methods: pooled ordinary least squares (OLS) and generalized least squares (GLS) random effects (RE) estimator. The lack of time-varying independent variables precludes the use of a fixed-effect (FE) estimator. Indeed, the inclusion of a cross-sectional FE dummy variable (for each bond) does not allow us to determine the impact of the bonds' characteristics, such as the sector of issuance. Instead, adding both FE and sectorial dummy variables provokes issues of multicollinearity, as individual characteristics are both captured by FE dummy and sectorial dummy variables. Therefore, the study prefers the RE estimator, in line with [17]. Since we are interested to see the effect of the banking sector on green bonds, we further include interaction terms between each region and the banking dummy variable. Regional and interaction dummy variables for North America are used as references and excluded so as not to cause a multicollinearity issue.

**Figure 4.** *Cont*.

**Figure 4.** Mean-variance analysis of rate of return of green bonds, by sector of issuance. Source: Authors' calculation.

The regression analysis provides further information on the characteristics of green bonds, as the level of significance of the variables tends to vary depending on the region of issuance. It is interesting to note the difference in significance, depending on the analyzed sample. First, and regardless of the sample, the level of significance of the control variables is in line with the literature on the topic. For instance, the coupon rate was also found to be a significant variable in [22,24]. Similarly, maturity is often used as a control variable in studies assessing bonds' performance but is generally not found to be significant [23,24]. However, apart from these control variables, no sectorial dummy, or regional dummy, or even their interaction terms, appears to be significant. This is all the more surprising that, when conducting regressions on regional sample, sectorial dummy variables show significance, to an extent. This could potentially be explained and improved by using a larger sample of analysis.



The majority of sectoral dummy variables show a lack of significance, with the notable exception of banking and finance in the Asia-Pacific. Our results prove that bonds issued by companies in the banking and financial sector consistently display lower rates of return. Not only does this sector issue low-performing bonds, but the size of the associated coefficient (0.62 or 0.57, depending on the method of estimation) is relatively large, as the average return of Asian bonds is 3.52. Even when using the full sample, being a green bond issued from the banking sector in Asia is shown to have a slightly significant negative sign. This is all the more striking as it appears that no other sectoral dummy variable shows such high levels of significance in other regions. This result confirms that the dominance of traditional forms of banking in the Asian financial sector has an impact on the characteristics of green bonds, specifically on the performance of bonds.

The significance of year dummy variables also provides a few other takeaways from this study. As the rate of return is measured on 10 January each year, each dummy captures the state of the market at the beginning of the year. Keeping this in mind, it comes as no surprise that bonds performed relatively poorly at the beginning of 2020 in the Asia-Pacific. As the majority of Asian bonds were issued in China, their performance took a severe hit from the outbreak of the coronavirus disease (COVID-19) at the end of 2019, as shown by the negative and large coefficient linked with the 2020 dummy variable. The negative sign of the same variable in the North American sample could reflect the level of dependence of the United States economy on China: the negative expected performance of Asian bonds could therefore bring down American bonds as well.
