population size is 1000 individuals. This result confirm that the matching process is independent from the algorithm used. In conclusion, as long as the matching process is using the prosperity score distance, it will return the best outcome.
Model ”Difference in difference specification”
To examine how firm-level outcomes evolve following the issuance of a corporate green bond, I use the Difference-in-Differences statistical framework, and a combination of four different fixed effects. The indices of the regression are the following:
i=ric(f irm) t=year
c=primary country of risk s=SIC(industry)
yi,t=αi+αt+αs+αc+β×iGreenBondi,t+i,t (6.2) where the dependent variables are presented in section 4.5.1. αiis firm fixed effect, i.e., the effect that remain fixed across entities, but that evolves over time. αt is the variation in the time. The fixed effect comes from omitted variable that vary over time but not across entities. The intercept is excluded to prevent perfect multicollinearity. αs is the industry fixed effect, i.e., the fixed effect constant across industries. Because industry regulation is introduced nationally, it affects all entities belonging to that industry. Finally, αc represents the countries’ fixed affect. As aforementioned, country regulation affects all entities at the same time as it is introduced nationally.
Copenhagen Business School Master Thesis 15 Sept 2021
Matching char-acteristic
Type Count Mean Median Std. Dev. P-value (diff.
in means)
roa 1 Green 51 0.084 0.087 0.033 0.356
Vanilla 51 0.092 0.090 0.052
leverage 1 Green 51 0.500 0.531 0.166 0.711
Vanilla 51 0.512 0.508 0.160
tobinq 1 Green 51 1.175 1.139 0.405 0.472
Vanilla 51 1.126 1.094 0.266
size 1 Green 51 23.510 23.720 1.141 0.378
Vanilla 51 23.750 23.910 1.563
size dif Green 51 0.089 0.033 0.234 0.197
Vanilla 51 0.043 0.030 0.094
roa dif Green 51 -0.009 0.000 0.053 0.455
Vanilla 51 -0.003 -0.002 0.021
tobinq dif Green 51 -0.026 -0.005 0.274 0.439
Vanilla 51 0.006 0.012 0.106
leverage dif Green 51 0.008 0.002 0.038 1.000
Vanilla 51 0.008 0.005 0.037
gps 1 Green 51 54.640 55.870 20.993 0.737
Vanilla 51 56.060 57.700 21.606
sps 1 Green 51 51.825 53.126 23.746 0.417
Vanilla 51 55.870 59.100 26.349
eps 1 Green 51 51.857 52.085 26.455 0.416
Vanilla 51 56.070 59.230 25.624
eps diff Green 51 4.737 0.734 15.153 0.679
Vanilla 51 3.727 0.564 8.474
gps diff Green 51 3.751 0.784 14.770 0.527
Vanilla 51 2.139 1.467 10.533
sps diff Green 51 5.627 2.878 12.086 0.400
Vanilla 51 3.800 1.879 9.579
Table (6.7) This table presents descriptive statistics comparing treated and matched control firms. Levels (e.g., size) are measured in the year preceding the bond issue (t – 1), while pre-trends (e.g., size diff) are measured in the two-year window preceding the bond issue (changes from t – 2 to t – 1). The last column report the p-value of the difference-in-means. *, **, and *** denotes significance at the 10%, 5%, and 1% level, respectively.
Page 78 of 198
interviews opens the possibility to ask inquisitive questions as opposed to structured in-terviews which would have added comparatively little value. The data is compiled in the appendix and consists of five interviews which can be split into three categories: fixed income specialist (2), energy sector expert (2), and climate finance specialist (1). Pref-erences regarding anonymity have been fully respected. Besides providing insights, the interviews provided opportunities to present views from our literature review and hear views from industry practitioners.
According to (Carmines & Zeller, 1979), a study can only gain scientific acceptance if the underlying data is reliable and valid. Reliability refers to whether the data collection tech-niques and analytical procedure would reproduce consistent findings if they were repeated on another occasion or if they were replicated by another researcher. To ensure reliability, the methodology and analysis of this paper were designed to possibly avoid any error or bias. Although the data collection and methodology were already explained, this section reflects on the errors and biases that this scientific paper may have encountered, and it clarifies the solutions implemented to avoid possible fallacies.
When addressing the causal effect, unmeasured or time-varying common causal effect could affect the measurement of the causal relationship between the variables which we focus on. By selecting a global and spread across different energy sectors bond population, we exclude the threat of participant biases, i.e., unmeasured and time-varying common causes.
A similar type of biases is omitted variables bias (OVB). To minimize OVB, different fixed effects are integrated with the differences-in-differences models, e.g., the firm fixed effect, the country fixed effect or year fixed effect. Although considering such fixed effects, other factors may influence the relationship between our randomized treatment and the outcome variable. For instance, Flammer (2021) uses a combination of fixed effects, i.e., a product between two single fixed effects, e.g., the industry by year fixed effects, or country by year fixed effects.
Regarding data credibility, this study uses Eikon as a database for quantitative analysis.
Eikon is a valid source for scientific studies, due to its reliance on accurate representation of data for the future of their businesses. The study can be repeated by other researchers since the data is publicly available, and we rely on data of primarily a quantitative nature.
Talking about errors that may have occurred during the research, the following paragraph addresses some. To avoid typos in writing the formulas for the data collection ourselves, we received those directly from the help desk of Eikon. Moreover, the scripts written in R
Copenhagen Business School Master Thesis 15 Sept 2021
were double-checked by an extensive search for obvious errors in the analysis outcomes.
Finally, for a research paper to be reliable, research biases cannot be ignored. The match-ing method used to pair green bond observations with comparable vanilla bonds could be a source of bias. The research chooses the variables involved in the matching equation.
However, in choosing such variables, we rely on previous peer-reviewed literature which has dealt with the problem before.
We believe that our research respects validity since the paper follows a highly structured research approach. Such a research approach is typical in the research of causal effects, i.e., the identification of the extent and nature of the cause-and-effect relationship. In this case, the cause-and-effect relationship relays between the issuance of a green bond and the post-issuance firm-level performances of energy companies (business research methodology).
We are aware that causal research comes with both advantages and disadvantages. For in-stance, coincidences in events may be measured as cause-and-effect relationships. Another threat to the validity of the paper is the difficulty of reaching an appropriate conclusion based on casual reference findings. This is due to the impact of a wide range of factors and variables in social environments. In other words, while causality can be inferred, it cannot be proved with a high level of certainty ().
However, our research design follows the best practices, which are hereafter explained, of explanatory research. Firstly, we conduct an experiment, which is the most popu-lar primary-data-collection method in studies with causal research design (). Moreover, the three main requirements of any causal relationship that claims to have validity were considered when designing the experiment. Our experiment complies with the temporal sequence component, i.e., the issue of the green bond occurs before the firm-level per-formance measurement. The second component, namely concomitant variation, is also respected as the variation of the two variables occurs systematically. Last but not least, the issuance of green bonds is endogenous with respect to firm outcome, i.e., unobservables may drive a spurious relationship between the issuance of green bonds and firm outcomes (Flammer, 2021). In an ideal setting, the threat of endogeneity would be addressed using an instrument for the issuance of green bonds (Flammer, 2021). However, an empirical setting in which companies randomly issue green bonds has not been found. Our paper makes a plausible counterfactual of how firm-level outcomes would involve absenting the green bond issuance, using the matching approach.
Page 80 of 198
Finally, multivariate OLS regression is chosen, as a significant and causal inference is sought. This cannot be achieved, to the same extent, using descriptive research solely.
However, a causal inference could be confused with a simple correlation and lead to misin-terpreted conclusions (Stock & Watson, 2012). Time series and panel data regressions may have alleviated some of these shortcomings through Granger causality tests, but sufficient historical data does not exist to allow for these approaches.
Chapter 7
Analysis
7.1 Summary statistics of energy-sector population
The method to create a data set for corporate energy bonds is described in 6.1. In essence, I extract all the corporate bonds in Eikon database that were issued by energy companies.
In particular, industry selection is performed by means of the SIC code, i.e., the Standard Industrial Classification (SIC) codes are four-digit numerical codes that categorize the industries that companies belong to base on their business activities. Besides the industry criterion, a series of other criteria is described in 6.1. Given the comprehensive coverage of Eikon’s fixed income database, the resulting data set is likely to closely map the full universe of energy green bonds. To facilitate comparisons, I convert all amounts into US dollars, and in the following section, three tables summaries key statistics of the energy green bond universe.
Table 7.1 is a proof of the rapid rise of the energy green bond market over the past decade.
From 2013 until 2015, few energy firms issued green market debt, whereas from 2016 on, the growth of the market has undertaken a positive trend, although in 2017, the yearly growth rate was negative. Table 7.2 groups the energy green bonds by industry sector of the issuer. As can be seen, energy green bonds are more common in industries where the environment is likely core to the firms’ operations, e.g., electric utilities and alternative electric utilities. Finally, Table 7.3 provides a breakdown by countries. As is shown, green bonds are especially prevalent in the US, Europe and Asian countries.
82
Year # Bond $ Tot Amount (billion)
2013 1 0,424
2015 1 0,360
2016 10 3,061
2017 6 0,465
2018 12 3,293
2019 30 9,219
2020 38 8,035
Total 98 24,857
Table (7.1) Energy green bond over time
This table reports the sum of the amount issued (in $B) as well as the count of energy green bonds issued on an annual basis from 2013 until 2020
Copenhagen Business School Master Thesis 15 Sept 2021
Industry Sector # Bond $Tot Amount (billion)
Electric Utilities 46 16,687
Alternative Electric Utilities 12 0,799
Renewable Energy Equipment & Services (NEC) 6 0,464
Hydroelectric & Tidal Utilities 6 0,585
Multiline Utilities 5 1,857
Independent Power Producers (NEC) 4 0,257
Water & Sewage Construction 4 0,309
Fossil Fuel IPPs 3 0,160
Oil & Gas Refining and Marketing (NEC) 3 0,325
Renewable IPPs 2 0,156
Geothermal Electric Utilities 2 1,580
Electrical Transmission & Grid Equipment 1 0,601
Coal (NEC) 1 0,439
Courier, Postal, Air Freight & Land-based Logistics (NEC) 1 0,077
Photovoltaic Solar Systems & Equipment 1 0,200
Electrical Components & Equipment (NEC) 1 0,360
Total 98 24,857
Table (7.2) Energy green bonds by sector
This table reports the average annual amount issued (in $B) as well as the number of energy green bond by industry sector issued on an annual basis from 2012 until 2020. The industry SIC codes are listed in 10.1
Page 84 of 198
Country # Bond $Tot Amount (billion)
United States 23 12,946
Thailand 15 0,909
Norway 11 0,653
China (Mainland) 10 1,412
France 7 0,824
Japan 5 0,402
New Zealand 5 0,500
Sweden 4 1,025
India 2 1,000
Spain 2 0,390
United Kingdom 2 0,840
Bermuda 2 1,580
Switzerland 2 0,365
Netherlands 2 0,961
Belgium 1 0,721
Singapore 1 0,200
Portugal 1 0,061
Argentina 1 0,035
Latvia 1 0,030
Italy 1 0,001
Total 98 24,857
Table (7.3) Energy green bond issuer by country of incorporation
This table reports the annual average amount issued (in $B) as well as the number of energy green bond by country issued on an annual basis from 2012 until 2020.