Green Bond Alignment, Certification and Corporate Credit Risk: Evidence from Global Issuers

Abstract

This study examines which bond-level, issuer-level, and institutional characteristics are associated with external recognition or alignment in the global corporate green bond market and whether these associations differ between Climate Bonds Initiative (CBI) alignment and formal CBI certification. The database contains 6009 green bond issuances by 822 corporate issuers between 2016 and 2025. The main complete-case analysis uses 4875 bonds issued by 682 firms and estimates a multilevel logistic model with issuer random intercepts, Fitch-rating, sector, and issuance-year controls. Larger issuance amounts are positively associated with external recognition or alignment across every specification. Issuance in a developed market and average operating margin, which is treated as an exploratory covariate, are also positively associated with the broad outcome, whereas issuer size is negatively associated; however, these relationships are more sensitive to outcome definition, temporal measurement, or estimator choice. Financial leverage and credit-rating categories show no consistent association. Separate aligned-versus-self-labelled, certified-versus-aligned, and multilevel multinomial analyses reveal substantial heterogeneity between CBI alignment and formal certification, confirming that the two categories should not be interpreted as equivalent verification mechanisms. These findings identify transaction scale as the most stable correlate of external recognition or alignment and show that issuance-market context and issuer characteristics operate differently across recognition categories.

1. Introduction

The green bond market has become an important financing channel for the energy transition and other environmentally beneficial investments. Since the first green bond issued by the European Investment Bank in 2007, issuance has expanded markedly, particularly after the Paris Agreement strengthened international commitments to climate finance (Fatica & Panzica, 2021; Lebelle et al., 2020). Corporate green bonds now finance renewable energy, sustainable infrastructure, clean mobility, energy efficiency, and related projects across developed and emerging markets (Fatica & Panzica, 2021; Teti et al., 2022).

Market expansion has not eliminated concerns about environmental credibility. Because the green label refers to the intended use of proceeds rather than to a homogeneous legal or verification regime, investors may face substantial information asymmetry and greenwashing risk (Flammer, 2021; Kapraun et al., 2019; Lian & Hou, 2024; Tang & Zhang, 2020). Issuers can respond through different recognition mechanisms. In this study, CBI-aligned bonds are instruments identified as consistent with Climate Bonds Initiative criteria, whereas CBI-certified bonds have obtained formal certification under the relevant CBI framework. Self-labelled bonds rely on the issuer’s own green designation without either CBI classification. Alignment and certification therefore differ in verification procedures, costs, and signal strength and should not be treated as substantively identical, even when they are combined in a broad indicator of external recognition or alignment.

Signalling theory provides a useful framework for studying these choices. A costly and verifiable signal can convey otherwise unobservable information and reduce adverse selection (Connelly et al., 2025; Spence, 1973). Applied to green bonds, external recognition or formal certification may strengthen the credibility of environmental claims and increase comparability across issuances (Bachelet et al., 2019; Kapraun et al., 2019; Pietsch & Salakhova, 2022). Two competing issuer-level mechanisms are plausible. A financial-capacity mechanism predicts that issuers with stronger credit quality and operating performance are better able to bear verification costs. A credibility-need mechanism instead predicts that financially riskier or less visible issuers may have stronger incentives to use external recognition to mitigate adverse perceptions. The relative importance of these mechanisms is ultimately an empirical question.

Existing research has concentrated primarily on green bond pricing, corporate performance, and the greenium, while evidence on the determinants of alignment and certification remains comparatively limited and heterogeneous (Baker et al., 2018; Flammer, 2021; Rodrigues Loiola et al., 2025; Zerbib, 2019). Moreover, combining CBI-aligned and CBI-certified bonds without separate contrasts can conceal meaningful heterogeneity. This study therefore asks: Which bond-level, issuer-level, and institutional characteristics are associated with external recognition or alignment in corporate green bonds, and do these associations differ between CBI alignment and formal CBI certification?

The study makes three contributions. First, it distinguishes the broad outcome of external recognition or alignment from formal certification and directly compares aligned, certified, and self-labelled bonds. Second, it models the hierarchical structure of 6009 issuances nested within 822 corporate issuers, using an issuer-random-intercept framework with issuance-year, rating, and sector controls and a common complete-case sample for the principal specifications. Third, it evaluates the stability of the findings across alternative size, leverage, profitability, temporal, governance, environmental, and estimator specifications. The evidence shows that issuance amount is the most stable correlate of external recognition or alignment, whereas issuance-market context, issuer size, and operating performance require explicit qualification because their results vary across recognition categories or robustness designs.

The remainder of the article is organised as follows. Section 2 reviews the literature and develops the theoretical framework. Section 3 states the research hypotheses. Section 4 describes the data, variables, and empirical strategy. Section 5 presents the results, Section 6 discusses their theoretical and practical implications, and Section 7 concludes with limitations and directions for future research.

2. Literature Review and Theoretical Framework

2.1. Corporate Credit Risk

Corporate credit risk is directly related to the probability that a firm will fail to honor its financial obligations, determining the cost of capital and the conditions of access to financing. The classical literature already demonstrates that this risk is priced by the market based on economic and financial fundamentals, being reflected in debt spreads and in the financing decisions of firms (Ballester et al., 2024; Caballero et al., 2019; Elton et al., 2001; Hu et al., 2022; Merton, 1974). The centrality of credit risk in the funding dynamics of firms is due to its direct role in determining the marginal cost of debt, defining the investor base, and, in extreme cases, affecting market access (Ballester et al., 2024; Karas et al., 2025). The literature shows that small changes in risk perception can significantly impact the spreads investors require, reflecting not only default probability but also the premiums for systematic risk and liquidity (Collin-Dufresne et al., 2001; Elton et al., 2001; Iannotta et al., 2019).

In general, corporate decisions are shaped by concerns about credit ratings, as downgrades may increase financing costs, limit market access, and trigger adverse contractual clauses (Kisgen, 2006). Additional evidence indicates that credit risk affects firms’ capital structure and refinancing capacity, particularly during periods of financial stress, when investor risk aversion intensifies (Hung et al., 2017; Karas et al., 2025). In this process, credit ratings play an important role by synthesising information on leverage, liquidity, profitability, and exposure to sectoral and macroeconomic risks into an observable metric (Bongaerts et al., 2012; Chodnicka-Jaworska, 2021; Kladakis & Skouralis, 2024).

With regard to green bonds, evidence indicates that environmental, social, and governance (ESG) aspects can influence both perceived risk and the cost of debt, suggesting that environmental performance and the credibility of sustainable initiatives become, although heterogeneously, part of credit risk assessment (Höck et al., 2020; Oikonomou et al., 2014). Complementary evidence from NYSE-listed firms indicates that green-finance initiatives and emission-reduction policies can improve accounting-based performance, whereas environmental investments may impose short-term costs (Ighrarah & Khalifa, 2025).

At the time of issuance, spreads and credit ratings reflect traditional financial factors together with qualitative governance and sustainability information (Capelli et al., 2021; Christensen et al., 2021; Elton et al., 2001; Tang & Zhang, 2020). External recognition, independent review, and formal certification may reduce environmental information asymmetry by increasing the verifiability of the use-of-proceeds framework. Their credibility-enhancing roles are related, but the mechanisms differ in procedural stringency and should not be treated as interchangeable (Bachelet et al., 2019; Flammer, 2021; Hyun et al., 2023; Yu et al., 2024).

2.2. Information Asymmetry and Signalling Theory

Information asymmetry occurs when the relevant attributes of assets or issuers are not fully observable to investors. Under these circumstances, adverse selection tends to intensify, with direct implications for price formation, liquidity, and required risk premiums (Choi, 2018; Lee, 2021). The recent literature has emphasised that even in relatively developed markets, informational heterogeneity remains significant and affects investment decisions and the cost of capital, especially for assets associated with greater complexity or non-financial dimensions (Biais et al., 2015; George et al., 2016; Goldstein & Yang, 2017).

It is in this context that signalling theory applies. The core idea is that better-informed agents may, under certain conditions, transmit their quality to the market through costly, verifiable signals. Recent studies advance this discussion by examining more realistic settings where multiple signals coexist, and credibility depends not just on cost but also on verifiability and consistency over time (Bafera & Kleinert, 2023; Baier et al., 2022; Connelly et al., 2025). In practical terms, weak or low-cost signals tend to lose informational content, while more robust signals help distinguish higher-quality issuers, allowing the market to incorporate this differentiation into asset pricing (Connelly et al., 2025).

Another aspect that gains relevance in this literature is the role of intermediaries and independent verification mechanisms. When information is difficult to observe or requires specific technical knowledge, the involvement of third parties can substantially increase the credibility of the signals issued. By reducing uncertainty about the accuracy of disclosed information, these agents help mitigate informational asymmetry and make the price formation process more informative (Biais et al., 2015; Gao et al., 2020). In this market, the environmental quality of financed projects and the issuer’s commitment to sustainable practices are not fully observable to investors, which heightens adverse-selection and greenwashing concerns (Huynh et al., 2022). Beyond signalling, green bond issuance may also influence firm-level outcomes by promoting green innovation and sustainability-oriented management and by alleviating financing constraints (Wang et al., 2026). In this scenario, alignment, certification, and third-party verification mechanisms provide independent validation of environmental attributes and reduce informational opacity (Bachelet et al., 2019; Flammer, 2021; Tang & Zhang, 2020).

More recent empirical evidence suggests that recognition and verification mechanisms are associated with greater perceived credibility and, in some cases, with better financing conditions, such as lower spreads or greater institutional-investor demand (Copelovitch et al., 2018). These patterns are consistent with signalling theory, but they may also reflect self-selection if issuers with better prior environmental performance or lower risk are more likely to seek stronger verification. Observed associations must therefore be interpreted cautiously rather than as causal effects of certification (Carlson & Palmer, 2016; Richards et al., 2017).

2.3. Institutional and Sectoral Context

The finance literature has highlighted that the institutional environment directly influences the quality of information available to investors and, consequently, market efficiency (Çam & Özer, 2021; Eldomiaty et al., 2023; Fiechter & Novotny-Farkas, 2017). More robust institutional structures, characterised by stronger investor protection, effective regulatory enforcement, and high disclosure standards, tend to reduce informational asymmetry and improve the accuracy of asset pricing (Bushman et al., 2004; Porta et al., 1998).

Conversely, weaker institutional contexts, in which transparency is limited and monitoring mechanisms are less effective, increase informational uncertainty and raise the cost of capital, reflecting greater risk perception among investors (Fernández-Ramos & Tamayo, 2017; Goldstein & Yang, 2017; Hail & Leuz, 2006; Hartwell & Malinowska, 2019). In this scenario, the informational value of external verification mechanisms is not uniform but depends on the level of institutional quality. In environments of high regulatory quality, the information disclosed by firms tends to be more reliable, so additional verification may play a more incremental role, reinforcing already established disclosure and monitoring mechanisms (Baalouch et al., 2019; Cahan et al., 2016; Lisowsky et al., 2017). In contrast, in less mature institutional contexts, third-party validation becomes more relevant by reducing uncertainty about information accuracy and providing an additional mechanism of market discipline (Biais et al., 2015; Bouvard & Lévy, 2018; Desai, 2018; Lamin & Livanis, 2020).

Another important aspect analysed by capital providers is sectoral characteristics. Sectors with a higher intensity of intangible assets, greater operational complexity, or greater exposure to regulatory and environmental risks tend to exhibit greater informational opacity, thus increasing the relevance of mechanisms that enhance transparency (Christensen et al., 2021; Fosu et al., 2018; Leuz & Wysocki, 2016). Additionally, sectors subject to greater public and regulatory scrutiny face stronger incentives to adopt practices that reinforce their legitimacy before investors and other stakeholders, which may be reflected in greater use of verification and disclosure instruments (Gipper et al., 2025; Mura et al., 2018; Roszkowska-Menkes et al., 2024).

The interaction between the institutional environment and sectoral characteristics suggests that the effects of informational mechanisms are not uniform but vary across contexts. In particular, the ability of a mechanism to reduce informational asymmetry and affect risk perception depends on both institutional quality and the nature of the issuer’s economic activities, which justifies empirical approaches that consider multiple levels of analysis (Ballester et al., 2024; Tröster & Hiete, 2018).

The non-financial nature of environmental attributes intensifies the dependence on institutional and verification mechanisms to ensure the credibility of disclosed information. Evidence indicates that the effectiveness of certifications and external reviews varies across countries/regions and sectors, reflecting differences in regulatory quality, the degree of standardisation, and market maturity (Flammer, 2021; Oya et al., 2018; Tröster & Hiete, 2018; Wolff & Schweinle, 2022).

For green bonds, the informational value of external recognition may vary across jurisdictions. In more developed institutional settings, it can complement established disclosure systems, whereas in less transparent markets, stronger third-party verification may play a more substantive uncertainty-reduction role (Mutarindwa et al., 2024; Yu et al., 2024). Issuers in environmentally intensive industries such as energy, manufacturing, and transportation also face greater scrutiny and may benefit more from verifiable recognition mechanisms (Bai, 2025; Block et al., 2024; Fatica & Panzica, 2021; Hassan et al., 2020). Formal certification may therefore strengthen environmental credibility, but its procedural meaning remains distinct from alignment (Prajogo et al., 2016; Tang & Zhang, 2020; Zerbib, 2019; Zhang et al., 2021).

Sectoral patterns in the green bond market operate through two related but distinct margins: pricing and issuance. On the pricing margin, Fatica, Panzica, and Rancan find a greenium for non-financial corporations but not for financial institutions, a difference that may reflect the greater ability of non-financial firms to link bond proceeds transparently to identifiable environmental projects (Fatica et al., 2021). On the issuance margin, Flammer shows that corporate green bonds are more prevalent in industries in which environmental issues are financially material to firms’ operations (Flammer, 2021). Löffler, Petreski, and Stephan document that utilities, real estate, and construction account for larger shares of green than conventional bonds, although their multivariate propensity model identifies no single sector with a significantly higher or lower probability of green issuance (Löffler et al., 2021). The literature therefore supports sectoral heterogeneity but does not establish a stable or universally applicable category of ’greenium sectors’; reported patterns depend on issuer type, market, period, sector classification, and whether greenium is measured in the primary or secondary market.

This distinction also defines the scope of the present study. A sector-specific greenium is a relative pricing outcome conditional on bond issuance, whereas the propensity to issue a green rather than a conventional bond concerns a different extensive-margin decision. Because the database contains green bonds only, it does not provide the conventional bond or non-issuer comparison group required to estimate green bond issuance propensity. The existing sector indicators therefore control for sectoral heterogeneity in external recognition or alignment among issued green bonds; they are not interpreted as estimates of the likelihood that firms in a given sector issue green bonds. A separate ’greenium-sector’ indicator was not constructed because the literature does not provide a consistent ex ante classification and such an indicator would not identify issuance propensity in the present conditional sample.

It is worth noting that the recent evolution of the international regulatory framework, with initiatives to standardise and harmonise environmental information, has reinforced the role of institutions in shaping investors’ expectations (Bernini & Rosa, 2024; Fu et al., 2023; Lebelle et al., 2020). In this way, greater regulatory harmonisation tends to increase the reliability of the signals associated with green bonds and reduce informational heterogeneity across jurisdictions, with direct implications for risk assessment and the cost of capital (Christensen et al., 2021; Fatica & Panzica, 2021; Goldstein & Yang, 2017; Hyun et al., 2023).

3. Research Hypotheses

The hypotheses refer to the broad binary outcome of external recognition or alignment—CBI-aligned or CBI-certified bonds relative to self-labelled bonds. They do not equate alignment with formal certification. The directional predictions follow the financial-capacity interpretation of signalling theory, while the competing credibility-need mechanism is considered when interpreting results that depart from these predictions.

Operating performance may be related to an issuer’s capacity or incentive to obtain external recognition. However, the financial-capacity and credibility-need mechanisms imply competing directions, and the literature does not support an unambiguous ex ante prediction. We therefore do not formulate a directional hypothesis for operating performance. Mean operating margin is retained as an exploratory issuer-level covariate, and its estimates are interpreted as associational and assessed across alternative temporal and profitability specifications.

Transaction scale and issuer organisational scale represent distinct mechanisms. At the bond level, a larger offering can spread the fixed costs of external review, documentation, and recognition over a greater financing volume and may attract greater investor attention and scrutiny. By contrast, issuer organisational scale concerns persistent reporting, compliance, and sustainability capacity. Issuance amount is not a proxy for firm size: large firms may undertake relatively small offerings, while smaller firms may issue comparatively large bonds. Both variables are included simultaneously so that the transaction-scale association is estimated conditional on issuer size and the issuer-size association is estimated conditional on issuance amount. Table 1 summarises the research hypotheses and their theoretical expectations.

Table 1. Research hypotheses and theoretical expectations.

Hypothesis	Description	Relationship/Association	Theoretical Assumption
H1—Credit quality	Issuers with stronger credit quality and lower financial leverage have greater capacity to adopt external recognition or alignment mechanisms.	Stronger ratings and lower leverage increase the probability of external recognition or alignment.	(Ballester et al., 2024; Flammer, 2019; García et al., 2023; Huynh et al., 2022; Kapraun et al., 2019; Pietsch & Salakhova, 2022)
H2—Transaction scale (issuance amount)	Larger offerings can spread fixed recognition costs over a greater financing volume and attract greater visibility and scrutiny.	Conditional on issuer size and the remaining covariates, larger offerings have a higher probability of external recognition or alignment.	(Flammer, 2019; Kapraun et al., 2019; Löffler et al., 2021; Pietsch & Salakhova, 2022; Tang & Zhang, 2020)
H3—Institutional environment and issuance-market type	The institutional and regulatory maturity of the market in which the bond is issued increases demand for standardised and externally recognisable green instruments.	Green bonds issued in developed markets have a higher probability of external recognition or alignment.	(Ballester et al., 2024; Kapraun et al., 2019; Lebelle et al., 2020; Löffler et al., 2021; Pietsch & Salakhova, 2022)
H4—Sectoral differences	Issuers in sectors facing greater environmental exposure and scrutiny have stronger incentives to adopt external recognition or alignment mechanisms.	Environmentally scrutinised sectors have a higher probability of external recognition or alignment.	(Ballester et al., 2024; Flammer, 2019; García et al., 2023; Janda et al., 2023; Kapraun et al., 2019; Löffler et al., 2021; Pietsch & Salakhova, 2022; Tang & Zhang, 2020)
H5—Issuer organisational scale (asset size)	Larger issuers may possess more extensive reporting systems, specialised sustainability personnel, and compliance resources with which to implement recognised frameworks.	Conditional on issuance amount and the remaining covariates, greater issuer asset size increases the probability of external recognition or alignment.	(García et al., 2023; Kapraun et al., 2019; Löffler et al., 2021; Pietsch & Salakhova, 2022)

Because CBI alignment and formal CBI certification differ substantively and the direction of their contrast is not unambiguously implied by the existing literature, their heterogeneity is addressed through an exploratory research question rather than a post hoc hypothesis: Do the associations identified for the broad outcome differ across CBI-aligned, CBI-certified, and self-labelled bonds?

4. Materials and Methods

4.1. Data and Sample Construction

The database covers corporate green bonds issued between 2016 and 2025, a period that begins after the Paris Agreement and captures the expansion and institutional consolidation of the international green bond market. Sovereign, municipal, multilateral, and other public-sector issuers were excluded because their financing structures, pricing determinants, and institutional objectives differ from those of corporate issuers (Tomczak, 2024). Each observation represents one bond issuance, and multiple bonds may therefore be nested within the same issuer.

The full database contains 6009 green bonds issued by 822 companies. The bonds were classified into three mutually exclusive categories: 3874 CBI-aligned bonds, 256 CBI-certified bonds, and 1879 self-labelled bonds. CBI-aligned bonds are treated as externally recognised or aligned with CBI criteria but not as equivalent to bonds that obtained formal CBI certification. Self-labelled bonds rely on the issuer’s own green designation without either CBI classification. Table 2 reports both the full database and the common complete-case sample used in the main model.

Table 2. Green bond recognition categories in the full database and main analytical sample.

Recognition Category	Full Database	Main Sample
CBI-aligned	3874	3114
CBI-certified	256	213
Self-labelled	1879	1548
Total bonds	6009	4875
Issuers	822	682

The principal models use the same complete-case sample of 4875 bonds issued by 682 firms. Of the 1134 bonds excluded from the full database, 1119 lack all three principal issuer-level financial variables, and 15 lack only operating margin. No missing financial values were imputed. To document the robustness samples, Appendix A, Table A1 reports descriptive statistics for the alternative variables and restricted samples, while Appendix A, Table A2 reports a maximum-sample baseline using all 6009 bonds and a restricted baseline using the same 4875 bonds as the full main model.

The issuance-market indicator is assigned at the bond level from the database’s Country of Issue field and may therefore vary across bonds issued by the same firm. For country and regional entries, the developed/emerging grouping follows the analytical classification used in the IMF World Economic Outlook, which considers per-capita income, export diversification, and integration into the global financial system rather than a single mechanical threshold. When the provider records “Eurobond” instead of a country, that issuance-market label is retained in the developed group under the database’s original coding; it must not be interpreted as the issuer’s domicile. The indicator is consequently a broad issuance-market proxy rather than a direct measure of issuer location, regulatory quality, or institutional strength. Table 3 reports the groups and corrects the emerging-market count presented in the previous version.

Table 3. Classification of country/region of issue and provider market labels by market type.

Market Type	Bonds	Country/Region of Issue or Provider Label
Emerging	1572	Argentina, Brazil, Chile, China (Mainland), Colombia, Dominican Republic, India, Kenya, Malaysia, Mexico, Morocco, Panama, Philippines, Poland, Romania, South Africa, Thailand, and Turkey.
Developed	4437	Australia, Austria, Belgium, Canada, Denmark, Eurobond, Finland, France, Germany, Greece, Hong Kong, Italy, Japan, Latvia, Liechtenstein, New Zealand, Norway, Portugal, Singapore, Slovakia, South Korea, Spain, Sweden, Switzerland, Taiwan, the United Kingdom, and the United States.

Note: counts refer to the full 6009-bond database. Country entries follow the IMF World Economic Outlook analytical grouping; see https://data.imf.org/Datasets/WEO/Groups-and-Aggregates-April-2025 (accessed on 1 July 2026). “Eurobond” is a provider issuance-market label retained in the developed group and is not an issuer-domicile category.

4.2. Variable Construction

The main dependent variable, external recognition/alignment, equals one for CBI-aligned or CBI-certified bonds and zero for self-labelled bonds. This binary outcome captures whether a bond has CBI recognition or alignment beyond the issuer’s own label; it is not described as a measure of formal certification. Heterogeneity analyses preserve the three original categories and separately compare aligned with self-labelled bonds (R5), certified with aligned bonds (R6), and all three outcomes jointly (R7).

Transaction scale is measured at the bond level by the natural logarithm of the amount issued in U.S. dollars and may vary across bonds issued by the same firm. Issuer organisational scale is measured by the mean of annual log total assets and represents a structural issuer-level profile; the median is used only in robustness specification R1. Both measures are entered simultaneously to estimate each association conditional on the other. Issuance year is represented by indicators for 2017–2025, with 2016 as the reference year.

The principal financial variables are issuer-level structural profiles calculated from annual observations over 2016–2025. They should therefore be interpreted as medium-term issuer characteristics, not as contemporaneous financial conditions at each issuance date. Financial leverage is measured by mean total debt divided by total assets (TDTA); mean total liabilities divided by total assets (TLTA) is used only in R2. Both measures are expressed as ratios on a consistent scale. Values slightly above one can occur when reported debt or liabilities exceed assets; gross ratio values outside the audit interval from zero to ten were treated as unavailable. Operating profitability is measured by the provider-reported operating margin, which relates operating income to revenue, and is averaged across available annual observations. Negative margins were retained because they represent economically meaningful operating losses. Annual ratio data were screened for gross invalid values and winsorised within year at the first and ninety-ninth percentiles before issuer-level summaries were constructed.

The temporal robustness model replaces the structural financial measures with log total assets, TDTA, and operating margin observed in the year preceding issuance ($t-1$). Bonds issued in 2016 are not assigned unavailable 2015 values. Alternative profitability tests replace operating margin with return on invested capital (ROIC) or return on equity (ROE) and compare each alternative with an operating-margin baseline estimated on exactly the same restricted sample. ROIC source values reported in percentage points were divided by 100; ROE was already expressed as a ratio. Zero values in the annual ROIC and ROE fields were treated as unavailable because, under the source-data coding used in this extract, they represented absent observations rather than observed zero returns.

Credit quality is represented by the grouped Fitch ratings available in the archived integrated Refinitiv research extract. S&P ratings were not available in that extract. Constructing a temporally comparable S&P measure would require access to an additional proprietary historical-ratings source and the development of a new instrument- and issuer-level linkage matching the applicable rating to each bond’s issuance date. Current ratings collected ex post were not used because they are not methodologically equivalent to ratings observed at issuance and could introduce temporal misclassification and look-ahead bias. Accordingly, an S&P specification was not estimated within the archived-data design of this study. Fitch observations were grouped into the categories shown in Table 4; missing, withdrawn, and unavailable ratings form an explicit category so that the rating control does not mechanically discard unrated bonds. The main specification also includes eight issuer-sector groups.

Table 4. Fitch rating groups used in the models.

Original Rating	Analytical Group
AAA, AA+, AA, AA−	High investment grade
A+, A, A−	Upper medium grade
BBB+, BBB, BBB−	Lower investment grade
BB+, BB, BB−, B+, B, CCC	Speculative grade
NR, RD, WD, or unavailable	No rating/withdrawn

Table 5 summarises the variables used in the main and robustness models.

Table 5. Variables in the main and robustness models.

Role/Level	Variable	Definition
Outcome/bond	External recognition/alignment	One for CBI-aligned or CBI-certified; zero for self-labelled.
Outcome/bond	Three-category recognition type	CBI-aligned, CBI-certified, or self-labelled; used in R5–R7.
Predictor/bond	Transaction scale: ln(Amount issued)	Natural logarithm of the bond’s issuance amount in U.S. dollars.
Control/bond	Issuance year	Indicators for 2017–2025; 2016 is the reference year.
Predictor/bond	Developed issuance market	One for a developed and zero for an emerging issuance market, based on the Country of Issue field.
Predictor/issuer	Issuer organisational scale: mean annual ln(Total assets)	Main structural proxy for issuer size; median used in R1.
Predictor/issuer	Mean TDTA	Mean total debt divided by total assets; main leverage proxy.
Exploratory covariate/issuer	Mean operating margin	Structural operating-performance measure retained without a directional hypothesis.
Controls/issuer	Fitch rating and sector	Grouped Fitch rating and eight sector indicators.
Robustness/issuer	TLTA, ROIC, ROE, and $t-1$ measures	Alternative leverage, profitability, and issuance-timed financial measures.
Robustness/issuer	Governance Pillar Score at $t-1$	Governance score rescaled so that one unit equals ten index points.
Robustness/issuer	Environmental Pillar Score at $t-1$	Environmental score rescaled so that one unit equals ten index points.

4.3. Empirical Specification

The data are hierarchical because several bonds may be issued by the same company. The main analysis therefore uses a multilevel logistic model with a random intercept for issuer. Let $Y_{ij}$ equal one when bond i issued by firm j is externally recognised or aligned and zero when it is self-labelled. The model is specified in Equation (1):

$$\begin{array}{cc}\hfill \mathrm{logit}\left[\mathrm{Pr}(Y_{ij}=1\mid u_j)\right]& ={\beta }_0+{\beta }_{1}Develope{d}_{ij}+{\beta }_2\ln \left(Amoun{t}_{ij}\right)\hfill \\ \hfill & + {\beta }_{3}MeanLnAsset{s}_j+{\beta }_{4}MeanTDT{A}_j\hfill \\ \hfill & + {\beta }_{5}MeanOperatingMargi{n}_j+{\gamma }^{\prime }Ratin{g}_j\hfill \\ \hfill & + {\delta }^{\prime }Secto{r}_j+{\displaystyle \sum _{t=2017}^{2025}}{\tau }_t\mathbb{1}(Yea{r}_{ij}=t)+u_j,\hfill \end{array}$$

(1)

where $Develope{d}_{ij}$ is the bond-level issuance-market indicator, and $u_j\sim N(0,{\sigma }_u^2)$ is the issuer-specific random intercept. The specification assumes that the issuer intercept is approximately normally distributed, that bond observations are conditionally independent after covariates and the random intercept are taken into account, and that the random intercept is not correlated with included regressors through an unmodelled mechanism. These assumptions permit efficient modelling of repeated issuances but do not eliminate the possibility of residual firm-level confounding; this limitation is considered explicitly below.

Models were estimated in Stata 19 using maximum likelihood and 16 adaptive Gauss–Hermite integration points. The results are reported as odds ratios (ORs), with 95% confidence intervals. Model adequacy is documented through convergence status, the issuer-level intraclass correlation coefficient (ICC), the likelihood-ratio test comparing the random-intercept specification with ordinary logistic regression, log-likelihood, AIC, BIC, the number of bonds, and the number of issuers. The ICC uses the conventional level-1 logistic variance ${\pi }^2/3$. A pooled logit with issuer-clustered standard errors is reported as an estimator-sensitivity check.

Because issuance amount and issuer size represent distinct scale mechanisms, their empirical overlap was examined in the common main-model sample. Their bond-level Pearson correlation is positive but modest ($r=0.242$, $p<0.001$, $N=4875$), corresponding to approximately 5.9% shared variance. Multicollinearity was further assessed through auxiliary linear regressions based on the complete fixed-effect design matrix, including rating, sector, and issuance-year indicators. The VIFs were 1.24 for developed-market status, 1.21 for ln(issuance amount), 1.47 for mean annual ln(total assets), 1.09 for mean TDTA, and 1.36 for mean operating margin; corresponding tolerance values ranged from 0.68 to 0.92. All focal VIFs are well below the conservative threshold of 5, supporting the simultaneous inclusion of transaction scale and issuer organisational scale.

4.4. Robustness, Heterogeneity, and Post-Estimation Analyses

The robustness design changes one measurement or estimator feature at a time. R1 replaces mean annual log assets with the annual median. R2 replaces TDTA with TLTA; the two leverage proxies are not entered simultaneously. The temporal model uses financial variables at $t-1$ and is paired with a structural-profile model estimated on the same 4216 bonds. The estimator check uses pooled logit with issuer-clustered standard errors. R3 replaces the exploratory operating-margin covariate with ROIC, and an additional matched-sample model uses ROE. R8 and R9 add, respectively, the Governance Pillar Score and Environmental Pillar Score at $t-1$ to paired baselines estimated on the same observations as each score-augmented model.

Recognition-category heterogeneity is evaluated in three ways. R5 estimates aligned versus self-labelled bonds with the multilevel binary model. R6 estimates certified versus aligned bonds; because the corresponding issuer-random-intercept model was numerically unstable under the sparse formal-certification outcome, the retained R6 specification is a pooled logistic model with issuer-clustered standard errors and parsimonious grouped rating and sector controls. R7 is an additional multilevel multinomial model that uses self-labelled bonds as the reference category and a shared issuer-level random effect. R7 is interpreted as supplementary heterogeneity evidence rather than as a replacement for the main binary model.

Average marginal effects (AMEs) for the five focal predictors were calculated after the main model with the Stata command margins, dydx(…) predict(mu fixedonly). Thus, predicted means use the fixed portion of the model and set the issuer random intercept to zero; observation-specific marginal effects are then averaged over the 4875-bond estimation sample. Delta-method 95% confidence intervals are reported.

Before estimating the exploratory governance robustness model, we audited the ESG and governance indicators available in the database: the Governance Pillar Score, CSR Strategy Score, Management Score, Shareholders Score, and annual Governance Pillar Scores for 2016–2025. For each measure, we examined its scale and distribution, missing-data pattern, coverage across bonds and issuers, consistency within issuers, and availability by issuance year and green bond recognition category.

Correlations among the governance indicators were calculated using one observation per issuer to avoid assigning disproportionate weight to firms with multiple bond issues. Based on temporal alignment, conceptual coverage, and redundancy among the alternative indicators, the annual Governance Pillar Score at $t-1$ was selected for the exploratory robustness analysis.

A parallel environmental robustness check uses the annual Environmental Pillar Score measured at $t-1$. Exact zeros in the annual score fields were treated as unavailable under the source-data coding used in the archived extract. The score was divided by ten so that its odds ratio represents a ten-point increase. R9 compares a baseline containing the $t-1$ financial measures with an environmental-score model estimated on the identical sample of 1421 bonds issued by 224 firms.

5. Results

5.1. Sample Characteristics

Table 6 describes the common analytical sample. Of the 4875 bonds, 3327 (68.25%) have external recognition or alignment and 1548 (31.75%) are self-labelled. The broad recognised/aligned group comprises 3114 CBI-aligned bonds and 213 formally CBI-certified bonds. Developed issuance markets account for 74.65% of the main sample. Financial issuers and Utilities and Energy together represent 61.25% of observations, and 78.38% of the bonds fall in the no-rating/withdrawn Fitch category.

Table 6. Frequency distribution in the common analytical sample.

Variable	Category	N	%
Recognition category	CBI-aligned	3114	63.88
	CBI-certified	213	4.37
	Self-labelled	1548	31.75
Binary outcome	External recognition/alignment	3327	68.25
	Self-labelled	1548	31.75
Issuance-market type	Developed	3639	74.65
	Emerging	1236	25.35
Issuer sector	Financial	1838	37.70
	Utilities and Energy	1148	23.55
	Real Estate and Construction	702	14.40
	Consumer and Services	467	9.58
	Industrial and Manufacturing	308	6.32
	Transport and Logistics	305	6.26
	Others	89	1.83
	Technology and Telecom	18	0.37
Fitch rating group	High investment grade	133	2.73
	Upper medium grade	480	9.85
	Lower investment grade	376	7.71
	Speculative grade	65	1.33
	No rating/withdrawn	3821	78.38

Because the binary outcome and issuance-market type are already reported as counts and percentages in Table 6, Table 7 is restricted to continuous variables. It reports medians and interquartile ranges in addition to means and standard deviations and eliminates the coefficient of variation used in the previous version. The bond-level issuance amount is summarised across 4875 issuances, whereas structural financial profiles are summarised once for each of the 682 issuers to avoid mechanically weighting firms by their number of bonds. TDTA and operating margin are consistently expressed as ratios rather than percentages.

Table 7. Descriptive statistics for the continuous main analytical variables.

Unit	Variable	N	Mean	SD	Median	Q1	Q3	Min.	Max.
Bonds	ln(Amount issued)	4875	20.006	2.330	20.030	19.337	20.723	6.908	27.216
Issuers	Mean annual ln(Total assets)	682	22.449	2.367	22.649	21.435	23.853	10.789	28.707
Issuers	Mean TDTA	682	0.353	0.242	0.357	0.182	0.495	0.000	1.035
Issuers	Mean operating margin	682	0.178	0.330	0.124	0.020	0.338	−1.957	1.169

Note: financial ratios have not been multiplied by 100.

The three recognition categories also have distinct observed profiles (Table 8). Formally certified bonds are less frequently issued in developed markets and are issued by smaller firms on average than either aligned or self-labelled bonds. Aligned bonds have the highest average issuance amount and operating margin. These unconditional comparisons motivate the category-specific models but should not be interpreted as adjusted effects.

Table 8. Mean characteristics by green bond recognition category.

Category	N	Share Issued in Developed Markets	ln(Amount)	Mean ln(Assets)	Mean TDTA	Mean Operating Margin
CBI-aligned	3114	0.756	20.208	22.878	0.363	0.274
CBI-certified	213	0.521	19.991	21.292	0.351	0.235
Self-labelled	1548	0.759	19.600	22.944	0.408	0.181

Note: the developed-market entries are proportions. Statistics are bond-weighted; an issuer may contribute several bonds and may appear in more than one recognition category.

5.2. Main Multilevel Logistic Model

Table 9 presents the main model for external recognition/alignment versus self-labelling. Issuance in a developed market is positively associated with the outcome (OR = 4.608, 95% CI [2.102, 10.102], $p<0.001$). Issuance amount is also positive and statistically significant (OR = 1.303, 95% CI [1.193, 1.423], $p<0.001$), so a one-unit increase in log issuance amount is associated with approximately 30.3% higher odds, conditional on the remaining covariates and issuer effect.

Table 9. Main multilevel logistic model of external recognition/alignment.

Predictor/Statistic	OR	95% CI Lower	95% CI Upper	p-Value
Developed issuance market	4.608	2.102	10.102	<$0.001$
ln(Issuance amount)	1.303	1.193	1.423	<$0.001$
Mean annual ln(Total assets)	0.732	0.602	0.890	0.002
Mean TDTA	2.785	0.568	13.662	0.207
Mean operating margin	8.999	2.642	30.658	<$0.001$
Fitch rating indicators	Yes
Sector indicators	Yes
Issuance-year indicators	Yes
Bonds	4875
Issuers	682
ICC	0.872	0.837	0.901
Log-likelihood	−1879.385
AIC	3812.770
BIC	3988.051
Random-intercept LR test				<$0.001$

Note: OR = odds ratio. Emerging market, high-investment-grade Fitch rating, Financial sector, and 2016 are the reference categories. The model uses an issuer random intercept and 16 adaptive Gauss–Hermite integration points.

Issuer organisational scale is negatively associated with the broad outcome (OR = 0.732, 95% CI [0.602, 0.890], $p=0.002$), contrary to H5. Mean operating margin, retained as an exploratory rather than confirmatory covariate, is positively associated with recognition/alignment (OR = 8.999, 95% CI [2.642, 30.658], $p<0.001$), although the OR corresponds to a full one-unit change in a ratio and should therefore be interpreted with its measurement scale in mind. Mean TDTA is not statistically significant (OR = 2.785, $p=0.207$), and none of the individual Fitch rating indicators is significant; rating effects are not consistent across the robustness analyses.

Sector differences remain after adjustment. Relative to Financial issuers, Utilities and Energy has higher odds of external recognition/alignment (OR = 6.608, $p<0.001$), whereas Industrial and Manufacturing (OR = 0.144, $p=0.005$) and Others (OR = 0.041, $p=0.031$) have lower odds. The other sector contrasts are not significant at the 5% level. The ICC is 0.872 (95% CI [0.837, 0.901]), indicating that a large share of latent outcome variation lies between issuers. The likelihood-ratio test against an ordinary logistic model is significant ($p<0.001$), supporting the issuer-random-intercept specification. The model converged under the stated integration settings.

5.3. Robustness of the Binary Specification

Appendix A, Table A3 summarises R1, R2, the temporal model, and the clustered-logit check. Replacing mean with median log assets in R1 produces nearly identical results; median issuer size remains negative (OR = 0.736, $p=0.003$). Replacing TDTA with TLTA in R2 does not generate a significant leverage effect (OR = 5.285, $p=0.077$). The two ratios have a positive but modest correlation ($r=0.211$, $p<0.001$, $N=4890$) and are never entered simultaneously. The issuance-amount association remains positive in every specification.

In the $t-1$ model, issuance amount (OR = 1.309, $p<0.001$) and issuance in a developed market (OR = 5.761, $p<0.001$) remain significant. Issuer size weakens to the 10% level (OR = 0.869, $p=0.077$), while $t-1$ TDTA (OR = 2.103, $p=0.290$) and $t-1$ operating margin (OR = 0.848, $p=0.643$) are not significant. In the pooled logit with issuer-clustered standard errors, issuance amount remains positive (OR = 1.124, $p=0.041$) and operating margin remains positive (OR = 2.168, $p=0.010$), whereas issuance-market type and issuer size are not significant. These differences show that issuance-market type, firm size, and profitability are more estimator- or measurement-sensitive than issuance amount.

The alternative operating-performance tests in Appendix A, Table A4 use matched restricted samples. Neither ROIC (OR = 28.972, $p=0.605$; 1482 bonds) nor ROE (OR = 5.567, $p=0.391$; 1964 bonds) is statistically significant, and both estimates have very wide confidence intervals. Their paired operating-margin baselines are likewise imprecise in these smaller samples. These checks support retaining operating margin as the principal exploratory structural measure but do not establish a general operating-performance effect across alternative metrics.

5.4. Recognition-Category Heterogeneity

R5 and R6 directly address the distinction between alignment and formal certification (Table 10). In R5, aligned bonds are compared with self-labelled bonds using the multilevel model. Developed-market status (OR = 6.443, $p<0.001$), issuance amount (OR = 1.299, $p<0.001$), and operating margin (OR = 5.560, $p=0.007$) are positively associated with alignment; issuer size and TDTA are not significant.

Table 10. Binary recognition-category heterogeneity analyses.

Predictor	R5: Aligned vs. Self-Labelled			R6: Certified vs. Aligned
	OR	95% CI	p	OR	95% CI	p
Developed issuance market	6.443	[2.839, 14.622]	<$0.001$	0.372	[0.159, 0.868]	0.022
ln(Issuance amount)	1.299	[1.189, 1.419]	<$0.001$	1.205	[1.041, 1.395]	0.013
Mean annual ln(Total assets)	0.902	[0.742, 1.098]	0.305	0.815	[0.708, 0.939]	0.004
Mean TDTA	2.116	[0.409, 10.936]	0.371	0.480	[0.140, 1.653]	0.245
Mean operating margin	5.560	[1.606, 19.247]	0.007	1.302	[0.597, 2.837]	0.507
Estimator	Multilevel logit			Pooled logit; issuer-clustered SE
Bonds	4662			3327
Issuers/clusters	618			490

Note: R5 includes full rating, sector, and issuance-year controls. R6 uses parsimonious grouped rating and sector controls and issuance-year indicators because formal certification is sparse. The unstable R6 multilevel estimate is not reported.

R6 compares certified with aligned bonds using pooled logit with issuer-clustered standard errors. Relative to aligned bonds, certified bonds have lower odds of being issued in developed markets (OR = 0.372, $p=0.022$) and by larger issuers (OR = 0.815, $p=0.004$), but higher issuance amounts (OR = 1.205, $p=0.013$). TDTA and operating margin do not distinguish the two categories. Thus, variables that predict entry into the broad recognition/alignment group do not necessarily predict the more demanding transition from alignment to formal certification.

The supplementary multilevel multinomial analysis in Appendix A, Table A5 confirms this heterogeneity. Relative to self-labelled bonds, issuance amount is positively associated with both alignment (RRR = 1.293, $p<0.001$) and certification (RRR = 1.541, $p<0.001$). Issuance in a developed market is also positive in both comparisons, while issuer size is negative. The formal certified-versus-aligned contrast is negative for issuance in a developed market (RRR ratio = 0.391, $p<0.001$), issuer size (0.812, $p<0.001$), and TDTA (0.479, $p=0.012$), and it is positive for issuance amount (1.192, $p<0.001$). Operating margin does not distinguish certification from alignment ($p=0.148$). Because only eight issuers issued bonds in all three categories, this model is interpreted as additional heterogeneity evidence.

5.5. Average Marginal Effects

Figure 1 reports the AMEs from the fixed portion of the main model. Changing the issuance-market indicator from emerging to developed is associated with a 0.226 increase in predicted probability (95% CI [0.106, 0.347], $p<0.001$). A one-unit increase in log issuance amount is associated with a 0.035 increase (95% CI [0.022, 0.048], $p<0.001$), while a one-unit increase in mean log assets is associated with a 0.041 decrease (95% CI [−0.067, −0.015], $p=0.002$). The AME for TDTA is not significant. The exploratory operating-margin covariate has a positive AME of 0.288 (95% CI [0.139, 0.437], $p<0.001$), again corresponding to a full one-unit change in the ratio. Because the predictors use different units, AME magnitudes should not be mechanically ranked.

Figure 1. Average marginal effects from the main multilevel model, with 95% confidence intervals. Predictions use the fixed portion of the model and set the issuer random intercept to zero.

5.6. Exploratory ESG Analyses

Appendix A, Table A6 adds the Governance Pillar Score at $t-1$ to a paired baseline using 1427 bonds and 225 issuers. A ten-point increase in governance score is not associated with external recognition/alignment (OR = 1.031, 95% CI [0.840, 1.265], $p=0.769$). The focal coefficients change only minimally, and the likelihood-ratio test for adding governance is not significant (${\chi }^2\left(1\right)=0.08$, $p=0.784$). Because this selectively observed ESG sample is approximately 29% of the main sample, the result is exploratory and does not rule out governance effects in a more complete dataset.

Appendix A, Table A7 reports the paired environmental analysis. In the common sample of 1421 bonds from 224 issuers, a ten-point increase in the Environmental Pillar Score at $t-1$ is not associated with external recognition/alignment (OR = 0.950, 95% CI [0.752, 1.200], $p=0.668$). Adding the score does not improve fit relative to the paired baseline (LR ${\chi }^2\left(1\right)=0.18$, $p=0.673$), and the focal estimates change only minimally. The finding is therefore evidence of no detected independent association in this selectively observed subsample, not evidence of a negative environmental-score effect or proof that environmental performance is irrelevant.

6. Discussion

6.1. Transaction Scale and Signalling Capacity

The most consistent finding is the positive association between transaction scale and external recognition or alignment. It appears in the main model, R1, R2, the restricted structural and $t-1$ samples, the clustered logit, R5, R6, and both multinomial outcome equations. Conditional on issuer size, larger transactions attract greater visibility and scrutiny, while the fixed costs of external review or formal certification can be spread over a larger financing volume. This combination makes recognised structures economically more feasible and informationally more valuable in large transactions, supporting H2 and the signalling interpretation (Flammer, 2021; Spence, 1973; Tang & Zhang, 2020).

6.2. Institutional Environment and Category Heterogeneity

Issuance in a developed market is strongly positive in the broad multilevel outcome and in aligned-versus-self-labelled comparisons. This pattern is consistent with, but does not directly measure, deeper institutional-investor demand, more established sustainable-finance practices, and greater recognition of standardised green frameworks. The association disappears in the pooled clustered logit and reverses when formal certification is compared directly with alignment. Because the variable is constructed from the bond’s Country of Issue field, the result should be interpreted as an issuance-market association rather than as evidence about issuer domicile, a direct regulatory effect, or a systematic developed-market preference for formal certification.

The R5–R7 analyses materially change the article’s interpretation. CBI alignment and CBI certification share some correlates, especially transaction scale, but differ with respect to issuance-market type, issuer size, and leverage. Combining them remains useful for the principal question of whether a bond has recognition or alignment beyond self-labelling, but it cannot support claims specifically about formal certification. The category-specific evidence consequently justifies the revised terminology throughout the title, abstract, methods, results, and conclusions.

6.3. Issuer Organisational Scale

Issuer organisational scale is negatively associated with the broad outcome after transaction scale is held constant, contrary to H5 and to a simple resource-capacity argument. One interpretation is that larger firms already have established reputations, analyst coverage, and disclosure systems, which reduce the marginal value of an additional recognition signal. Smaller issuers may obtain greater legitimacy benefits from external alignment. Nevertheless, the size coefficient weakens with $t-1$ measurement and is not significant in the pooled clustered logit; it also differs between R5 and R6. The result is therefore informative but not estimator-invariant. The different signs and stability patterns for issuance amount and issuer assets further confirm that the two variables capture distinct transaction-level and organisational mechanisms.

6.4. Credit Risk and Exploratory Operating Performance

Neither mean TDTA nor the grouped Fitch ratings provides consistent evidence for H1. TLTA is also not significant when used as the sole alternative leverage proxy. These findings do not support either the prediction that financially stronger issuers have systematically greater capacity to obtain recognition or alignment or the competing idea that more indebted issuers use these mechanisms to offset financial-risk perceptions. Environmental recognition or alignment appears to complement rather than substitute for conventional credit-risk assessment.

Mean operating margin is positive in the main structural model, R1, R2, R5, the matched structural sample, and the clustered logit. Because no directional operating-performance hypothesis is retained, this pattern is interpreted as an exploratory association with the issuer’s medium-term operating profile rather than as confirmatory evidence. The association does not survive when performance is measured at $t-1$, and the ROIC and ROE models are too imprecise to establish a broader effect. Accordingly, the finding should not be presented as evidence that contemporaneous operating performance determines recognition or alignment.

6.5. Sectoral Differences

Utilities and Energy shows substantially higher odds than the Financial reference sector, consistent with greater environmental visibility and the importance of recognisable standards for energy-transition projects. Industrial and Manufacturing and the residual Others group show lower odds. These contrasts provide partial rather than uniform support for H4: environmental exposure can increase demand for credibility, but the applicability of recognised taxonomies and the measurability of financed projects also vary across industries.

6.6. Exploratory ESG Scores

Neither the Governance Pillar Score nor the Environmental Pillar Score measured at $t-1$ is independently associated with external recognition/alignment in its respective paired subsample. Adding either score leaves the focal estimates nearly unchanged and does not improve model fit. These analyses reduce concern that the main associations merely reflect the included ESG dimension within the observed subsamples, but their selective coverage—approximately 29% of the main sample—precludes a general conclusion that governance or environmental performance has no role in recognition decisions.

6.7. Theoretical and Practical Implications

The findings refine signalling theory in the green bond setting. Signal adoption depends not only on issuer quality or credibility need but also on transaction scale, issuance-market context, and the type of recognition mechanism. Alignment and formal certification represent different signal intensities and should be modelled accordingly. The evidence also cautions against inferring that recognition or alignment communicates lower default risk: leverage and rating measures are largely unrelated to the broad outcome after controls.

For issuers and advisers, the strongest practical implication is that large transactions are the most natural candidates for external alignment or certification. Smaller issuers may nevertheless obtain substantial reputational benefits when information asymmetry is high. Investors should distinguish a bond identified as CBI-aligned from one that has obtained formal CBI certification and should not treat either status as a substitute for independent credit analysis. Regulators and standard setters can improve market comparability by making the procedural differences among self-labelling, alignment, external review, and formal certification explicit.

Operating performance is not classified as a hypothesis in Table 11. Mean operating margin remains an exploratory covariate: it is positive in the structural-profile models but not at $t-1$, while the ROIC and ROE estimates are imprecise.

Table 11. Summary of hypotheses and findings.

Hypothesis	Main Evidence	Assessment
H1	TDTA/TLTA and Fitch ratings show no consistent association.	Not supported.
H2	Transaction scale is positive in every binary and multinomial specification.	Supported.
H3	Issuance in a developed market is positive in the broad multilevel and R5 models but sensitive to estimator and category contrast.	Qualified support.
H4	Utilities and Energy is positive; Industrial and Manufacturing and Others are negative relative to Financial.	Partially supported.
H5	Issuer organisational scale is negative in the main model and sensitive in temporal/clustered checks.	Not supported in the hypothesised direction.

7. Conclusions

This study examines the bond-level, issuer-level, and institutional correlates of external recognition or alignment in 6009 corporate green bonds issued between 2016 and 2025. The main complete-case analysis uses 4875 bonds from 682 issuers and distinguishes the broad binary outcome from formal CBI certification. This conceptual distinction is essential: CBI-aligned and CBI-certified bonds differ in their verification requirements and do not exhibit identical empirical associations.

Transaction scale is the most stable result. Larger green bond transactions have higher odds of recognition or alignment across alternative measurements, samples, estimators, and outcome definitions, conditional on issuer organizational scale. Issuance in a developed market and issuer organisational scale are also associated with the broad outcome, but their interpretation depends more strongly on whether alignment and certification are separated, whether financial conditions are measured structurally or at $t-1$, and whether issuer heterogeneity is modeled through a random intercept. Mean operating margin is retained as an exploratory association and is likewise measurement-sensitive. Leverage and Fitch rating categories provide no consistent evidence that external recognition or alignment is primarily a response to conventional corporate credit risk.

The results therefore support a more precise interpretation of green bond signaling. External recognition or alignment is associated with transaction scale and issuance-market context, while formal certification represents a distinct and comparatively sparse category. Recognition or alignment may enhance the credibility of environmental claims, but neither should be interpreted as evidence of superior credit quality or as a replacement for conventional risk analysis.

7.1. Limitations

Several limitations qualify the findings. First, the main financial covariates are long-run issuer averages and measure structural financial profiles rather than conditions at the exact issuance date. The $t-1$ analysis improves temporal alignment but has reduced coverage and weaker profitability results. Second, the random-intercept model assumes approximate normality, conditional independence, and no unmodelled correlation between the issuer effect and regressors. The clustered-logit comparison shows that some results are estimator-sensitive, and unobserved firm heterogeneity may remain.

Third, formal CBI certification is rare, which limits within-issuer category variation and makes the multilevel certified-versus-aligned binary model unstable. The multinomial specification uses a shared random effect and is therefore supplementary. Fourth, the rating-agency analysis is limited to the Fitch information contained in the archived research extract; a comparable S&P analysis would require a separate proprietary historical source matched consistently to each issuance date. Fifth, ESG data are selectively observed: the governance and environmental checks use 1427 and 1421 bonds, respectively. Sixth, the developed/emerging indicator is constructed from the Country of Issue field, includes ‘Eurobond’ as a provider market label, and is not a direct measure of issuer domicile or regulatory quality; it cannot capture all institutional differences within either group. Finally, the observational design identifies associations rather than causal effects. Because the dataset contains green bonds only, it provides neither the matched conventional-bond counterfactual required to estimate a greenium nor the comparison universe required to estimate whether firms in particular sectors are more likely to issue green bonds.

7.2. Future Research

Future research should link bond issuance dates to more complete annual or quarterly financial statements, use correlated-random-effects or fixed-effects alternatives where sufficient within-issuer variation exists, and collect richer governance and environmental measures. Integrated historical rating data could test whether the results are robust across Fitch, S&P, and other agencies using ratings matched consistently to each issuance date. Larger samples of formally certified bonds would permit more flexible multilevel multinomial structures and clearer separation of alignment, external review, and certification. Cross-market work could replace the broad issuance-market grouping with direct measures of taxonomy design, disclosure enforcement, investor protection, and institutional ESG demand. Finally, a new universe containing conventional bonds or non-issuing firms could distinguish sectoral issuance propensity from sector-specific greenium, while matched bond-level designs could investigate whether the distinct recognition categories are priced differently relative to comparable conventional debt.