Full Text Available

Access Repository Access Repository Access Repository
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Cover Image

Does Green Pay Less? Global Corporate Bond Evidence on Primary and Secondary Yields

The greenium, or green premium, refers to the lower yield that arises from a bond’s green label, conditional on otherwise identical contractual features and credit risk. In our study, we estimate the greenium by combining causal matching techniques with a neural network–based propensity score approa...

Full description

Saved in:
Main Author: El Kenawy, Youssef
Format: Thesis
Published: AUC Knowledge Fountain 2026
Subjects:
Keyword / Variable Definition is_green Dummy variable equal to 1 if the bond is classified as a green bond and 0 otherwise. Green bond indicator Binary measure identifying whether a bond qualifies as a green instrument. ytm_issue Yield to maturity at issuance
measured at the primary market offering date. Yield at issue Initial yield investors receive when the bond is first issued. ask_yield Secondary-market ask yield based on quoted ask-side yield-to-convention. Secondary-market ask yield Yield implied by the ask price in the secondary market. bid_ask Difference between bid and ask prices or yields
used as a liquidity proxy. Bid–ask spread Measure of market liquidity reflecting transaction costs and trading efficiency. Liquidity proxy Variable used to approximate the ease of trading a bond in the market. Winsorization Statistical treatment that limits extreme observations to reduce outlier influence. 1% tail winsorization Replacement of observations below the 1st percentile and above the 99th percentile with boundary values. currency Bond denomination currency represented as a categorical factor variable. Currency fixed effects Controls accounting for systematic differences across bond currencies. Exact matching covariate Variable used for strict one-to-one matching between treated and control observations. year_bin Grouped issue-year category used in matching and regression models. Issue-year category Classification of bonds according to issuance period (e.g.
≤2020
2021–2022
2023+). has_rating Indicator showing whether a valid credit rating exists for the bond. size_q Quintile ranking of issuance amount based on bond size. Issue size quintile Classification of bonds into five groups according to issuance volume. mat3 Maturity bucket grouping bonds by years to maturity. Maturity category Classification of bonds into short-
medium-
or long-term maturity groups. seniority_simple Simplified seniority classification distinguishing senior unsecured debt from other structures. Seniority classification Categorization of bonds according to repayment priority in case of default. issue_year Calendar year in which the bond was issued. issuer_id Unique identifier assigned to each bond issuer. Amt_Issued Total nominal amount issued for a bond offering. ttm_years Time remaining to maturity expressed in years. Time to maturity Number of years between issuance date and maturity date. rating_raw Original credit rating obtained from rating agencies before categorization. rating_bucket Grouped credit rating category such as High
Medium
Risky
or Unrated. rating_cat Ordered factor representation of rating categories. Moody’s rating Credit assessment assigned by Moody's Corporation. S&P rating Credit assessment assigned by S&P Global. Fitch rating Credit assessment assigned by Fitch Ratings. Unrated bonds Bonds without a valid external credit rating. High-risk bonds Bonds classified as having elevated default or credit risk. Senior unsecured debt Debt obligations with high repayment priority but no collateral backing. ps_nn Neural-network estimated propensity score representing the probability of being a green bond. Neural-network propensity score Propensity score estimated using a feed-forward neural network model. NN-PSM Neural Network Propensity Score Matching methodology combining machine learning with causal matching. Propensity score matching (PSM) Statistical matching method used to reduce selection bias between treated and control groups. Feed-forward neural network Machine learning architecture where information flows sequentially from input to output layers. weights Observation weights generated through the matching procedure. Matching weights Numerical adjustments applied to observations during treatment effect estimation. Nearest-neighbor matching Matching approach pairing each treated observation with the closest control observation. subclass Identifier for matched strata generated by the matching algorithm. Matching subclass Group of matched treated and control observations sharing similar characteristics. ATT estimation Estimation of the Average Treatment Effect on the Treated. Average Treatment Effect on the Treated (ATT) Average causal impact of treatment on treated observations only. yield_ng Average yield among conventional (non-green) bonds within the same matching subclass. Conventional bond benchmark Yield benchmark constructed from matched non-green bonds. yield_ng_c Centered version of the conventional bond benchmark yield. Centered benchmark yield Benchmark yield adjusted by subtracting its sample mean. yield_ng_c2 Squared centered benchmark yield used in nonlinear specifications. Quadratic heterogeneity model Regression specification including squared interaction terms to capture nonlinear effects. Linear interaction model Regression model incorporating interaction terms between explanatory variables. Heterogeneity analysis Examination of how treatment effects vary across groups or market conditions. Matching covariates Variables used to construct comparable treated and control groups. Fixed effects Regression controls accounting for unobserved group-specific characteristics. Issuer fixed effects Controls capturing issuer-specific unobservable characteristics. Clustering Statistical adjustment of standard errors for correlated observations. Greenium Yield difference between green bonds and comparable conventional bonds. Green bond pricing Analysis of yield formation and valuation of green debt instruments. Bond yield spreads Differences in yields between bonds with varying characteristics or risks. Secondary market analysis Examination of bond trading behavior after issuance. Primary market analysis Examination of bond characteristics and pricing at issuance. Causal inference Statistical framework for identifying causal relationships between variables. Treatment variable Variable representing exposure to an intervention or condition. Outcome variable Variable measuring the effect or response of interest. Machine learning in finance Application of artificial intelligence methods to financial analysis and prediction. Sustainable finance Financial activities supporting environmental or social sustainability objectives. ESG bonds Bonds linked to environmental
social
and governance objectives. Climate finance Financing directed toward climate mitigation and adaptation initiatives. Fixed-income econometrics Econometric analysis focused on bonds and debt securities.
Business Analytics
Business Intelligence
Corporate Finance
Finance and Financial Management
Tags: Add Tag
No Tags, Be the first to tag this record!
Summary:The greenium, or green premium, refers to the lower yield that arises from a bond’s green label, conditional on otherwise identical contractual features and credit risk. In our study, we estimate the greenium by combining causal matching techniques with a neural network–based propensity score approach to construct a closely comparable set of green and conventional bonds. Our empirical framework incorporates issuer fixed effects and currency × issuance-year fixed effects, ensuring that our estimates reflect the impact of the green label itself rather than differences in macro-financial conditions or issuer composition. Our findings indicate that, once currency-specific issuance-year conditions are absorbed, the primary market greenium becomes economically small and statistically insignificant, suggesting limited evidence of a systematic issuance-stage pricing advantage associated with the green label. In contrast, in the secondary market, our results reveal a statistically significant negative greenium of approximately 7–8 basis points, pointing to persistent valuation effects in trading markets. The effect is strongest under the neural network matching specification, while traditional matching and regression approaches yield qualitatively similar but less precise estimates. Overall, our study contributes to the green bond literature by distinguishing between issuance-stage pricing dynamics and secondary-market valuation effects, and by introducing a machine learning enhanced, causally oriented framework to assess the magnitude and robustness of green bond yield differentials.

Similar Items