Green Bonds Need More Attention
(2025) DABN01 20251Department of Economics
Department of Statistics
- Abstract
- This thesis applies deep learning methods to examine the forecastability of green bond yields relative to conventional (brown) bond yields. Using Transformer-based neural networks trained on a large panel of macro-financial indicators, selected via LASSO regularization, we develop forecasting models that differentiate between regular and crisis periods. The results suggest that green and brown bond yields exhibit structurally different dynamics: green bonds demonstrate greater forecastability and more consistent directional accuracy during periods of market stress. Complex architectures such as Transformers are not always necessary; simpler models like univariate multilayer perceptrons often achieve comparable performance. These findings... (More)
- This thesis applies deep learning methods to examine the forecastability of green bond yields relative to conventional (brown) bond yields. Using Transformer-based neural networks trained on a large panel of macro-financial indicators, selected via LASSO regularization, we develop forecasting models that differentiate between regular and crisis periods. The results suggest that green and brown bond yields exhibit structurally different dynamics: green bonds demonstrate greater forecastability and more consistent directional accuracy during periods of market stress. Complex architectures such as Transformers are not always necessary; simpler models like univariate multilayer perceptrons often achieve comparable performance. These findings highlight both the potential and limitations of deep learning in sustainable finance forecasting and underscore the value of regime-aware modeling. As the green bond market expands, the evidence supports treating green and brown bonds as distinct instruments, an important consideration for both investors and researchers. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9194899
- author
- Freidestam, Douglas LU and Ekström, Hugo LU
- supervisor
- organization
- alternative title
- A Deep Learning Approach to Forecasting Yield Dynamics across Market Regimes
- course
- DABN01 20251
- year
- 2025
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- green bonds, greenium, yield forecasting, deep learning, Transformer, neural networks, financial crises
- language
- English
- id
- 9194899
- date added to LUP
- 2025-09-12 09:04:13
- date last changed
- 2025-09-12 09:04:13
@misc{9194899, abstract = {{This thesis applies deep learning methods to examine the forecastability of green bond yields relative to conventional (brown) bond yields. Using Transformer-based neural networks trained on a large panel of macro-financial indicators, selected via LASSO regularization, we develop forecasting models that differentiate between regular and crisis periods. The results suggest that green and brown bond yields exhibit structurally different dynamics: green bonds demonstrate greater forecastability and more consistent directional accuracy during periods of market stress. Complex architectures such as Transformers are not always necessary; simpler models like univariate multilayer perceptrons often achieve comparable performance. These findings highlight both the potential and limitations of deep learning in sustainable finance forecasting and underscore the value of regime-aware modeling. As the green bond market expands, the evidence supports treating green and brown bonds as distinct instruments, an important consideration for both investors and researchers.}}, author = {{Freidestam, Douglas and Ekström, Hugo}}, language = {{eng}}, note = {{Student Paper}}, title = {{Green Bonds Need More Attention}}, year = {{2025}}, }