Comparative analysis of traditional econometric models and artificial intelligence models for forecasting the bond yield curve
DOI:
https://doi.org/10.34121/1028-9763-2025-2-45-52Keywords:
yield curve, forecasting, neural networks, econometric models, machine learningAbstract
The article is devoted to a comparative analysis of the effectiveness of traditional econometric models and artificial intelligence models for forecasting the bond yield curve by assessing the forecasting accuracy and resilience to changes in market conditions. The study uses quarterly data from the Federal Reserve Economic Database (FRED) for the period 2011–2024 for ten maturities of US government bonds. The study is relevant in the context of growing volatility in financial markets and the need for accurate forecasts for investment decisions and monetary policy making. Three traditional models (DNS, Svensson, VAR) and three artificial intelligence models (DNN, CNN-LSTM, XGBoost) were compared. Traditional models were parameterized using the least squares method, and artificial intelligence models were parameterized using a sliding window approach and cross-validation. To ensure the statistical significance of the results, artificial intelligence models were trained and tested in 20 independent experiments. The models were evaluated by RMSE, MAE, and the coefficient of determination R². Artificial intelligence models demonstrated significantly higher forecasting accuracy: DNN showed the best results with RMSE of 0.960 and MAE of 0.832, which is three times better than traditional models. At the same time, traditional models showed greater resilience to changes in market conditions with resilience coefficients close to 1, while AI models showed higher sensitivity to volatility. Statistical analysis confirmed the significance of differences in forecasting accuracy between the models (p<0.05). The lowest accuracy among all the models was demonstrated by the VAR model, which is explained by its limited ability to capture nonlinear dependencies. The proposed models are characterized by a different ratio of accuracy and stability, which allows for choosing the optimal approach depending on specific practical tasks.
References
1. Бєляєв Д. Математичне моделювання як складова розробки фінансового механізму формування золотовалютних резервів. SumDU Repository: Home. URL: https://essuir.sumdu.edu.ua/bitstreamdownload/123456789/52520/5/Bieliaiev_Matematychne_modeliuvannia.pdf (дата звернення: 18.02.2025).
2. Paientko T., Rudaia M. Methodological approaches to the discounting rate choice in forensic economics research. Criminalistics and forensics. 2021. N 66. URL: https://doi.org/10.33994/kndise.2020.66.80 (date of access: 18.02.2025).
3. Vector autoregression, structural equation modeling, and their synthesis in neuroimaging data analysis / G. Chen et al. Computers in biology and medicine. 2011. Vol. 41, N 12. P. 1142–1155. DOI: https://doi.org/10.1016/j.compbiomed.2011.09.004 (date of access: 18.02.2025).
4. Ткачик Д.А., Квєтний Р.Н. Використання економетричних моделей для прогнозування фінансових показників. Репозитарій Вінницького національного технічного університету. URL: https://ir.lib.vntu.edu.ua/bitstream/handle/123456789/38480/17505.pdf (дата звернення: 18.02.2025).
5. Bayraci S., Susuz O. A Deep Neural Network (DNN) based classification modelin application to loan default prediction by Selçuk BAYRACI, Orkun SUSUZ. Theoretical and Applied Economics. URL: https://www.ectap.ro/a-deep-neural-network-dnn-based-classification-modelin-application-to-loandefault-prediction-seluk-bayraci_orkun-susuz/a1421/ (date of access: 18.02.2025).
6. Deng L. et al. Analysis of the Effectiveness of CNN-LSTM Models Incorporating Bert and Attention
Mechanisms in Sentiment Analysis of Data Reviews. Proc. of the 2023 4th International Conference on Big
Data and Informatization Education (ICBDIE 2023). Dordrecht, 2023. P. 821–829. DOI: https://doi.org/10.2991/978-94-6463-238-5_106 (date of access: 18.02.2025).
7. Nielsen D. Tree boosting with xgboost — why does xgboost win «every» machine learning competition? NTNU Open. URL: https://ntnuopen.ntnu.no/ntnu-xmlui/handle/11250/2433761 (date of access: 18.02.2025).
110
22
