Data-Driven Duration Management -- Term Structure Forecasting Using Machine Learning — explained | StockTools Quant Research

quant research // summarizing the paperLOADING

Data-Driven Duration Management -- Term Structure Forecasting Using Machine Learning

Tobias Lausser, Joao Eduardo Vuolo, Rudi Zagst · 2026-06-25

what this means for investors // AI summary of the abstractSituationally useful

The paper tests different ways to forecast the yield curve (term structure) of U.S. and European government zero-coupon bonds, comparing classical models like Dynamic Nelson-Siegel and PCA against neural networks, sometimes adding macroeconomic data. It evaluates them both on statistical accuracy (RMSE, MAE, directional accuracy) and on how a bond trading strategy built on the forecasts performs. Neural networks consistently beat traditional models on both accuracy and portfolio results, but the best-performing setup differed by market.

Why it matters: Fixed-income managers doing duration management or bond allocation might find that ML-based yield curve forecasts could improve on classical Nelson-Siegel or PCA approaches. The finding that the optimal model differs between the U.S. and Europe suggests forecasting setups may need to be tailored per market rather than applied uniformly.

Neural networks outperformed classical DNS and PCA models in both forecast accuracy and simulated portfolio performance across both markets.
The best U.S. model used DNS factors plus an autoencoder for macro features; the best European model used PCA factors and no macro data — market-specific tuning mattered.
Hybrid designs that embed classical model structure (DNS/PCA factors) into neural networks were most effective, rather than pure black-box or pure classical approaches.
Adding macroeconomic variables helped the U.S. but not Europe, so more inputs aren't always better.

Applies to: yield curve / term structure forecasting and fixed-income duration management

⚠ Results come from historical backtests and simulated trading strategies, so the reported outperformance may not persist under live trading, transaction costs, or different regimes.

ABSTRACT

This paper compares different methods for forecasting the term structure of U.S. and European zero-coupon government bonds using both traditional econometric and Machine Learning (ML) approaches. We compare classical models (e.g., Dynamic Nelson-Siegel (DNS) and Principal Component Analysis (PCA)) with different Neural Network (NN) architectures, including those inspired by the classical models, on the U.S. Treasury market and bonds issued by the European Central Bank (ECB). To enhance predictive performance, macroeconomic variables are incorporated. The findings for both markets are separately analyzed and compared. To this end, we propose a robust model evaluation framework combining statistical accuracy metrics - such as RMSE, MAE, and directional accuracy - with the economic relevance of a quantitative bond trading strategy. Results show that NNs consistently outperform traditional models in both forecasting accuracy and portfolio performance. For the U.S., the most effective approach is a direct-forecasting NN that incorporates DNS factors to reduce the dimensionality of zero-rate data and an Autoencoder (AE) to extract macroeconomic features, while for Europe, the optimal model is a factor-based NN using PCA-derived zero-rate factors without the integration of macroeconomic variables. Overall, the paper demonstrates how combining traditional modeling approaches with modern ML techniques and evaluation can improve yield curve forecasts and support applications in fixed-income portfolio construction.

★ PRO · COMING SOON

Go deeper: a full research-committee breakdown of this paper, its assumptions and failure modes, and how its method would apply to a specific ticker or your watchlist. See StockTools AI →

AI summary generated from the paper’s public abstract via arXiv; it may miss nuance — read the source before relying on it. Thank you to arXiv for its open-access interoperability; StockTools is not affiliated with arXiv, and all rights remain with the authors. Educational only, not financial advice.