US Treasury Yield Curve Forecasting: Application of Machine Learning in Financial Decision-Making

This article introduces a GitHub open-source project that uses machine learning methods to predict the US Treasury yield curve. The project explores the application value of ML in financial decision-making and the significance of the yield curve as a leading economic indicator. The original author of the project is Shourya0609, released on June 14, 2026, with the link: https://github.com/Shourya0609/forecasting_the_US_treasury_yield_curve.

The yield curve describes the relationship between Treasury bond yields and maturity periods; normally, it slopes upward. Changes in its shape contain economic information: an upward slope indicates expansion, while a flat/inverted curve may signal a recession (historically, inversion is a reliable warning of recession). It integrates market expectations for interest rates, inflation, and growth, serving as a reference for central bank monetary policy and helping investors optimize asset allocation and identify risks, etc.

Traditional methods (VAR, dynamic Nelson-Siegel model) have limitations such as linear assumptions, curse of dimensionality, and difficulty adapting to dynamic changes. ML methods (e.g., neural networks, random forests) have advantages like nonlinear modeling, automatic feature extraction, multi-source data fusion, and adaptive learning, making them more suitable for capturing complex relationships in financial markets.

Based on the project description, the architecture is speculated to include: 1. Data preprocessing (processing historical yield, macroeconomic, monetary policy, and market data, including missing value handling, standardization, etc.); 2. Feature engineering (technical indicators, term structure features, macro factors, lag features, etc.); 3. Model selection (supervised learning like XGBoost, time series like LSTM, hybrid models); 4. Evaluation (rolling window validation, metrics like RMSE, MAE).

For financial institutions: asset-liability management (interest rate risk management), portfolio optimization (duration adjustment), derivative pricing; For policymakers: evaluating monetary policy transmission, monitoring systemic risks, predicting economic cycles; For academia: verifying ML effectiveness, exploring model architectures, researching feature importance.

Need to address overfitting (regularization, early stopping), market structure changes (regime switching, online learning), interpretability requirements (SHAP/LIME, interpretable models), and data quality issues (missing values, survivor bias).

Including deep learning architecture innovations (Transformer, graph neural networks), multi-task learning (joint prediction of economic recession, etc.), reinforcement learning applications (optimizing trading strategies), and real-time prediction systems (supporting high-frequency trading).

This project demonstrates the application of ML in yield curve forecasting and provides support for financial decision-making. Although ML cannot guarantee absolute accuracy, it offers a systematic method to extract information. For learners and researchers, it is a good starting point to understand the application of ML in finance.