Modeling and forecasting yield curves: A comparative analysis of parametric vs. machine learning techniques

Accurately modeling and forecasting yield curves in financial markets is vital for economic analysis and policymaking. Yield curves, reflecting the term structure of interest rates and market expectations, play a crucial role in predicting economic activities and guiding financial decision-making processes. Yield curve forecasts, therefore, are invaluable resources for investors, central banks, and other economic agents. This doctoral dissertation focuses on the modeling and forecasting of yield curves in the bond markets of Turkey, the United States (US), and the European Central Bank (ECB). The study evaluates three distinct methods: the Dynamic Nelson-Siegel Model (DL), Functional Principal Component Analysis (FPCA), and Dynamic Gaussian Processes (DGP), comparing their performances using a broad dataset and rigorously analyzing their forecasting accuracies. The DL model excelled in short-term predictions but displayed increasing error rates in long-term forecasts. The FPCA model demonstrated stable and low error rates, particularly with ECB data, highlighting its efficacy in complex market structures. The DGP model achieved the best results across all markets, especially excelling with low RMSE values in US datasets. Despite the high volatility in the Turkish market, which generally reduces prediction accuracy, the DGP model outperformed the other methods significantly. Beyond modeling the relationship between yields and maturities, the study also explored the implications of these models on economic and financial policies. The findings contribute to central bank policy formation and investor risk management by offering methodological recommendations and practical guidance on selecting suitable models for different market conditions. This research provides valuable theoretical and practical insights for future economic forecasting studies by evaluating the forecasting performance of both parametric approaches and machine learning techniques.