Macro Scope: Factor Selection in Dynamic Term Structure Models via Bayesian Methods
Project Description
The proposed project will develop new techniques to improve the forecasting of bond returns and yield curve dynamics by systematically identifying the most relevant macroeconomic factors that affect interest rates. Traditional dynamic term structure models (DTSMs) often assume that the information in the current yield curve fully captures all drivers of future interest rates. However, recent research reveals that various macroeconomic variables, such as measures of inflation and real activity, can possess additional predictive power beyond what yields alone reflect. At the same time, other studies show that including too many factors can lead to unstable forecasts and poor investment outcomes out-of-sample. This project addresses these conflicting findings by developing a Bayesian learning framework that dynamically selects the most important factors for forecasting in real time, while ignoring redundant predictors.
Using modern Bayesian methods, particularly Sequential Monte Carlo and stochastic variable search algorithms, we will let the data inform which macroeconomic variables truly matter for predicting excess bond returns. Our approach updates investors' beliefs as new data arrive, avoiding the pitfalls of overfitting and accounting for model uncertainty in a principled way. We will evaluate whether this data driven macro factor selection leads to improved predictive accuracy and higher economic value for bond investors. By possibly reducing vast macro universe to a few key predictive factors, this project has the potential to enhance both the understanding of what drives bond risk premia and the practical management of interest rate risk.
Using modern Bayesian methods, particularly Sequential Monte Carlo and stochastic variable search algorithms, we will let the data inform which macroeconomic variables truly matter for predicting excess bond returns. Our approach updates investors' beliefs as new data arrive, avoiding the pitfalls of overfitting and accounting for model uncertainty in a principled way. We will evaluate whether this data driven macro factor selection leads to improved predictive accuracy and higher economic value for bond investors. By possibly reducing vast macro universe to a few key predictive factors, this project has the potential to enhance both the understanding of what drives bond risk premia and the practical management of interest rate risk.
Relevance to Liechtenstein
The project creates value for Liechtenstein by strengthening both the research environment of the University of Liechtenstein's Financial Economics focus area and the analytical capabilities of the country's financial industry. The project is embedded in an active academic setting where research activities, discussions, and regular scholarly exchanges allow results to be shared with colleagues and practitioners based in Liechtenstein. This ensures that insights generated by the project contribute to the broader academic dialogue and support evidence-based decision-making in the local financial ecosystem. For Liechtenstein's internationally oriented financial institutions, the project offers concrete practical benefits. By identifying a small set of macroeconomic factors that are most relevant for predicting yield curve dynamics and bond risk premia, the outcomes support more robust forecasting and improved management of interest rate risk. The statistical methodology developed in the project reduces model uncertainty, enabling banks, asset managers, insurers, and pension providers in Liechtenstein to strengthen their fixed-income analytics and portfolio decisions. The project also contributes to knowledge transfer and capacity building within Liechtenstein. Its findings will be incorporated into courses and research-related activities at the University of Liechtenstein, ensuring that students and future professionals in the financial sector are trained in state-of-the-art empirical methods.
