Bayesian sequential inference and machine learning applied to yield curve forecasting

We use machine learning, applied mathematics and techniques from modern statistics to refine Dynamic Term Structure Models. Specifically, we propose alternative, sequential Monte Carlo based solutions for the model selection problem, the extraction of unobserved factors from the yield curve and the identification of nonlinear associations between bond yields and the economy. The computational algorithms improve interest rate forecasts, significantly. In particular, they considerably facilitate the process of turning predictive performance into economic benefits to bond investors, verified within a dedicated portfolio optimization framework. Empirical results for the US Treasuries are especially important for companies and institutions navigating global fixed-income markets, including university endowments, pension funds and insurance corporations.