Stripping the Discount Curve - a Robust Machine Learning Approach

The SAA Working Group on Yield curves was created in June 2022 with the goal to provide technical input to the FINMA-Arbeitsgruppe Zinskurven. Specifically, we apply the Kernel Ridge (KR) method developed by Filipović, Pelger, and Ye (paper available at SSRN: https://ssrn.com/abstract=4058150) to estimate the yield curve of the Swiss Government Bond market. 

In this talk, I present and compare the KR method to the benchmark methods of Nelson-Siegel-Svensson, and its implementation of the Swiss National Bank, and the Solvency II standard Smit-Wilson and its implementation for the Swiss Solvency Test. An extensive empirical analysis run on a sample of Swiss Government Bonds from 2010 to 2022 reveals the superior performance of the KR method, both in terms of accuracy and robustness.