This website provides government zero-coupon yield curve estimates for the United States, Germany, Japan, the United Kingdom, Canada, and Switzerland. All six markets include Kernel Ridge Regression (KR) yield curves. For the five non-U.S. markets, KR-Transfer (KR-T) estimates additionally incorporate information from U.S. Treasury and foreign-exchange markets. The United States page also provides discount bond returns, term structure factors, and bond market condition measures.
Across all six markets, our method dominates the Nelson-Siegel-Svensson benchmark used by many central banks in extensive out-of-sample comparisons, and for US Treasuries it also dominates all nonparametric benchmarks. It achieves substantially smaller out-of-sample yield and pricing errors while being robust to outliers and data selection choices. We attribute the superior performance to the optimal trade-off between flexibility and smoothness.
Data by country
Market conditions GermanyKR yields · KR-T yields JapanKR yields · KR-T yields United KingdomKR yields · KR-T yields CanadaKR yields · KR-T yields SwitzerlandKR yields · KR-T yields
Methods
Kernel Ridge Regression (KR)
KR estimates a market's discount curve from that market's own government bonds. Rather than imposing a parametric shape, it searches a general class of smooth curves and trades off how well the curve reprices the observed bonds against how much curvature it needs to do so. The penalty applies to curvature only, so the level and slope of the curve are left to the data.
Available for all six markets.
KR-Transfer (KR-T)
Local bond markets have maturity gaps. KR-T fills them using the deepest market available: FX spot and forward rates convert U.S. Treasuries into target-currency synthetic bonds, and the two curves are estimated jointly under a weak covered-interest-parity restriction that penalises only the curvature of the cross-currency spread.
Available for the five non-U.S. markets.
Which one? The two agree closely at maturities where local government bonds are densely observed. KR-T is the better choice where the local cross-section is sparse. See the empirical evidence →
Research
The database is documented in three papers. Cite the one matching the data you use (how to cite).
Stripping the Discount Curve - a Robust Machine Learning Approach
Damir Filipovic, Markus Pelger & Ye Ye · Management Science, 2026
Introduces the kernel ridge regression approach to discount curve estimation and shows that it dominates parametric and nonparametric benchmarks for U.S. Treasuries. This is the methodology behind the KR yield curves.
Shrinking the Term Structure
Damir Filipovic, Markus Pelger & Ye Ye · Review of Finance, 2026
Shows that the same framework unifies curve estimation with cross-sectional factor modelling, and identifies a fourth “complexity” factor carrying a large risk premium. This is the methodology behind the U.S. discount bond returns, term structure factors, and bond market condition measures.
Stripping Discount Curves Across Currencies: Transfer Learning from US Treasuries
Damir Filipovic, Markus Pelger & Rose Wang · Working paper, 2026
Shows that KR generalises to all six government bond markets and uniformly dominates the Nelson-Siegel-Svensson benchmark, then introduces KR-T, a transfer-learning extension that carries information from data-rich to data-poor markets through FX forwards. This is the methodology behind the KR-T yield curves.
Authors
Prof. Dr. Damir Filipovic
Ecole Polytechnique Fédérale de Lausanne (EPFL) and Swiss Finance Institute
