Data Documentation & Code › United States
U.S. KR yields
Kernel Ridge Regression zero-coupon yield curves for U.S. Treasuries, the reference dataset behind the U.S. discount bond database.
File format and coverage
All yield files begin with date, min_observed_ttm,
max_observed_ttm and max_curve_ttm - _days in
daily-maturity files, _months in monthly-maturity files - followed by one column per
maturity.
| Field | Meaning |
|---|---|
date | Observation date (YYYY-MM-DD) |
min_observed_ttm_days / _months |
Shortest maturity among the Treasuries observed that day |
max_observed_ttm_days / _months |
Longest maturity among those Treasuries; the boundary of the extrapolation region |
max_curve_ttm_days / _months |
Last maturity with a published value; it follows the U.S. publication horizon, which
steps 3,650 → 7,300 → 10,950 days across vintages, and so need not coincide with
max_observed_ttm in either direction |
Yields are annualized, continuously compounded, on a 365-day year, in
decimal units, and cells past max_curve_ttm are blank. Every variant is published as
CSV; the daily-observations, daily-maturities variant is also published as Parquet, with the same
columns.
Yield file layout changed in October 2026.Details →
The yield values are unchanged. The files now use the four columns above, the
same layout as the international panels. The old MAX_DATA_TTM column has been
removed; max_observed_ttm replaces it.
Data availability
Daily from June 14, 1961, on the U.S. Treasury trading calendar; the international date filter is not applied and no bond coverage detail file is published for the United States. 10-, 20-, and 30-year estimates begin on June 14, 1961, January 5, 1973, and November 25, 1985, respectively. Treasury data come from the CRSP Treasuries Time Series.
Construction
KR estimates the discount curve from observed U.S. Treasury prices, trading off repricing fit against curvature, and the published values are the previously released U.S. estimates carried over unchanged. This is a different pipeline from the international panels, with its own ridge scaling: the two share a file format and a yield convention, not an estimation run.
Stripping the Discount Curve - a Robust Machine Learning Approach · Paper & citation →