Words Beyond the Rate Decision? FOMC Statement Innovations and the U.S. Treasury Yield Curve, 2000-2026

This paper asks whether innovations in Federal Open Market Committee (FOMC) statement language explain high-frequency movements in the U.S. Treasury yield curve beyond conventional monetary-policy surprises. The analysis covers 211 scheduled statement releases from March 2000 through July 2026 and uses 30-minute asset-price changes from the Federal Reserve Bank of San Francisco U.S. Monetary Policy Event-Study Database. A transparent text-as-data pipeline classifies sentences as hawkish, dovish, or neutral using word- and character-level TF-IDF features with multinomial logistic regression. The meeting-level innovation score measures the net stance of language added to, removed from, or replaced relative to the preceding scheduled policy statement. Baseline regressions control for the futures-implied current-meeting surprise (MP1) and simultaneous Summary of Economic Projections releases; an expanded specification adds the next-meeting path surprise (MP2). Inference combines HC3 covariance estimates, a meeting-clustered maturity system, permutation and wild-bootstrap tests, multiplicity corrections, equivalence tests, minimum detectable effects, leave-one-year-out prediction, and resampling of the labeled text corpus. The estimated coefficients are small at all six Treasury maturities: 0.001, 0.121, 0.132, 0.119, -0.013, and 0.072 basis points per one-standard-deviation hawkish innovation at three months, six months, two years, five years, ten years, and thirty years. Every 95 percent confidence interval includes zero, and a joint six-maturity test yields p = 0.439. Adding MP2 does not materially change the estimates; the text score does not improve out-of-year prediction, and no result in the predeclared robustness family survives multiplicity adjustment. Equivalence tests reject average effects outside plus or minus 1 basis point at every maturity. Under this design, scalar statement hawkishness provides little incremental average information for Treasury-yield reactions once market-based policy surprises are controlled for, although smaller, state-dependent, or multidimensional communication effects remain plausible.