1% ============================================================================2\section{Conclusion}3\label{sec:conclusion}4% ============================================================================56The Nelson--Siegel factors can be observed at the one-minute frequency with data that are continuously available, by estimating7on daily data the portfolios of Treasury futures that track each factor and applying them to the minute returns of the8most-traded contract of each session. Out of sample, the cumulated intraday series reproduces the daily level factor and the9implied 2- to 30-year yields almost completely and the slope and curvature factors to the extent that they describe the102--30-year shape of the curve; the bill segment and the Nelson--Siegel short rate are not identified, and the paper says so11rather than reporting them. An independent construction from Treasury ETFs sees the same level moves and the same direction of announcement12responses.1314With the factors in hand, four facts emerge. Scheduled releases concentrate yield-curve risk: a thirty-minute window around the15Employment report carries a third of the day's realized factor variance and the FOMC statement a fifth, and the mix of factors in16the response has shifted from the short rate during the zero lower bound to the two-year sector and the curvature in the current17cycle, so that CPI days now carry more curve risk than FOMC days. The moves differ in persistence: FOMC moves are not reversed18and extend over the following two weeks, CPI moves are partly reversed on average, and within the hour there is little cross-factor19dynamics, while the factors' daily volatilities are tightly connected with curvature volatility the receiver. Factor volatility20has a strong intraday periodicity with the 08:30 release minute as its mode. And a large part of the one-minute comovement of21level and slope is what tick-size noise would produce, a cautionary result for anyone who would compute factor covariances at the22highest frequency; at five minutes and beyond the comovement is real and matches the daily data. Realized factor variances built23from the futures tape forecast themselves better than GARCH models fitted to daily factor changes under the QLIKE loss, though24not under mean squared error.2526Three extensions follow directly. Short-rate futures (SOFR and, before 2018, Eurodollar and federal funds) would close the bill27gap and could be added to the instrument set over the whole sample with a vendor that carries their history at the minute frequency.28The announcement-window factor changes can be regressed on published monetary-policy surprise series to map the Nelson--Siegel29decomposition onto the target, path and asset-purchase factors of the identification literature. And the realized factor30covariances can replace the daily covariances in dynamic Nelson--Siegel models with time-varying volatility, where the factor31innovations are currently measured once a day.32