1% ============================================================================2\section{Related literature}3\label{sec:literature}4% ============================================================================56\paragraph{Nelson--Siegel factors.} \citet{nelson1987parsimonious} proposed the three-component exponential curve as a7parsimonious fit to the cross-section of yields; \citet{diebold2006forecasting} fixed the decay parameter, interpreted the three8coefficients as level, slope and curvature, and showed that autoregressive forecasts of the factors forecast yields well.9\citet{diebold2006macroeconomy} linked the factors to macroeconomic variables in a state-space system,10\citet{christensen2011affine} derived the arbitrage-free version, and \citet{diebold2013yield} survey the approach. The11factor interpretation goes back to the principal-component analysis of \citet{litterman1991common}. All of this work is at the12daily or monthly frequency; \citet{gurkaynak2007treasury} describe the daily zero-coupon curve that is the usual input. The only13intraday application of the Nelson--Siegel form we are aware of, \citet{jeleskovic2020comparing}, fits curves to interbank14quotes on the Italian e-MID market and is concerned with the fit rather than with the dynamics of the factors.1516\paragraph{Announcements and the Treasury market.} The intraday response of Treasury prices to macroeconomic releases is well17documented. \citet{fleming1999price} describe price formation and liquidity in the cash market around announcements;18\citet{balduzzi2001economic} measure the response of bond prices at several maturities to the surprise component of19releases; \citet{andersen2003micro} document the response of exchange rates and \citet{andersen2007real} that of bond, equity and20currency futures, showing that the price response is concentrated in the first minutes after the release and differs across the21business cycle; \citet{faust2007high} examine exchange and interest rates jointly. \citet{brandt2004price} study price discovery and order flow in the cash market, \citet{mizrach2008information} information22shares across cash and futures venues, and \citet{jiang2011information} information and liquidity shocks around announcements; \citet{fleming2003measuring}23measures Treasury market liquidity. These papers work with individual maturities or contracts. Our contribution relative to24them is to move the unit of analysis from the maturity to the factor, so that the response to a release can be described as a25move in the level, the slope and the curvature of the curve rather than as a vector of yield changes.2627\paragraph{High-frequency identification of monetary policy.} \citet{kuttner2001monetary} used federal funds futures to28separate expected from unexpected policy changes; \citet{cochrane2002fed} and \citet{piazzesi2005bond} studied the term-structure29response. \citet{gurkaynak2005actions} showed that two factors---a target factor and a path factor---are needed to describe30asset-price moves in FOMC windows, and \citet{swanson2021measuring} added a third, large-scale asset purchase, factor;31\citet{nakamura2018high} used the same windows to identify information effects, \citet{hanson2015monetary} the response of32long-term real rates, and \citet{bauer2023reassessment} reassess the predictability of the surprises. \citet{lucca2015pre}33document the pre-FOMC drift in equities. These papers extract factors from the \emph{announcement-window surprise}; our34factors describe the \emph{curve}. The two are related---a target surprise is largely a move in the short rate, which in35Nelson--Siegel terms is the sum of level and slope, and a path surprise is largely a move in the 2-year yield relative to the36short rate---and we make the relation explicit in Section~\ref{sec:results}. We do not attempt identification.3738\paragraph{Impulse responses and connectedness.} The persistence of announcement moves is usually studied with daily or39monthly local projections; we apply the local-projection idea to minute and multi-day horizons around the release, which is40close in spirit to the drift analysis of \citet{lucca2015pre}. Spillovers across the factors' volatilities are measured with the41generalized forecast-error variance decomposition of \citet{pesaran1998generalized} as organised by \citet{diebold2012better}.4243\paragraph{Realized covariation of yields.} The realized-volatility literature \citep{andersen2003modeling,barndorffnielsen2004econometric}44is mostly concerned with equities and exchange rates. For bonds, \citet{andersen2010bonds} ask whether the yield curve spans45volatility risk and \citet{collindufresne2002bonds} whether bonds span the fixed-income market; \citet{cieslak2016information} study the46information in the term structure of yield volatility; \citet{hansen2023joint} builds a joint model of the term structure and of realized yield covariation computed from47intraday Treasury data, which is the closest antecedent of our realized factor measures.48\citet{litterman1991volatility} relate volatility to the curve. Our realized measures differ in that they are computed for the49Nelson--Siegel factors themselves, from futures, over the full 23-hour session, and are validated against the daily factors. The50microstructure issues we encounter---the Epps effect \citep{epps1979comovements} and tick-size noise---are the standard ones,51and we address them with the signature-plot logic of the realized-volatility literature. The forecasting comparison uses the HAR52model of \citet{corsi2009simple}, the loss functions discussed by \citet{patton2011volatility}, GARCH \citep{bollerslev1986generalized}53as the daily benchmark, and the test of \citet{diebold1995comparing} with the covariance estimator of \citet{newey1987simple}.5455\paragraph{Futures-specific issues.} The mapping from a futures price to a yield runs through the cheapest-to-deliver bond,56the delivery option and the basis \citep{burghardt2005treasury}; the March 2020 cash--futures basis dislocation is a reminder that the mapping is not constant, and the slow-moving-capital57logic of \citet{duffie2010presidential} is the usual reading of such episodes. The mimicking-portfolio approach absorbs these features into the estimated58weights and re-estimates them quarterly, and the cash-yield ceiling of Section~\ref{sec:results} measures what they cost.59