spb/wp47_uqo

1\begin{table}[tp]2\centering3\footnotesize4\begin{threeparttable}5\caption{Out-of-sample fidelity of the futures mimicking portfolios for the daily Nelson--Siegel factor changes, 2011--2026 and sub-periods (H1)}6\label{tab:fidelity}7\setlength{\tabcolsep}{3pt}8\begin{tabular}{lrrrrr@{\hspace{8pt}}rrrr@{\hspace{8pt}}rrrr}9\toprule10Period & $n$ & \multicolumn{4}{c}{Level} & \multicolumn{4}{c}{Slope} & \multicolumn{4}{c}{Curvature} \\ \cmidrule(lr){3-6}\cmidrule(lr){7-10}\cmidrule(lr){11-14} & & $R^2$ & $b$ & RMSE & SD & $R^2$ & $b$ & RMSE & SD & $R^2$ & $b$ & RMSE & SD \\11\midrule122011--2026 & 3,963 & 0.95 & 1.00 & 1.3 & 5.7 & \textbf{0.82} & 0.99 & 2.7 & 6.3 & 0.75 & 1.03 & 7.7 & 15.5 \\132011--2015 & 1,240 & 0.95 & 0.98 & 1.4 & 6.2 & 0.92 & 0.98 & 1.9 & 6.6 & 0.79 & 1.00 & 4.8 & 10.5 \\142016--2019 & 1,010 & 0.94 & 0.95 & 1.0 & 4.2 & \textbf{0.79} & 0.95 & 2.3 & 4.9 & 0.66 & 1.07 & 6.0 & 10.2 \\152020--2021 & 502 & 0.94 & 1.00 & 1.5 & 6.2 & \textbf{0.83} & 0.96 & 2.7 & 6.4 & \textbf{0.57} & 0.81 & 6.9 & 10.6 \\162022--2026 & 1,211 & 0.95 & 1.04 & 1.4 & 6.0 & \textbf{0.73} & 1.04 & 3.7 & 7.1 & 0.78 & 1.05 & 11.0 & 23.3 \\17\bottomrule18\end{tabular}19\begin{tablenotes}[flushleft]\footnotesize20\item \textit{Notes:} Each row evaluates, over $n$ out-of-sample yield days, the quarterly re-estimated tracking weights (OLS of the daily CMT-based factor change on the yield-day returns of the available Treasury futures over the trailing three years, applied to the following quarter only) on the daily change of the indicated factor. $R^2$ is the out-of-sample $1-\sum(\Delta\beta-\widehat{\Delta\beta})^2/\sum(\Delta\beta-\overline{\Delta\beta})^2$ of the realised daily change on the prediction formed with the weights only (no intercept); $b$ is the slope of actual on predicted; RMSE and SD (standard deviation of the actual change) are in basis points. A root enters the regression when it has at least 100 one-minute returns on at least 85\% of the training days (root sets: ZT ZF ZN US in 2011--2012, plus UB from 2012Q4, plus TN from 2018Q4). Pre-registered thresholds: $R^2\geq0.90$ for level and slope, $\geq0.60$ for curvature; bold marks $R^2$ below the threshold. Entries are point estimates without confidence intervals.21\end{tablenotes}22\end{threeparttable}23\end{table}24