1\begin{table}[htbp]\centering\small2\caption{Final (held-out): QLIKE relative to Random walk (trailing RMS)}3\label{tab:final_cmp_qlike_heldout}4\begin{adjustbox}{max width=\textwidth}5\begin{tabular}{lcccc}6\toprule7Model & $H=5$ & $H=15$ & $H=30$ & $H=60$ \\8\midrule9Seasonal volatility & -7881292606331.87\% (-8.1) & -8896570178640.64\% (-17.9) & -7805191930785.72\% (-9.8) & -6357895926512.99\% (-6.1) \\10EWMA-$t$ & -14368387124373.51\% (-32.3) & -12886749273533.74\% (-35.1) & -10692226543788.80\% (-19.3) & -7804644926948.44\% (-10.3) \\11HAR-RV & -11991153797857.57\% (-29.6) & -15408356847389.03\% (-47.2) & -14085402915508.40\% (-25.9) & -12829576065389.08\% (-15.8) \\12GARCH(1,1)-$t$ & -9150495905868.49\% (-18.9) & -8241640138160.87\% (-22.8) & -6060670156897.87\% (-9.2) & -2677412926831.07\% (-3.1) \\13Ridge & -1054018044397.35\% (-1.1) & -14400258822980.35\% (-34.5) & -14083112880296.95\% (-22.5) & -13172287575370.57\% (-15.2) \\14Elastic net & -1045208980438.32\% (-1.1) & -14378492224890.53\% (-34.1) & -14070621964178.67\% (-22.3) & -13168125458512.58\% (-15.1) \\15LightGBM & +1438442754994027.25\% (3.0) & -11074436016941.67\% (-4.1) & -14510585977166.56\% (-21.1) & -13818631535817.14\% (-15.8) \\16MLP (1m) & -17448222120920.47\% (-34.2) & -16543617998332.09\% (-47.5) & -15403641955970.65\% (-31.0) & -14313813763982.13\% (-21.4) \\17TCN (1m) & -18443585062096.97\% (-35.2) & -17109454564114.94\% (-47.4) & -15760046569402.44\% (-32.0) & -14452672740764.15\% (-22.8) \\18MRPT-S (1m+5m+30m) & -18624546657634.33\% (-29.4) & -17586756948614.74\% (-43.5) & -16319208002183.26\% (-29.8) & -15332871126334.91\% (-22.1) \\19MRPT-S (1m only) & -18362352193653.54\% (-30.5) & -17231017095898.39\% (-43.5) & -15813797009598.51\% (-27.2) & -14506694663658.71\% (-18.9) \\20MRPT-B (1m+5m+30m) & -18449196322165.06\% (-38.4) & -17319294892507.10\% (-50.0) & -16088668851044.35\% (-35.1) & -14988730328170.58\% (-25.6) \\21\bottomrule22\end{tabular}23\end{adjustbox}24\begin{minipage}{0.95\textwidth}\scriptsize Relative change in mean QLIKE versus the reference (negative = better); Diebold--Mariano $t$-statistic with HAC variance in parentheses.\end{minipage}25\end{table}