1\begin{table}[htbp]\centering\small2\caption{Ablations: mean scaled CRPS of the cumulative-return forecast by horizon (minutes)}3\label{tab:ablations_crps_scaled}4\begin{adjustbox}{max width=\textwidth}5\begin{tabular}{lcccc}6\toprule7Model & $H=5$ & $H=15$ & $H=30$ & $H=60$ \\8\midrule9Random walk (trailing RMS) & 0.4699 {\scriptsize[0.4648, 0.4754]} & 0.4575 {\scriptsize[0.4515, 0.4639]} & 0.4469 {\scriptsize[0.4412, 0.4538]} & 0.4347 {\scriptsize[0.4271, 0.4441]} \\10Seasonal volatility & 0.4582 {\scriptsize[0.4527, 0.4641]} & 0.4453 {\scriptsize[0.4390, 0.4519]} & 0.4347 {\scriptsize[0.4285, 0.4418]} & 0.4221 {\scriptsize[0.4139, 0.4320]} \\11EWMA-$t$ & 0.4547 {\scriptsize[0.4492, 0.4606]} & 0.4425 {\scriptsize[0.4363, 0.4491]} & 0.4326 {\scriptsize[0.4264, 0.4396]} & 0.4214 {\scriptsize[0.4133, 0.4314]} \\12MRPT-S (1m+5m) & 0.4537 {\scriptsize[0.4482, 0.4596]} & 0.4418 {\scriptsize[0.4355, 0.4483]} & 0.4316 {\scriptsize[0.4253, 0.4384]} & 0.4191 {\scriptsize[0.4110, 0.4289]} \\13MRPT-S cross-attention fusion & 0.4535 {\scriptsize[0.4480, 0.4594]} & 0.4416 {\scriptsize[0.4353, 0.4481]} & 0.4313 {\scriptsize[0.4251, 0.4382]} & 0.4190 {\scriptsize[0.4109, 0.4288]} \\14MRPT-S Gaussian head & 0.4546 {\scriptsize[0.4492, 0.4606]} & 0.4424 {\scriptsize[0.4361, 0.4489]} & 0.4320 {\scriptsize[0.4258, 0.4389]} & 0.4195 {\scriptsize[0.4115, 0.4292]} \\15MRPT-S no regime & 0.4534 {\scriptsize[0.4479, 0.4593]} & 0.4414 {\scriptsize[0.4352, 0.4480]} & 0.4312 {\scriptsize[0.4249, 0.4380]} & 0.4188 {\scriptsize[0.4107, 0.4286]} \\16MRPT-S learned regime & 0.4531 {\scriptsize[0.4476, 0.4591]} & 0.4410 {\scriptsize[0.4347, 0.4476]} & 0.4306 {\scriptsize[0.4243, 0.4376]} & 0.4182 {\scriptsize[0.4100, 0.4281]} \\17MRPT-S returns only & 0.4542 {\scriptsize[0.4487, 0.4601]} & 0.4424 {\scriptsize[0.4362, 0.4489]} & 0.4322 {\scriptsize[0.4260, 0.4392]} & 0.4201 {\scriptsize[0.4119, 0.4299]} \\18MRPT-S time only & 0.4542 {\scriptsize[0.4488, 0.4601]} & 0.4424 {\scriptsize[0.4362, 0.4489]} & 0.4323 {\scriptsize[0.4260, 0.4392]} & 0.4200 {\scriptsize[0.4119, 0.4299]} \\19MRPT-S OHLCV (no rel. volume) & 0.4536 {\scriptsize[0.4481, 0.4595]} & 0.4418 {\scriptsize[0.4355, 0.4482]} & 0.4316 {\scriptsize[0.4253, 0.4385]} & 0.4193 {\scriptsize[0.4113, 0.4291]} \\20MRPT-S shuffled minutes & 0.4538 {\scriptsize[0.4483, 0.4598]} & 0.4420 {\scriptsize[0.4357, 0.4484]} & 0.4318 {\scriptsize[0.4255, 0.4388]} & 0.4196 {\scriptsize[0.4115, 0.4294]} \\21MRPT-S + ticker id & 0.4534 {\scriptsize[0.4479, 0.4593]} & 0.4418 {\scriptsize[0.4356, 0.4483]} & 0.4318 {\scriptsize[0.4256, 0.4387]} & 0.4199 {\scriptsize[0.4118, 0.4297]} \\22MRPT-S + copula (joint sampling) & 0.4534 {\scriptsize[0.4479, 0.4593]} & 0.4417 {\scriptsize[0.4355, 0.4481]} & 0.4316 {\scriptsize[0.4254, 0.4385]} & 0.4194 {\scriptsize[0.4113, 0.4291]} \\23MRPT-S reference (4k steps) & 0.4533 {\scriptsize[0.4478, 0.4592]} & 0.4415 {\scriptsize[0.4352, 0.4480]} & 0.4312 {\scriptsize[0.4249, 0.4381]} & 0.4188 {\scriptsize[0.4107, 0.4286]} \\24MRPT-S 1m only (4k) & 0.4538 {\scriptsize[0.4483, 0.4597]} & 0.4420 {\scriptsize[0.4358, 0.4485]} & 0.4318 {\scriptsize[0.4256, 0.4386]} & 0.4195 {\scriptsize[0.4113, 0.4293]} \\25MRPT-S reference, seed 2 & 0.4531 {\scriptsize[0.4476, 0.4591]} & 0.4411 {\scriptsize[0.4347, 0.4476]} & 0.4308 {\scriptsize[0.4245, 0.4379]} & 0.4187 {\scriptsize[0.4103, 0.4287]} \\26MRPT-S reference, seed 3 & 0.4531 {\scriptsize[0.4476, 0.4590]} & 0.4413 {\scriptsize[0.4350, 0.4479]} & 0.4312 {\scriptsize[0.4250, 0.4380]} & 0.4194 {\scriptsize[0.4115, 0.4290]} \\27MRPT-S + copula, independent sampling & 0.4535 {\scriptsize[0.4480, 0.4593]} & 0.4416 {\scriptsize[0.4354, 0.4481]} & 0.4314 {\scriptsize[0.4252, 0.4383]} & 0.4191 {\scriptsize[0.4110, 0.4288]} \\28\bottomrule29\end{tabular}30\end{adjustbox}31\begin{minipage}{0.95\textwidth}\scriptsize Cluster-bootstrap (trading-day) 95\% intervals in brackets.\end{minipage}32\end{table}