\begin{table}[htbp]
\centering\scriptsize
\caption{Real-data ablations at the municipality level: RMSE of monthly changes against the full model (×100) and one-step-ahead standardized innovation diagnostics}\label{tab:ablation_real}
\begin{threeparttable}
\resizebox{\textwidth}{!}{%
\begin{tabular}{lrrrrrrrrr}
\toprule
Variant & H: $\Delta$ vs full & H: OOS $z$ & H: \\% in band & C: $\Delta$ vs full & C: OOS $z$ & C: \\% in band & P: $\Delta$ vs full & P: OOS $z$ & P: \\% in band \\
\midrule
Full model & 0.00 & 1.38 & 86 & 0.00 & 1.22 & 90 & 0.00 & 1.33 & 87 \\
No EB pooling of $\tau^2$ & 0.71 & 1.42 & 85 & 0.35 & 1.22 & 90 & 0.96 & 1.33 & 87 \\
Fixed quality prices & 0.48 & 1.38 & 86 & 0.44 & 1.21 & 90 & 0.42 & 1.32 & 87 \\
Local linear trend & 1.04 & 1.42 & 85 & 0.88 & 1.19 & 90 & 1.24 & 1.32 & 87 \\
Gaussian innovations & 3.16 & 1.27 & 88 & 1.74 & 1.08 & 92 & 1.92 & 1.24 & 88 \\
Municipality-level location FE & 3.49 & 1.35 & 87 & 1.55 & 1.11 & 91 & 1.64 & 1.27 & 88 \\
No MRC level & 3.66 & 1.34 & 87 & 1.47 & 1.21 & 90 & 1.71 & 1.31 & 87 \\
\bottomrule
\end{tabular}}
\begin{tablenotes}\footnotesize\item OOS $z$ = root mean square of the standardized one-step-ahead prediction error of the cell's monthly mean deviation (1 if calibrated); \% in band = share of innovations inside $\pm 1.96$.\end{tablenotes}
\end{threeparttable}
\end{table}