\begin{table}[htbp]
\centering\scriptsize
\caption{Task B — Diebold--Mariano statistics against the no-change forecast (negative = better than no change) and Holm-adjusted $p$-values across the 18 models of each horizon}\label{tab:tests_b}
\resizebox{\textwidth}{!}{%
\begin{tabular}{lrrrrrrrr}
\toprule
Model & h=1: DM & h=1: Holm $p$ & h=3: DM & h=3: Holm $p$ & h=6: DM & h=6: Holm $p$ & h=12: DM & h=12: Holm $p$ \\
\midrule
Mean growth & +0.26 & 1.00 & -0.59 & 1.00 & -0.73 & 1.00 & -0.26 & 1.00 \\
Momentum & -1.00 & 1.00 & -0.96 & 1.00 & +0.57 & 1.00 & +0.13 & 1.00 \\
Random walk with drift & +2.40 & 0.33 & +1.14 & 1.00 & +0.90 & 1.00 & +0.29 & 1.00 \\
Seasonal naive & +5.00 & 0.00 & +1.12 & 1.00 & +1.00 & 1.00 & +0.29 & 1.00 \\
AR(2), cell & -1.10 & 1.00 & -1.86 & 0.87 & -1.67 & 1.00 & -0.38 & 1.00 \\
Local linear trend & +7.19 & 0.00 & +8.08 & 0.00 & +10.47 & 0.00 & +4.88 & 0.01 \\
Ridge (panel) & -1.90 & 0.80 & -2.02 & 0.75 & -0.92 & 1.00 & +0.13 & 1.00 \\
LightGBM (panel) & -3.85 & 0.01 & -2.91 & 0.11 & -1.79 & 1.00 & -0.19 & 1.00 \\
MLP (panel) & +0.73 & 1.00 & -1.23 & 1.00 & -0.52 & 1.00 & +0.55 & 1.00 \\
LSTM & +1.88 & 0.80 & -1.02 & 1.00 & -1.17 & 1.00 & +0.07 & 1.00 \\
GRU & +1.58 & 0.99 & -1.67 & 1.00 & -1.23 & 1.00 & +0.14 & 1.00 \\
TCN & +0.68 & 1.00 & -1.29 & 1.00 & -1.23 & 1.00 & +0.15 & 1.00 \\
TFT-lite & +1.24 & 1.00 & +0.23 & 1.00 & -0.16 & 1.00 & -0.04 & 1.00 \\
BHPF & -1.79 & 0.82 & -1.92 & 0.83 & -0.74 & 1.00 & +0.11 & 1.00 \\
RAREF & -1.99 & 0.72 & -1.74 & 1.00 & -0.93 & 1.00 & +0.11 & 1.00 \\
QANSS & -2.27 & 0.42 & -2.18 & 0.59 & -1.27 & 1.00 & +0.27 & 1.00 \\
DSGFN & -0.52 & 1.00 & -1.24 & 1.00 & -0.91 & 1.00 & +0.10 & 1.00 \\
QREFA & -1.74 & 0.82 & -2.04 & 0.75 & -1.27 & 1.00 & +0.05 & 1.00 \\
\bottomrule
\end{tabular}}
\par\vspace{3pt}\parbox{\textwidth}{\footnotesize HAC variance with $h-1$ Bartlett lags and the Harvey--Leybourne--Newbold correction; $t_{n-1}$ reference distribution; Holm step-down over the 18 two-sided $p$-values of each horizon.}
\end{table}