\begin{table}[htbp]\centering
\begin{threeparttable}
\caption{Rolling-origin forecasts of next-week log posted price (storefront listings)}\label{tab:fprice}
\small
\begin{tabular}{lrrrr}
\toprule
Model & Origins & N & MAE & RMSE \\
\midrule
banner\_cat\_drift & 4 & 215\,264 & 0.0248 & 0.1084 \\
lightgbm & 4 & 215\,264 & 0.0301 & 0.1000 \\
mean\_reversion\_ref & 4 & 215\,264 & 0.0284 & 0.1025 \\
naive\_last & 4 & 215\,264 & 0.0226 & 0.1083 \\
ridge & 4 & 215\,264 & 0.0299 & 0.0992 \\
\bottomrule
\end{tabular}
\begin{tablenotes}[flushleft]\footnotesize
\item Strict time-based holdouts: train on weeks $\le t-1$, test on week $t$, origins $t = 6,\dots,9$; features are lags only. Lower is better.
\end{tablenotes}
\end{threeparttable}
\end{table}