Tourism demand forecasting presents unique challenges when data exhibit hierarchical structures across geographic levels, requiring forecasts that are accurate and coherent across aggregation levels. This study presents a comprehensive empirical evaluation of hierarchical forecasting approaches using Australian tourism data spanning 77 regions, seven states, and national totals from 1998 to 2016. We systematically compared three automated base forecasting methods (Auto-ARIMA, Auto-ETS, and Auto-CES) combined with four reconciliation strategies (Bottom-Up, Top-Down, and two MinTrace variants) across single-step and multi-step forecasting horizons. Our analysis reveals that the bottom-up approach consistently underperforms owing to error propagation from disaggregated levels, whereas reconciliation methods that enforce coherence constraints produce superior results. The combination of the Auto-ETS base forecaster with top-down reconciliation emerged as the optimal strategy, achieving the lowest error rates (RMSE: 160.13 for single-step, 299.41 for multi-step forecasts) while maintaining computational efficiency. Notably, the theoretically advanced MinTrace methods did not outperform the more straightforward top-down approach in this context. These findings challenge the assumption that complex reconciliation methods necessarily yield superior performance, demonstrating that well-captured aggregate-level patterns can be effectively disaggregated using top-down strategies. The results provide actionable guidance for tourism industry practitioners and extend their applicability to other domains with hierarchical data structures, including retail sales and energy demand forecasting.