Reproducible National Tourism Accommodation-Demand Forecasting under Temporal Extrapolation and Leakage Audits
Year: 2026
Type: preprint
Abstract: Confirmatory computational forecasting study using official aggregated Eurostat TOUR_OCC_NIM observations. The primary unit is country-month and the primary series is the German national monthly slice (geo=DE; freq=M; c_resid=TOTAL; unit=NR; nace_r2=I551-I553). The study tests whether a sample-size-matched random row split yields optimistic errors relative to strict expanding-window temporal extrapolation, whether three fixed candidate models show a stable relative advantage under matched data, origins, seeds and CPU budgets, and whether results remain stable across pre-specified repeated seeds. The primary outcome is MAE; RMSE, MASE, non-zero-target MAPE, runtime, RAM and model rank are secondary or audit measures. The analysis is pre-specified before complete data retrieval; the NUTS2 monthly empty probe remains out of scope. Full methods, thresholds, provenance and deviation rules are frozen in the local protocol; no results are reported here.
Author: Wang Qiushi
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7204579404