Timely monitoring of tourism demand is essential for destination management, yet official monthly arrival statistics are often released with delays and can be difficult to use for near-real-time decision-making, particularly under structural shocks such as coronavirus disease 2019 (COVID-19). This study develops a fully reproducible, open-data nowcasting pipeline for Ecuador’s international tourist arrivals using a Python workflow. The framework integrates (i) the official monthly arrivals series published by Ecuador’s Ministry of Tourism (MINTUR), (ii) open online attention proxies from Wikipedia pageviews retrieved via the Wikimedia REST application programming interface (API), and (iii) open climate covariates derived from the ERA5-Land land reanalysis. Multiple forecasting models are evaluated under a rolling-origin, one-step-ahead backtest, with a mandatory seasonal naïve benchmark and a regime-aware assessment that separates a stress-test window (2019–2021) from an operational post-COVID window (2022–2025). Forecast accuracy is summarized using root mean squared error (RMSE), mean absolute error (MAE), and symmetric mean absolute percentage error (sMAPE), and statistical significance of performance differences is assessed using the Diebold–Mariano (DM) test. Results show that a ridge-regularized autoregressive model (ridgeₐr) achieves the best overall accuracy, reducing RMSE by approximately 79% relative to the seasonal naïve baseline over the full evaluation window. Windowed results confirm robust performance during the shock period and sustained improvements in the post-2022 operational regime, while the incremental benefit of broader exogenous signals is heterogeneous across windows, underscoring the importance of regularization and regime-aware reporting. The proposed approach provides a transparent, low-cost blueprint for reproducible tourism monitoring that is transferable to other destinations using open data and standard computational tools.
Guerra et al. (Mon,) studied this question.