Analysis · Finance
A Forecast Can Miss Less Yet Get the Direction Wrong
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
For Cyprus, the model with the lowest average forecasting error was not the best at predicting whether tourist arrivals would rise or fall. That split in a 2017 study by Hossein Hassani and colleagues shows why a claim of forecast accuracy needs a clearly named yardstick.
In Forecasting Accuracy Evaluation of Tourist Arrivals, the researchers compared nine forecasting methods using monthly Eurostat data from January 2000 to December 2013. They used observations through April 2011 for training and testing, reserving May 2011 onward for validation. Forecast horizons ranged from one to 24 months.
One measure was root mean squared error, which penalizes larger numerical misses more heavily. Another counted how often forecasts correctly predicted the direction of change. The researchers also applied statistical tests to compare forecasting errors. These measures answer different questions: how far off was the estimate, and did it point the right way?
For Cyprus, recurrent Singular Spectrum Analysis, or SSA-R, had the lowest average error across the five horizons. Yet its average direction score was 41%, compared with 78% for the related vector method, SSA-V. These are averages across horizons, not a single pooled success rate. SSA separates patterns from components treated as noise before forecasting; its recurrent and vector versions extend those patterns differently.
The comparison has boundaries. It tested particular implementations using historical arrivals alone, without additional economic predictors. Its validation period was limited, and some narrative claims conflict with the tables. Poor performance by the tested neural network therefore cannot establish that neural networks generally fail. Nor does predictive performance identify what causes tourism demand to change.
For readers assessing market forecasts, the useful lesson is to ask for the target, horizon, benchmark and scoring rule. A smaller numerical error does not automatically mean a more reliable growth signal, and neither result establishes a company's future revenue.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Forecasting accuracy evaluation of tourist arrivals Hossein Hassani, Emmanuel Sirimal Silva, Nikolaos Antonakakis …
