Abstract
Tourist decision to visit attractions is a complex process influenced by multiple factors of individual context. This study investigates how the accuracy of tourism demand forecasting can be improved at the micro-level by predicting the number of visits to London museums. The number of visits to London museums is forecasted and the predictive powers of Naïve I, seasonal Naïve, SARMA, SARMAX, SARMAX-MIDAS and artificial neural network models are compared. The empirical findings extend understanding of different types of data and forecasting algorithms to the level of specific attractions. Introducing the Google Trends index on pure time-series models enhances forecasts of the volume of arrivals to attractions. However, none of the applied models outperforms the others in all situations. Different models’ forecasting accuracy varies for short- and long-term demand predictions. The application of higher-frequency search query data allows generation of weekly predictions, which are essential for attraction- and destination-level planning.