How well can ChatGPT forecast tourism demand?

ChatGPT has demonstrated remarkable capabilities across various natural language processing (NLP) tasks. However, its potential for forecasting tourism demand from temporal data, specifically historical tourism arrivals data, remains an unexplored frontier. This research presents the first attempt to conduct an extensive Zero-shot and Chain-of-Thought analysis of ChatGPT's capabilities in tourism demand forecasting, under various temporal scenarios. Based on the Macau inbound tourism arrivals dataset, our empirical findings indicate that the predictive capability of ChatGPT-4 is noteworthy compared to the three benchmark time series models (Naïve, Exponential Smoothing, SARIMA) and the three benchmark machine learning models (Random Forest, Multi-Layer Perceptron, Long Short-Term Memory), especially when the forecast horizon is relatively short. Furthermore, compared to Zero-shot prompts, engaging in continuous dialogue can enhance the forecast accuracy of ChatGPT-4. This performance of ChatGPT highlights its potential for quantitative data prediction as a new user-friendly and cost-effective management tool.