High-frequency analysis in tourism: a novel approach to crisis management and demand forecasting

Abstract

Technological advancements are transforming city operations and urban planning. The COVID-19 pandemic highlighted the critical need for timely, granular and accurate tourism data and forecasts, as the limited availability of real-time information on tourism demand complicated the ability to make rapid and informed decisions. Traditional applications and research in tourism demand forecasting predominantly rely on irregular, low-frequency survey data to analyse international visitor flows. As a result, these methods are primarily targeted at high level policymakers and long-term planners. This presents a substantial research gap in the investigation of granular, high-frequency tourism demand forecasting applications that assist destination managers in making informed and targeted decisions in the short term. This dissertation addresses this gap by utilising high-granularity, hourly-level credit card and mobility data to measure tourism demand to enhance short-term decision-making through the development of novel methodologies within crisis impact assessment and destination demand forecasting applications. Specifically, we focus on a major tourist destination, the Gold Coast, Australia, where prior to the COVID-19 pandemic, tourism accounted for 12.3% of employment and contributed $12 billion to the local economy annually (EconomyId, 2021).

We develop a novel integrated intervention and forecasting methodology for crisis impact assessment that leverages daily anonymised and aggregated Visa card data to quantify rapid changes in expenditure. This methodology is then applied to investigate the economic effects of border restrictions on tourism on the Gold Coast. We find that the impact of border restrictions placed on Gold Coast visitor markets is influenced by policy severity, proximity to destinations and broader pandemic conditions. Importantly, we provide evidence that domestic visitors drive recovery, but this primarily occurs via intrastate visitors as interstate visitors face barriers through non-reciprocal travel arrangements, frequently changing restrictions and unstable pandemic conditions in visitor origins. Further, we find that travel restrictions disproportionately affect tourist-oriented businesses and alter the spatial distribution of spend of visitors within the destination.

The existing literature’s focus on low-resolution applications is reflected in both tourism demand forecasting and impact assessment methodologies. These methodologies are not well suited for practical high-frequency applications due to increased task complexity, such as larger scopes, longer forecast horizons, and multiple seasonal patterns. Furthermore, high-frequency applications introduce operational challenges, including the need for automated procedures, ongoing performance monitoring and model updating. To address these needs, we introduce a high-frequency forecasting framework tailored for high-resolution data, which is aligned with our proposed impact assessment methodology to meet the dynamic and practical needs of high-frequency forecasting in tourism. The framework is applied to tourism demand forecasting across a destination at an hourly level, which has received very limited attention in the literature. Utilising anonymised and aggregated DSpark mobility data, we perform a comparative analysis between traditional statistical time-series forecasting models fit locally on individual series and emerging temporal neural network architectures fit globally on multiple series. This novel analysis includes evaluating forecast decay over long horizons across different techniques and evaluating the effect of regular model updates on performance. The findings demonstrate that neural network approaches, capable of handling multiple connected time series, long-term forecasting horizons and capturing complex seasonal patterns through their non-linear structures, significantly outperform traditional time-series statistical methods in this setting. Additionally, the results provide evidence that regularly scheduled model updates, conducted weekly to integrate new data, enhance forecast accuracy over extended periods.

This research establishes the importance of integrating high-resolution data into impact assessment and demand forecasting in smart tourism contexts, offering robust methodological tools for improving real-time decision-making in tourism management.

Date of Award	3 Dec 2025
Original language	English
Supervisor	Adrian Gepp (Supervisor), Bruce Vanstone (Supervisor) & James Todd (Supervisor)