Search query and tourism forecasting during the pandemic: When and where can digital footprints be helpful as predictors?

Highlights

• We predicted daily tourism demand across 74 countries in 2020.

• A lasso method was used to incorporate Google search queries as predictors.

• We evaluated the usefulness of search queries in boosting forecasting accuracy.

• Usefulness of search queries in forecasting is associated with pandemic severity.

Abstract

During the COVID-19 pandemic, daily tourism demand forecasting provides actionable insight on tourism operations amid intense uncertainty. This paper applies the lasso method to predict daily tourism demand across 74 countries in 2020. We evaluate the usefulness of online search queries in boosting forecasting accuracy. The lasso method is used to select appropriate predictors and their lag orders. Results indicate that, in general, no evidence supports the usefulness of Google Trends data in generating more accurate forecasts. However, in some countries, the data can be useful for reducing the forecasting errors. Regression analysis further demonstrates that the relative usefulness of online search queries is associated with pandemic severity, the dominance of inbound tourism, and island geography. Lastly, implications are provided.

Introduction

The outbreak of COVID-19, a contagious disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has significantly affected the global tourism and travel industry. Declared a Public Health Emergency of International Concern on January 30, 2021, the pandemic has led to 180,817,269 confirmed cases and 3,923,238 deaths as of June 28, 2021 (WHO, 2021). Tourism and travel involve human movement and interaction and may contribute to the spread of disease (Farzanegan et al., 2021). Therefore, many countries have shut down their borders and enforced mobility restrictions; these policies dealt a devastating blow to the global tourism industry. As demand plummeted, many tourism businesses, such as airlines, hotels, and attractions, laid off and furloughed employees. A supply shrink followed. According to UNWTO statistics, the pandemic led to a loss of US $910 billion to US $1.2 trillion from tourism exports and was projected to reduce global GDP by 1.5% to 2.8% in 2020 (UNWTO, 2021).

During this highly uncertain time, tourism forecasting has come to play a more important role than ever in informing decisions among government personnel, industry professionals, and other tourism stakeholders. First, forecasting allows for tourism recovery predictions and can provide valuable insight for tourism policy design and implementation. Second, as demand has been volatile during the pandemic, accurate forecasts can help industry players better allocate their resources (e.g., inventory and staffing) to serve incoming demand. In addition to industry forecasting reports, tourism researchers have striven to develop and evaluate multiple forecasting methods to predict tourism demand at different levels (Kourentzes et al., 2021; Liu et al., 2021; Qiu et al., 2021). Various forecasting methods have since been leveraged and compared in terms of forecasting accuracy. Examples include time series models (Wickramasinghe & Ratnasiri, 2021), artificial neural network (ANN) models (Polyzos et al., 2021), stacking models (Qiu et al., 2021), forecasting combinations (Kourentzes et al., 2021), and the scenario-based mixed judgmental forecasting method (Liu et al., 2021).

Many tourism studies have highlighted the importance of big data indicators in enhancing tourism forecasting accuracy (Tian et al., 2021). Digital footprints, such as online search queries, web traffic, and social media posts, reflect prospective travelers' tourism interests and can reasonably predict tourism demand. Customer information processing theory suggests that consumers' path to purchase entails a funnel-like process from search to purchase. Digital footprints can reflect information about consumers' potential choices, some of which will be bought. Empirical studies have confirmed the usefulness of these footprints in improving tourism forecasting accuracy in various contexts (Pan et al., 2012; Pan and Yang, 2017a, Pan and Yang, 2017b); however, it remains unclear whether this usefulness applies in the COVID-19 era due to the volatility of tourism demand and travelers' unique information search behavior during this crisis.

To bridge this knowledge gap, we adopted the lasso method to forecast daily tourism demand across 74 countries in 2020. The lasso method is particularly useful when selecting variables from high-frequency, high-dimensional predictors of tourism demand (Tian et al., 2021). Different predictors were incorporated into our model to improve forecasting accuracy. More specifically, we compared the forecasting performance of a model with and without Google search query data. Doing so enabled us to examine the effectiveness of digital footprints in reducing forecasting error during the pandemic. Following our forecasting performance comparison, we employed regression analysis to understand when and where Google search query data were most helpful in improving forecasting accuracy. Results make at least three major contributions to the literature. First, we evaluated models' forecasting performance for daily tourism demand during the COVID-19 pandemic, providing guidelines for model and predictor selection in a time of great uncertainty. Our findings, based on a global sample, offer generalizable implications for academics and practitioners. Second, we scrutinized whether Google search query data (as a typical form of digital footprints) can improve forecasting accuracy across different forecasting horizons. Our conclusions helped evaluate the potential usefulness of digital footprints in tourism forecasting during this chaotic time. Last but not least, we systematically assessed factors explaining the effectiveness of Google search query data in tourism forecasting via rigorous regression analysis of results from 11,396 forecasting models. Although previous meta-analyses addressed potential determinants of forecasting accuracy (Kim & Schwartz, 2013; Peng et al., 2014), no single study has systematically evaluated these contributions in a context allowing for large-scale model comparison. Our findings thus clarify predictors' applicability in tourism forecasting.