Google Trends data and transfer function models to predict tourism demand in Italy

1. Introduction

As a result of the COVID-19 epidemic, which has caused a worldwide health, social and economic emergency, travel and tourism are among the most adversely affected industries, with planes grounded, hotels shuttered and travel restrictions implemented in almost every nation on earth. The number of foreign visitors in 2021 compared to 2019 is drastically down: 73% from Africa, 60% from the Americas, 94% from Asia and the Pacific, 62% from Europe and 75% from the Middle East (Unwto, 2022). The epidemic has highlighted the weaknesses of the current capabilities in the tourism industry when it comes to dealing with the fallout from a crisis of this magnitude and reviving the industry.

Tourist destinations need to seek a new normal and find new balances. In having to adapt to periods of uncertainty, in looking for alternative sources of financing, in repositioning and diversifying markets, products and services, it becomes not only essential to restart but also do it so in a more inclusive, innovative and sustainable way (Unwto, 2021). In this context, a notable contribution comes from the analysis of the megatrends such as the evolving visitor demand, the sustainable tourism growth, the enabling technologies and travel mobility. It is needed to adapt tourism in an increasingly fast-moving, inter-connected and dynamic environment following a strategic foresight approach (OECD, 2018). It is highly desirable to establish a new governance model between the Institutions, strengthening the organizational, financial and human resources and promoting a data-driven decision process. Good policy and planning decisions are made with authoritative, reliable and accessible data (WTTC, 2022). Recent tourism data indicate that these transformations in the tourism sector must take place in a timely manner not only to adequately address negative shocks but also unsustainable over-tourism phenomena already detected in some locations such as Croatia, Montenegro and Italy in last years (Knezevic et al., 2021). The foresight approach helps tourism to take proactive measures to reduce uncertainty, turbulence and ambiguity. Managerial interest in accurate forecasts of future events and changes in tourism demand is related to the extent to which actions can be implemented to influence the demand, monitor fluctuations for planning or resource capacities as well as manage the (for example, socioeconomic) effects of demand. Hence, the success of tourism businesses largely depends on the ability to predict tourism demand most accurately (Hopken et al., 2019) and timely.

Additionally, given that search queries are increasingly seen as a contemporaneous snapshot of collective consciousness due to the increased usage of the internet for news and information in recent years, what people search for is a good indicator of what they will do in the future (Goel et al., 2010). As a result, information from internet searches is recognized as a significant source of linked contemporaneous data (Jackman and Naitram, 2015).

Besides, emerging tech-savvy generations will become increasingly dominant consumers in the tourism market and recent studies also indicate that post-COVID travelers would have an even more digital profile and explore vacation alternatives on internet platforms. Digital solutions that deliver customized and real-time information will be more appealing to tourists (Knezevic et al., 2021).

The above discussion drivers our motivation to answer the following research question: “How to obtain future scenarios using real-time information provision.” The pandemic strengthens the need for more integrated, timely, comparable and granular data. Improving the evidence base, looking beyond traditional survey approaches using alternate and complementary data sources, to inform policy and business decisions is a key policy priority (OECD, 2022).

Therefore, this paper, following a big data-driven foresight approach (Saritas et al., 2022), predicts international tourism demand using a transfer function model on Google Trends data. Google Trends has become a very popular data source among researchers of a wide variety of fields over the last decade, despite some critical issues in quality (Cebriàn and Domenech, 2023).

The work focused on international tourism considering that although domestic travel has been a key factor in the recovery to date and remains essential (Aldao et al., 2022), the full recovery of the global economy will require the full return of international travel (WTTC, 2022).

In detail, it focuses on the US tourism demand in Italy, a country in which the total contribution of tourism and travel to GDP is 13%, to employment is 15% and foreign visitor spend is 8% of total exports (WTTC, 2022). International tourism accounts for 49% in terms of arrivals and the United States of America are the main extra-European market. Americans represent 9% of international arrivals and 30% of extra-European arrivals (Istat, 2019). The US tourist expenditure represents 13% (Bank of Italy, 2020).

Additionally, a rise in interest in national tourism attractions was seen in Italy between January and April 2021, with a proportion of searches for these locations increasing from 55% to 79% (Knezevic et al., 2021). For the public and private operators of the Italy tourist destination, the availability of reliable and accurate forecasts of the US tourism source market is of crucial importance, also in light of the results of recent surveys which show that the impact of the war in Ukraine could hurt travel sentiment to Europe in addition to COVID-19 health and safety concerns (European Travel Commission, 2022).

In the direction of tourism future research, we believe that our study contributes to the theme “tourism and technology” (Sharma et al., 2021) because it answers the research question on how e-tourism can be used for better markets forecast and furthermore raises awareness of foresight literacy in the tourism sector. Secondly, it complements the research on tourism demand forecasting by evaluating the performance of quantitative forecasting techniques on new data sources. The pandemic has reinforced the urgency to have new tourism data measures and complementary data sources to react quickly under uncertainty (OECD, 2022).

Thirdly, it is the first paper (to the best of our knowledge) that makes the United States arrival predictions in Italy. Despite its significance for destination and attraction management, the ability of the current approaches to precisely forecast visitor arrivals at particular attractions is still lacking (Volchek et al., 2019). Many works stop at the evaluation of the models, but few proceed in the forecast of the tourism demand (Huarng and Yu, 2019; Li and Law, 2020; Antolini and Grassini, 2019). Finally, the findings of this study provide immediate valuable scenarios to private and public stakeholders in the tourism origin and destination markets that could be used to make decisions. The projected scenarios, derived with increased forecasting abilities, can help tourist managers choosing which management methods to consider and implement (Clark et al., 2019).

Bergman et al. (2010) distinguish different typologies of future studies according to making truth claims and/or indicating mechanisms: prediction, prognosis, science fiction and utopia (Bergman et al., 2010). However, a claim that is said to be true may be more or less conclusive. A forecast that assumes many outcomes, such as scenarios, based on which mechanisms are active is an example of this. Our work falls into this typology because we think it makes more sense to present a range of potential outcomes rather than a single-point forecast in the context of economic forecasting. The paper is structured as follows: Section 2 provides a brief background Section 3 presents Google Trends data collection and the statistical analysis. Finally, in Section 4 we discuss the result and draw some concluding remarks.

2. Background

Significant losses in service activity were brought on by COVID-19, particularly in the hospitality, food, entertainment and retail sectors and not just in the nations that strongly rely on tourists (Kanchana and Shyama, 2021). The pandemic period highlighted that tourism needs to be better measured, including its economic, environmental and social performance and it needs having of accurate, reliable and timely predictions. Achieving all this is certainly not easy due to the lack of historical data, seasonal fluctuations, influences of unexpected events like hurricanes and eclipses that affect visitation but are unrelated to the previous year’s visitation, the variety of input factors and the complexity of visitors’ travel decision-making process (Song et al., 2010; Clark et al., 2019). Certainly, these results can be achieved more easily favoring the digital transformation of the tourism industry (Akhtar et al., 2021) considered as information intensive (for instance with the implementation of tourism intelligence systems) and facilitating data access, fundamental elements of a big data-driven foresight approach. Foresight is defined as the process of foresight towards the future to try to find out what could be. This process appears significantly in future studies, where the foresight carries it out by applying the research methods based on orientation toward the future (Aldulaini et al., 2022).

As a result, demand modeling and prediction continues to receive a lot of attention from academics and industry professionals and are among the most important areas of tourist research (Hopken et al., 2019).

Several methods have been suggested to increase the forecasting precision of tourist demand: time-series analysis, econometric models, hybrid techniques, nonlinear modeling and artificial intelligence algorithms (Tian et al., 2021; Li and Law, 2020; Jun et al., 2018; Li et al., 2005). Among the quantitative methods applied to tourism demand forecasting transfer function models (describing the relationship between an output variable Y_t at time t and k input variables X_1t, …, X_kt) have also attracted the attention of scholars (Dinis et al., 2017; Gnanapragasam and Cooray, 2016; Kulendran and Witt, 2003). But traditional studies based on historical statistics, have a certain hysteresis quality, which restricts the accuracy of prediction over the short term. Additionally, forecasting based on the conventional historical data is frequently impacted by unpredictable events like natural catastrophes, which lowers the accuracy of prediction (Feng et al., 2019). Furthermore, the official statistical systems rarely generate tourism data that can be used to drive policy decisions. The current tourist statistics directive suggests gathering monthly data on inbound and outbound travels, including regional (NUTS2) specificity. More punctual information (i.e. at provincial or municipal level) is published annually – monthly only in some countries – with a time delay that in the more recent Italian official survey ranges from 11 months to 23 months (Emili et al., 2020). In order to obtain more precise predictions of visitor arrivals at the destination-level, new data sources are being used increasingly frequently. The amount of information that the individuals can access online has greatly expanded thanks to the internet’s quick development and the proliferation of connected devices like laptops and mobile phones, which has also caused a revolution in the travel and tourism sector. The relationship between travelers and travel agencies has evolved as a result of new technology and online services. For instance, consumers may now compare thousands of hotels, arrange their entrance to a museum, or reserve a table at a restaurant using the internet. Tourism intelligence systems or destinations with dashboards, which provide real-time aggregate data on tourist arrivals, overnights, transport and spending, are important for smart and data-driven decision-making and are a vivid example of data utilization for both business and destinations. The tourist interactions with the internet, such as search engine data, web traffic, social media and mobile data, have generated structured and unstructured online data that can be captured, saved and analyzed and have gained popularity in predicting tourism demand (Li et al., 2021; Vu et al., 2018; Yang et al., 2015). Traces of access can be recorded when a tourist searches or makes a reservation online. Because travelers typically make plans in advance, aggregated internet behavioral data can be used to gauge demand for travel (Pan et al., 2012).

Tourism demand often has a natural link with new web-based data sources like search engine traffic, Website traffic or customer reviews on online review sites. These huge data sources have become more widely available, and as a result, forecasting tourist demand has also been done using them (Hopken et al., 2019). Literature has recently concentrated on taking advantage of the useful information search query data on tourist behavior provides. The number of travelers who use the internet to prepare for and book their business and leisure travels has dramatically increased over the past few years (Camacho and Pacce, 2018). Even for official statistics, big data are a vital source of information. International and national statistical offices are really conducting pilot studies on various big data sources today to see how well they would be able to provide or integrate official statistical data. Such data are primarily of importance because they can offer more time-saving and cost-effective information, especially at the local-level (Antolini and Grassini, 2019; Naccarato et al., 2018). Online data provide specific advantages over traditional data (e.g. survey-based data), such as higher frequency (i.e. daily, weekly and monthly) and the capacity to reflect people’s behavior from different angles (Pan and Yang, 2017). By examining the themes that emerge from the online discourse and researching the dynamics of user interaction, big data from online social networks can be used to predict rapid changes in tourist preferences and popularity trends for destinations and local attractions (Franzetti Colladon et al., 2019). As a result, academics have added online data to traditional data, extracted business value from it and used it in the predicting models (Andariesta and Wasesa, 2022). The ability to track and reliably predict visitor behavior has been transformed by the availability of real-time, high-volume and high-frequency data (Yang et al., 2015). Search data for tourist information has been extensively exploited to improve the forecasting capability of the current models (Park et al., 2017). A range of methods to apply search queries to predict tourist arrivals at a destination-level has been proposed (Antolini and Grassini, 2019; Artola et al., 2015; Bangwayo-Skeete and Skeete, 2015; Choi and Varian, 2009; Dergiades et al., 2018; Hopken et al., 2019; Li et al., 2018; Onder, 2017; Gunter and Onder, 2016; Park et al., 2017; Yang et al., 2015). The majority of existing studies demonstrate that online search engine data can improve the forecasting performance of tourism demand (Li and Law, 2020). Since search query data reveals information about the interests, intentions and opinions of travelers, it is useful for predicting travel demand. Tourists utilize search engines to learn more about their vacation destinations, routes, must-see attractions and other travelers' reviews and opinions of those locations. To predict the number of visitors who will actually visit a location, data from search queries may be employed.

Since ICTs and mobile computing devices have developed, the internet has become the primary source of travel information (Claude, 2020). Search engines, such as Google that dominate over a billion of worldwide Internet, Baidu and Yandex that represent the fastest growing emerging markets (i.e. China and Russia), shape tourist perceptions of attraction image and encourage travel decisions. In this paper, Google’s data were analyzed considering that most Americans have been using this search engine for some time to plan travel (Pan et al., 2011). Google began to release information on its users’ search queries through the publicly available Google Insights interface, now rebranded as Google Trends.

Since it began operating in 2004, Google Trends has provided search frequency of search keywords, which displays the comparison between the volume of searches for a given keyword and all other keywords over a given period of time. It then further normalizes search frequency into the range [0, 100], preventing changes in the volume of searches caused by an increase in users. The software displays the frequency of a specific keyword being used as a search term for the entire number of searches across multiple nations and languages (Choi and Varian, 2009). As a result, recent research has remarkably shown that Google Trends data reflect important elements of visitors' keyword-based queries, which in turn opens up a wealth of options to research and predict travelers' planned behavior (Dinis et al., 2019; Hopken et al., 2019; Yang et al., 2015; Kim et al., 2016; Hopken et al., 2021).

3. Google Trends data collection and analysis

4. Conclusions

This work adopts a big data-driven foresight approach to predict the international visitor arrivals in Italy using Transfer Function on the US Google Trends searches data. Our analysis remembering the emerged tech–savvy generations in the evolving visitor demand megatrend demonstrate Google Trends’ excellent potential for forecasting tourism demand in timely manner, particularly useful in a situation where events, like the COVID-19 outbreak or the Russian–Ukrainian conflict, have increased uncertainty in travel planning. Airline tickets are being booked increasingly close to departure, with a reduction between purchase and actual passage, for example, on average for Americans it has gone from 103 to 80 days, while for Japanese even from 76 to 32 days. Keeping in mind that the theory of forecasting is based on the premise that the current and past knowledge can be used to make predictions about the future (Petropoulos et al., 2022), the results obtained from complementary between theoretical developments and applications using transfer function on Google Trends data are in line with the opportunity to implement foresight approach based on new instruments in the tourism industry.

A digital transformation based on digital infrastructure, data strategy, augumented/virtual reality would allow the tourism industry to benefit from timely forecasts for smart and better decision-making.

In this paper, two different models using monthly data have been evaluated. The first model is an ARIMA(0,1,0) × (2,0,0)₁₂ model without any intervention. The second model allows transient-level changes (in particular a transient level change is expected in March 2020). An automatic procedure for the detection of possible transient level changes has been applied. Two scenarios emerged from the analysis characterized by different specific conditions. The first scenario, could be defined as prudent where the effect of perceived potential risk on travel and tourism decision-making after the COVID-19 crisis are still present (Matiza, 2022) and where forecasts are still struggling to reach pre-pandemic levels and the second, defined as optimistic in which the desire to be able to return to travel means that the average forecasts are in fact exceeded by the reality. The second scenario envisaged is reflected both in the data of the first half of 2022 which see are turn of tourist flows in Italy, including international ones and among these, above all Americans (also thanks to a favorable euro/dollar exchange rate) and in the greater confidence in the future on the part of tour operators. If the hypothesis of prudent scenario is compatible with niche phenomena such as tourism zero kilometer and slow (Monaco, 2021), with a resilient approach to adaptation and transformation (Kwok and Koh, 2022) and with the promotion of domestic tourism to fill the gaps left by international tourism, the optimistic one need evaluation of possible over-tourism phenomena, of safety in tourism (Monaco, 2021) and in general, of tourism well-being above all in a sustainability perspective (Larry, 2020).

While aware that factors such as subjectivity in the choice of keywords, Google Trends data quality and limitations of transfer function used only for linear systems (Lai, 1979) could have influenced the results obtained, it is believed that the path identified can make a significant contribution to take proactive measures to reduce uncertainty and effectively prepare for megatrends that will significantly impact tourism. In a future perspective, it could be interesting to observe the scenario generated by the application of the methodology to other origin markets of the international tourism in Italy or by data carried out on other search engines. Furthermore, external variables could be included in the analysis, considering a transfer function in which some external variables are added to the Google Trends values. In addition, obtaining forecasts through the use of transfer function models on Google Trends data could be a valid methodological tool also for other socioeconomic or environmental phenomena. For example, stakeholders of new emerged forms of tourism (i.e. wellness, health, volunteering and bleisure) could benefit from the forecasts obtained with the transfer function models from Google searches. Moreover, it could be useful to mitigate the vulnerability of the tourism sector due to climate change (Leal Filho, 2022).

In conclusions, the availability of a plurality of scenarios allows stakeholders to evaluate in a more coherent way the implications of political choices that can range from investment to restore consumer confidence, to seize the marketing opportunities offered by virtual reality (Mohanty et al., 2020; Monaco and Sacchi, 2023), to improve digital skills and infrastructure (Knezevic et al., 2021) to produce quick response and effective decision-making for future crises, to use of targeted and personalized marketing channels and quality improvement.