Treffer: Modeling and forecasting the 'new' flow of tourism in Portugal: envisioning a strategic management.

Purpose: Tourism, as a concept and activity, has evolved over time, specially these last decades (Caldas et al., 2020). Being a fundamental economic activity to generate employment and wealth (Ramos et al., 2019), it is an engine for the growth of several economies, motivating the creation and development of companies, becoming a strategic pillar (Zhang et al., 2021). In Portugal, the tourist activity employs approximately half a million people (Ramos & Costa, 2017). However, we are experiencing an adverse scenario due to the Covid-19 pandemic, whose impact is important to understand. The need to adapt and learnis explained by globalization, uncertainly and change (Machado et al., 2020). Forthat, it is necessary to analyze and forecast the flow of tourism (internal and external) so that the entities involved in the sector can define an action plan, endowed with strategic and competitive management. Given the changes in the 'recent' dynamics of tourism data, which are a clear obstacle to the modeling andforecasting process (Pesaran et al., 2006), this article discusses the contributionsand limitations of using classical forecasting methodologies, pointing out possible alternative methodologies. Methodology: The study focuses on the modeling and forecasting of time series between January 2002 and March 20229 . According to our research (Ramos, 2021; Lopes et al., 2021), we decided to use Exponential Smoothing (ETS) methodologies. These are a classic in time series modeling, for presenting reliable forecasts that are fast to implement and simple to understand, which is an advantage for the business world (Ord, 2004; Hyndman & Athanasopoulos, 2018). Hyndman & Athanasopoulos (2018) consider three types of ETS models,such that the simplest method assumes a stationary behavior, the secondexplicitly deals with the existence of a trend, while the third includes alsoseasonality. In computational terms, we used the Jupyter Notebook computing environment, with the Python programming language (specifically version 3.7.3).The code was the one developed in Ramos (2021), being open source in Lopes and Ramos (2020). Results: The results were presented and discussed through the analysis of two time series: (1) Total number of overnight stays in tourist accommodation establishments in Portugal – Total series; (2) Number of overnight stays of residents in tourist accommodation establishments in Portugal – Residents series. Overall, from the analysis of time series (descriptive and inferential analysis), there has been growth in Tourism in Portugal since 2002, with a visiblebreak in 2020 (Covid-19 pandemic). It is important to mention that it was “Residents” tourism that, in the pandemic period, dictated the dynamics in “Total”tourism, with the recovery mainly due to domestic tourism. The 'best models' ETSwere adjusted and selected, obtaining out-of-sample forecasts for the months ofJanuary, February and March of 2022. The error metric used in the evaluation was Mean Absolute Percentage Error (MAPE). It was decided to consider, for each series, all models that showed convergence, in parallel with the informationobtained from the AIC (Akaike, 1974) and BIC (Schwarz, 1978) criteria. In Total series, the selected ETS models corresponded positively, although with some error (MAPE – Jan: 9.29%; Feb: 11.05%; Mar: 10.44%). However, the same wasnot verified for the Residents series. As it has a completely atypical recent dynamic, it appears that the potentially more suitable ETS models (models with trend and seasonality) do not generally converge. The only models that show convergence are models without a seasonal componente, which does not agree with historical data. In this case we have considerably larger errors (MAPE – Jan: 30.81%; Feb: 16.8%; Mar: 16.15%). Research limitations: We recognize how reductive it is to analyse the flow of tourism based on the “number of overnight stays in tourist accommodation establishments in Portugal”. However, as it is a difficult phenomenon to measure,it was necessary to define an analysis strategy that would allow us to quantify it.In terms of the methodology used, the literature points to the success of ETS methodologies in time series with the characteristics present in the series under study. However the limitations of these methodologies, in dealing with abrupt breaks in structure in the data history, was evident in this study. In this sense, it is necessary to search for alternative forecasting methodologies. Machine Learning methodologies, namely Deep Learning (Deep Neural Networks) have been pointed out as quite promising (Ramos, 2021). This will be the next path inthis investigation to obtain more accurate models, which can be used in practicalterms. Originality: Given the importance that Tourism has both in the economic and social dimension of Portugal, and being a very volatile and constantly changing sector, it is imperative to define a strategy for future action. We need to understand how we can reinvent ourselves and being able to cope with situationsof external dependence. The use of (classical) forecasting models can be an excellent tool. If, on the one hand, the modeling process helps a better understanding of the phenomenon, the predictive aspect is a powerful support measurement tool to decision making. Although the next step will be the search for more accurate models, from a strategic point of view, it is important to emphasize that this study carried out points out that policies such as “Vá para fora cá dentro” (Go ‘outside’ inside) cannot be neglected. From this study, it becomes evident the need for a strategic management capable of promoting internal tourism, alerting us not to become dependent on external tourism. [ABSTRACT FROM AUTHOR]

Copyright of International Conference on Applied Business & Management is the property of ISAG: European Business School and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)