Abstract
The United Kingdom constitutes the principal tourist source market for Spain. This research note analyzes the immediate impact of the Brexit on British tourism in Spain using the Bayesian structural time series models framework. The results obtained show that between July 2016 and September 2017, Brexit has not produced any initial negative effect on the arrival of British tourists or on their spending in Spain.
Keywords
On June 23, 2016, a referendum was held in the United Kingdom about whether this country should stay in the European Union (EU). The result in favor of leaving the EU and its notification through a letter on March 29, 2017, triggered the negotiation process which will culminate, in a maximum period of 2 years, the exit process.
From a theoretical point of view (see Maesso Corral, 2011), in the short term, economic integration processes generate static effects such as the creation and deviation of trade (Viner, 1950), and in the long term, dynamic effects linked to the use of comparative advantages, the emergence of economies of scale, intensified competition, and an increase in competitiveness (Balassa, 1961). It could, therefore, be expected that an economic disintegration process, such as Brexit, would give rise to opposite effects, with potential reductions in growth and trade creation.
For Spain, and particularly for its tourism industry, Brexit represents a serious threat and, in econometric terms, a structural shock. Espasa, Gómez-Churruca, and Jareño (1990) revealed that, after Spain’s entry into the European Economic Community in 1986, it was precisely the tourists from the United Kingdom which gained the most weight (see Table 1). The United Kingdom’s exit from the EU could affect—although now negatively—the flows of British tourists toward Spain to a greater extent than the tourist flows from other countries.
Tourists Entering Spain, Different Nationalities: Relative Weight of the Total.
Note. Average percentages for the period.
The threat is even more relevant because today, the United Kingdom constitutes the principal tourist source market for Spain. During 2016, it is estimated that more than 17 million British tourists visited Spain (representing 23% of this country’s total international tourists), accounting for more than 16 billion euros (20.80% of total international tourist spending for this year). The United Kingdom is followed by France (with a little over 11 million tourists and almost 7 billion euros) and Germany (also with just over 11 million tourists and more than 11 billion euros) (Instituto Nacional de Estadística de España [INE], 2017a, 2017b). British tourism in Spain is concentrated in a few regions, principally the Canary Islands, the Balearic Islands, Andalusia, the Region of Valencia, Catalonia, and Madrid.
The impact of Brexit on British tourism in Spain is surrounded by uncertainty, given that in addition to the insecurity associated to the already complex phenomenon, we must also add the overall variability of the evolution of the main competing destinations of the Spanish Mediterranean coast. In this respect, it can be expected that the final impact that the process has on the Spanish tourism industry may be influenced (enlarged or attenuated, depending on the case) by the evolution of the situation of countries such as Turkey, Greece, or Egypt, traditional competitors of Spain in the tourism market. Similarly, the impact will also depend on whether the negotiating process is hard or soft.
A hard Brexit is understood as a process that will lead to the complete rupture between the institutions and the acquis communautaire and the United Kingdom. This country would be considered as a third state, and as such, its relations would be subject to a series of bilateral agreements currently in force or to the regulations of private international law of each of the member states. A soft Brexit, on the contrary, would be associated to maintaining the majority of the legal relations between the United Kingdom and the rest of the EU, following the model of associated states such as Norway or with a special regime negotiated through agreements in some, several, or all of the sectors currently regulated by EU regulations. Logically, the consequences of each of the two scenarios are, necessarily, divergent. The outcome of the hard Brexit scenario would be much more traumatic than that of a soft Brexit in which the effects would probably be marginal.
However, Brexit began to have an impact since its announcement. On the same day as the results of the referendum were known, the pound depreciated significantly with respect to the dollar and the euro (Allen, Treanor, & Goodley, 2016) and a downward review of the economic growth forecasts for the United Kingdom was announced of between 1% and 2% (International Monetary Fund, 2016). This note analyzes the immediate or initial impact of Brexit on the arrivals of British tourists to Spain and their spending. Immediate impact is understood as that arising from the result of the vote, and the subsequent activation of the process through the notification to the Commission of the British will to abandon the EU. To determine this impact, we have used the Bayesian structural time series models framework proposed by Brodersen, Gallusser, Koehler, Remy, and Scott (2015) and implemented in the CausalImpact R-package (R Core Team, 2017).
Although the origin of the Bayesian methods dates back to the mid–18th century, their use has become more widespread in the fields of statistics, engineering, philosophy, and econometrics with the development of more and better equipment that enable a greater computing capacity (Gelman et al., 2013). One of the fields in which the Bayesian methods are most extensively used is causal analysis, not only with experimental data but also with observational data (Pearl, Glymour, & Jewell, 2016), whereby, through probabilistic graphical models and the establishment of counterfactuals, the effects that different interventions have on the outcome variables are analyzed using machine learning estimation and prediction algorithms (Bellot, 2016; Koller & Friedman, 2009).
Another field of application in which the Bayesian developed notably is time series analysis, in which they help to model the nonlinearities that characterize the behavior of many of these series. Different examples of the application of these methods are the Bayesian Vector Autoregressive models (Bayesian VAR models), Bayesian Autoregressive Integrated Moving Average models (Bayesian ARIMA models), change point, or anomaly detection literature. This is also the case of nonlinear state space models, the category to which the Bayesian structural time series model belongs, where the standard way of evaluating the implicated integrals is via Monte Carlo simulation which resolves numerical instabilities due to complex (and multivariate) problems (Teräsvita, TjØstheim, & Granger, 2010). The principal uses of the technique are the long-term and short-term (nowcasting) prediction of time series and inferring causal impact, as in the case of this article.
In short, the Bayesian structural time series are those used to approach the analysis of structural time series. Structural time series models can be defined in terms of a pair of equations:
where
Structural time series models are flexible and represent an advantage with respect to intervention analysis using solely ARIMA models. This is because the former enable the creation of counterfactual predictions by constructing a synthetic control based on a combination of markets that have not been treated. This is not possible using the ARIMA methodology, which only takes into account the evolution of the target country under consideration before and after the event without controls.
In Bayesian structural time series analysis, a Bayesian approach is used for inference and estimates. In a nutshell, the Bayesian version of the model consists of three main parts: the Kalman filter—used for time series decomposition, adding different state variables such as trend, seasonality, regression, and others; the spike-and-slab method—where the most important regression predictors are selected; and the Bayesian model averaging process where a combination of the results and predictions is performed (Scott & Varian, 2014). Despite its complicated mathematical underpinning, Bayesian structural time series analysis could be performed using several R packages such as BSTS (Scott, 2017) or CausalImpact (Brodersen et al., 2015). In the case of the use of this methodology for causal analysis, the latter package is used, and the causal impact is estimated by subtracting the predicted response obtained from a semiparametric Bayesian posterior estimation and the observed response during the post-intervention—in our case, post-Brexit (Brodersen et al., 2015).
Recent examples of the application of the Bayesian methods, associated to the analysis of large amounts of data (Big data) or time series, can be found in the field of marketing (Rossi, Allenby, & McCulloch, 2005), branding and retailing (Pancras, Gauri & Talukdar, 2013), efficiency studies (Griffin & Steel, 2007), trend prediction (Scott & Varian, 2014), price evolution analysis (Ludwig, Feuerriegel & Neumann, 2016), the impact of housing conservation public policies (Schmitt, Tull, & Atwater, 2017), and so on. In tourism, where the use of the Bayesian methods has also become more widespread, recent applications can be found in the context of measuring the efficiency of destinations, infrastructures, and tourism companies (A. Assaf, 2010; A. G. Assaf, Oh, & Tsionas, 2016; Tsionas & Assaf, 2014) and in the analysis of the impact of the Arab uprisings in certain Mediterranean tourist destinations (Perles, Ramón, Moreno, & Torregrosa, 2016).
In this study, as a preliminary step before estimating the impact using the methodology proposed by Brodersen et al. (2015), and to test the robustness of the results obtained using this method, a univariate analysis of the time series is made using the Box–Jenkins methodology (ARIMA). The estimate is made using the automated procedures in the forecast 9.1 library (Hyndman, 2017) for the programming language R 3.4.1 (R Core Team, 2017).
With respect to the data, it can be said that, for the estimate, monthly data provided by the Spanish National Institute (INE, 2017a, 2017b) and Instituto de Estudios Turísticos (Turespaña) through FRONTUR (Estadística de Movimientos Turísticos en Frontera) on number of tourism arrivals (in thousands) and EGATUR (Encuesta de Gasto Turístico) on tourism expenditures (in millions of euros) for the period between January 2012 and September 2017 have been used. The purpose of selecting this time frame is twofold. First, it is expected to include data until at least the end of the main season of 2017, given that the pre-booking of travel prior to Brexit may mean that it would be unlikely that decision making would change as would be the case if only that prior to the main season was considered. Second, the Brexit effect will be isolated from other potential structural changes caused by the tourism crisis of 2008-2009 (Perles, Ramón, Rubia, & Moreno, 2016) and the distorting effects of the outbreak of the Arab uprisings of 2011 before the selected sample period. In any event, the composition and size of the sample are adapted to the recommendations of Brodersen (2016) when using this methodology, taking a pre-intervention period of approximately 2 or 3 times the length of the post-intervention period.
A difficulty associated to the use of these series of arrivals and tourist spending resides in the break caused by the methodological change introduced by the INE when it took over the survey that had formerly been undertaken by the Instituto de Estudios Turísticos in October 2015 (INE, 2015). Fortunately, the INE has provided linked growth rates and a detailed procedure to give continuity to both of the series (INE, 2015). This procedure has been followed in this research note. It may be observed that in comparison with the original series provided by Turespaña, the linking procedure provided by the INE substantially raises the series relating to arrivals and spending from Germany and slightly reduced those from France between the years 2012 and 2015. The series of total tourist arrivals and expenditure in Spain and those from the United Kingdom remain practically unchanged.
Accordingly, France and Germany are used as control countries to create the counterfactual prediction in the methodology of Brodersen et al. (2015). As already mentioned, these two countries follow the United Kingdom in terms of international arrivals. These countries have not experienced the Brexit phenomenon with respect to their issue of tourists to Spain, which makes them ideal as control countries. Accordingly, the break point is established in Observation 54, which is equivalent to June 2016 when the Brexit referendum took place. To soften the variance, the logarithmic transformation has been made of the variables of arrivals and spending.
In line with the results obtained, Figure 1 reflects the impact of Brexit according to the univariate analysis of the series of arrivals and spending of tourists from the United Kingdom through the ARIMA methodology. For each of the series (arrivals and tourist expenditure), an optimal model has been estimated with the data corresponding to the period between January 2012 and June 2016. Based on this optimal model, the corresponding predictions have been made for the rest of the period, which constitute the counterfactual of the non-Brexit scenario.
Tourist Arrivals and Expenditure: ARIMA Models.
Specifically, the graph reflects the producing of the series (solid line), the prediction of their optimal models (point estimate, dashed line), and the confidence intervals of prediction (dotted lines) for the period July 2016-September 2017. The optimal model that fits the arrivals series is an ARIMA (0,1,1)(0,1,0), in other words, a first-order moving average model, with an order of integration in both the seasonal component and the non-seasonal component. The optimal model for the tourist expenditure series is an ARIMA (0,0,0)(0,1,0) with drift.
As we can see in the graph on the left, the effective arrivals are, in general, in line with or below the arrivals predicted by the model. If the accumulated impact is calculated for the whole of the period between July 2016 and September 2017, the effective arrivals would represent a relative fall of 1.07% with respect to those predicted by the model. With respect to expenditure, the graph on the right indicates that effective spending is in line with or above that predicted by the optimal model. Again, calculating in accumulative terms for the whole of the period, the effective expenditure of British tourists in Spain would represent an increase of 0.34% with respect to that predicted by the model. In any event, the effective production of both series is within the prediction intervals of the models, so the above-mentioned impacts would not be significant from a statistical point of view and would indicate the absence, to date, of a negative impact of Brexit on the arrivals and expenditure of British tourists in Spain.
Below, we will estimate the impact of Brexit using the model proposed by Brodersen et al. (2015), using the arrivals and expenditure of tourists from France and Germany as control series. Figure 2 reflects the results obtained in terms of volume of arrivals and Figure 3 reflects the results obtained in terms of tourist spending. The first panel shows the data and a counterfactual prediction for the post-treatment period. The second panel shows the difference between observed data and counterfactual predictions. This is the pointwise causal effect, as estimated by the model. The third panel adds up the pointwise contributions from the second panel, resulting in a plot of the cumulative effect of the intervention (Brodersen et al., 2015).
The Immediate Impact of Brexit on the Number of Tourists From the United Kingdom Visiting Spain.
The Immediate Impact of Brexit on U.K. Tourist Spending in Spain.
In relative terms, arrivals coming from the United Kingdom showed an increase of +1.0%. The 95% interval of this percentage is [–1.0%, +3.0%]. This means that, although the intervention appears to have caused a positive effect, this effect is not statistically significant when considering the entire post-intervention period as a whole.
Individual months within the intervention period may of course still have had a significant effect, as indicated whenever the lower limit of the impact time series (lower plot) was above zero. The apparent effect could be the result of random fluctuations that are unrelated to the intervention. This is often the case when the intervention period is very long and includes a considerable length of time when the effect has already worn off. It can also be the case when the intervention period is too short to distinguish the signal from the noise.
In the case of expenditures, an increase in relative terms of +2.0% during the post-intervention period may be observed and the 95% interval of this percentage is [–0.0%, +3.0%]. Conversely as in the case of arrivals, the probability of obtaining this effect by chance is very small (Bayesian one-sided tail-area probability p = .027). This means the causal effect can be considered statistically significant.
The results obtained show that, considering the whole period that has elapsed between the vote and the conducting of this analysis (between July 2016 and September 2017), Brexit has not produced any initial effect on the arrival of British tourists and a counterintuitive favorable effect on their spending in Spain. This gives cause to believe that, if an impact were to have occurred, it would have done so during the months immediately following the vote (July and August 2016) in relation to the depreciation of the pound. However, when the same analysis is replicated considering the calculation period of the impact as only the months of July and August, or July, August, and September, it may be observed that the impact obtained is not statistically significant. Therefore, it can be concluded that whatever the impact that Brexit has had until the present day, it does not seem to have translated—at least based on the methodology of analysis used in this study—into a reduction of British arrivals or tourist spending in Spain.
The result obtained is somewhat surprising because it is generally expected that consumers will purchase less of a luxury good when either their expected income goes down (see Crouch, 1995; Smeral & Weber, 2000; Smeral & Witt, 1996; Li, Song, & Witt, 2005, for income elasticities of greater than one in several tourism origin markets) or the relative price of a good or service increases (Crouch, 1995; Durbarry & Sinclair, 2003; Patsouratis, Frangouli, & Anastasopoulos, 2005; Witt &Witt, 1995, among others, highlighted the negative price elasticity of tourism demand). In this way, both a reduction in expected future income and a relative increase in the cost of vacationing in Spain appear to be the likely result of Brexit in the short run (relative price increase) and long run (relative price increase and decrease in British income). The result of this study probably reflects the very short time period that has elapsed since the Brexit referendum, making it difficult to draw conclusions. An analysis conducted several years after Brexit is likely to show a significant shift in tourism patterns.
Furthermore, this does not imply that different results could not be obtained with alternative methods of analysis. In addition, it would be appropriate to determine whether there are any differences in terms of the different markets, that is, business, leisure, or other, but unfortunately no disaggregated data are available before October 2015, when the INE took over the management of the survey, so replicating the analysis with the technique used in this article is not possible with the available data at this time. This would only be possible within 2 or 3 years, when the negotiation period is expected to finish and a sufficiently long disaggregated series becomes available. But these impacts will have to be tested in other future studies.
In addition to the above, in view of these results, future lines of research could examine how Brexit affects different destinations—both within and outside the EU, in accordance with their level of dependence on the British market. In this way, it could be determined whether what in principle is or should be a symmetrical crisis, caused by an impact in the market of origin, could lead to asymmetrical effects depending on the differentiating features of each destination.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, or publication of this article.
