Persistence of an external shock to domestic tourism demand

ABSTRACT

This study investigates two aspects of how an external shock in the guise of the Covid-19 pandemic affects domestic tourism demand: (1) If the impact varies across regions and over time or (2) whether a permanent change (hysteresis) occurs anywhere. By doing so, a presumptive change in domestic tourism demand during three summers is quantified based on timely official data for all Nordic countries (Denmark, Finland, Iceland, Norway and Sweden) and Estonia, encompassing a total of 76 NUTS3 regions. These regions are divided into five groups from large metropolitan to remote. Tourism demand is approximated by the number of domestic overnight stays in accommodation establishments in the summer months 2016–2022. Dynamic panel data estimations, including household consumption and hotel price index, reveal that all non-metropolitan regions experience a strong increase in domestic tourism flows in the first summer of the pandemic compared with the three years preceding 2020. In contrast, the largest metropolitan areas encounter a substantial decline. The surge in demand for non-metropolitan areas continues in 2021, while the large metropolitan regions return to their pre-pandemic level. After this, demand no longer deviates from its pre-2020 pattern across regions, confirming that the effects are temporary.

KEYWORDS:

• Domestic tourism demand
• hysteresis
• Regional typology
• Panel data
• External shock

1. Introduction

There is an ongoing discussion about the influence and strength of the Covid-19 pandemic on tourism demand (Mensah & Boakye, 2023). Several studies project tourism flows during the height of the pandemic (e.g. Liu et al., 2021; Kourentzes et al., 2021). Others yet, analyse the relationship between the number of Covid-19 cases and revenues of accommodations or dummy variables for the pandemic and find that there are significant and negative relationships (Ghosh, 2022; Yang et al., 2022; 2021 for a review of the literature). Particular attention is also paid to destinations that manage holding back the pandemic-related decline in demand (Duro et al., 2022; Boto-García & Mayor, 2022). The ability to withstand an unexpected shock is commonly referred to as resilience (Calgaro et al., 2014). Studies suggest that destinations with natural attractions, a strong focus on the domestic market, rural areas and regions with successful tourism prior to the pandemic are more resilient and also have a better recovery trajectory (Duro et al., 2022; Boto-García & Mayor, 2022; Falk et al., 2022a, 2022b; Gyimothy et al., 2022).

Hysteresis is a concept that contrasts resilience in that it implies that the system affected by an external shock has a memory that may lead to a persistent change even when its underlying causing factors disappear (Røed, 1997; Martin, 2012). Empirical evidence on how regions across a group of countries are affected in the different phases of the pandemic is scarce, as is the presumptive occurrence of hysteresis. After earlier SARS outbreaks, hysteresis is identified in international tourism demand (Mao et al., 2010).

This study investigates two aspects of how an external shock in the guise of the Covid-19 pandemic affects domestic tourism demand: (1) If the presumptive impact varies across regions and over time or (2) whether a permanent change (hysteresis) occurs in any of the regions. By doing so, a presumptive shift in domestic summer tourism demand during the different phases of the pandemic is quantified based on timely official data for the five Nordic countries (Denmark, Finland, Iceland, Norway and Sweden) and Estonia, encompassing a total of 76 NUTS-3 regions. Latvia and Lithuania also belong to the northern part of the Baltic Sea region, although there is no open information on the number of domestic overnight stays at the regional and monthly level for these countries.

The summer season is chosen because in the earlier stages of the pandemic, this was the time with the lowest spread of the virus and the least restricted domestic mobility (2020 and 2021) but still with some limitations on large gatherings in place including cancellations of public events (OxCGRT, n.d.). Generally, there are also fewer business tourists in the summer, implying that the structure of visitors is relatively homogeneous over time independent of the pandemic-related cancellations of business events. Tourism demand is approximated by the number of domestic overnight stays in hotels and similar accommodations in the months of July and August during the years 2016–2022. Domestic tourism demand is chosen for the analysis, since international travel is almost completely banned in the initial phase of the pandemic and from there on the restrictions vary vastly across countries.

Conceptually, the analysis makes use of a typology of regions developed by the OECD with a five-point falling scale from large metropolitan to non-metropolitan remote areas (Fadic et al., 2019). This division considers both the population density of a region and its connectivity to metropolitan areas and is thus particularly suitable for analyses relating to the effects of the Covid-19 pandemic. Since it is common with elements of persistence in tourism demand (Song et al., 2019), dynamic panel data methods are applied where also the time effects are allowed to vary across the regions.

Literature identifies several features of tourism demand during the different stages of the pandemic. Firstly, domestic tourism is doing better than international, which is not only related to mobility restrictions but also because increasing economic uncertainty tends to keep individuals closer to home (Arbulú et al., 2021; Provenzano & Volo, 2022; Gyimothy et al., 2022; Nguyen et al., 2022). Secondly, there is a move towards rural, sparsely populated and nature areas, while the demand for domestic city visits is below its pre-crisis levels in many European countries (national statistical offices and OECD, 2020).

However, the course of events reveals that rural and peripheral areas are not protected from the virus, as might initially have been believed (Florida et al., 2023). Data from the national public health agencies confirm this (The Directorate of Health and The Department of Civil Protection and Emergency Management, Iceland, Finnish Institute for Health and Welfare, THL, Norwegian Institute of Public Health, FHI, Public Health Agency of Sweden, FHM and Statens Serum Institut, SSI, Denmark).

This study is the first to use a dynamic panel data framework with five different layers of regions (Fadic et al., 2019) and timely comparable data on overnight stays for six countries for the investigation of the presumptive direct or lasting effects on domestic overnight stays following the external shock of the Covid-19 pandemic. Another contribution is the inclusion of regional dummy variables that vary over time with the purpose to account for effects during different stages of the pandemic.

Previous studies mainly examine the determinants of international tourism flows to the Nordic countries. Examples include Aalen et al. (2019) for Norway; Xie and Tveteraas (2020b) for Chinese tourists to Norway; Xie and Tveteraas (2020a) for international and total leisure tourists in Norway; Nordström (2004) for Sweden; Kronenberg et al. (2016) for a Swedish mountain destination and Khalik Salman et al. (2010) for Norway and Sweden as well as Falk and Vieru (2019) concerning international tourism demand to Finland in the winter season. A rare exception to this is the study by Coenen and van Eekeren (2003) who investigate the determinants of demand for domestic tourism by Swedish households.

After this introductory section, the conceptual background and the empirical approach are presented, followed by sections on data, results as well as concluding remarks.

2. Conceptual background

Destination resilience is commonly defined as the response of tourism to a negative shock (Calgaro et al., 2014; Boto-García & Mayor, 2022; Duro et al., 2022). The resilience of domestic tourism demand can vary across regions, with, for example, competitive areas in Spain showing a strong ability to withstand the first year of the pandemic (Boto-García & Mayor, 2022). Duro et al. (2022) document that regional tourism resilience is higher in the northern areas of Spain, specialised in nature tourism. Provinces that are already oriented toward the domestic market and less densely populated areas also perform better, despite increased immobility (Duro et al., 2022). Falk et al. (2022b) show that remote and poorly connected NUTS3 regions in the southernmost and northernmost parts of Europe endure the pandemic better in terms of domestic tourism demand in the first summer of the pandemic.

Medeiros et al. (2022) find that urban regions with a high proportion of business and congress tourists are particularly affected by the pandemic. This is consistent with Florida et al. (2023) who suggest that metropolitan regions are most affected in the initial stages of the Covid-19 pandemic. Wang and Ackerman (2019) conclude that pathogen concerns are stronger than those of other threats from natural disasters, for instance, implying that crowded areas could be perceived negatively during a pandemic. This could change the travel and tourism flows toward more remote and less populated destinations (Zenker & Kock, 2020). Helgadóttir and Dashper (2021) argue that tourists try to escape from densely populated to remote areas during a pandemic. Using municipality data for Bavaria, Schmude et al. (2021) show that domestic tourism flows, and population density are highly correlated. This means that regions that are far away from populated centres and potential sources of infection might be particularly attractive to visitors during times of pandemics if mobility is indeed allowed.

Studies covering the initial phase of the pandemic demonstrate that urbanisation is the factor most strongly correlated with the number of Covid-19 cases and deaths (e.g. Viezzer & Biondi, 2021). Nevertheless, Florida et al. (2023) emphasise that kind of density may be more important than scale, workplace compactness versus residential mass, for instance. These aspects have implications for tourism flows, as it is easier to avoid crowds in remote than in urban areas. The partial closure of cultural attractions in the cities due to the pandemic makes visiting museums, galleries, restaurants and bars indoors less attractive, even if they are open (Sharifi & Khavarian-Garmsir, 2020). Tveteraas and Xie (2021), for instance, find that many Norwegians prefer to holiday in their own country or close to home in July 2020.

In contrast to resilience, hysteresis is a concept that relates to how persistent the effect of a temporary external shock is, that is, if it remains even when the underlying causes disappear (Martin, 2012). Common examples in economics are external disturbances to the financial sector or the labour market (Dixit, 1992; Røed, 1997). In the latter case, for instance, it is shown that the natural rate of unemployment tends to settle on a new level of equilibrium that is higher than before the shock.

By considering a temporary change in the income of foreign visitors as an external shock, Schubert and Brida (2009) demonstrate that this leads to a permanent positive impact on tourism demand indicating the occurrence of hysteresis. In contrast, Romão (2020) mentions that tourism demand recovers fast after the financial and economic crisis, implying an absence of hysteresis. The concept of hysteresis is also applied to the case of the impact of an epidemic outbreak on tourism demand (Mao et al., 2010), where certain countries exhibited a slower and others a faster recovery. Besides this, evidence points to a relatively fast recovery of tourism demand as soon as the main threat is eliminated or under control by for instance vaccination programmes (Choe et al., 2021; Kourentzes et al., 2021). A forecast by Škare et al. (2021) indicates either a long-lasting or more permanent downward change in tourism demand world-wide following the Covid-19 pandemic.

Based on the perspectives of resilience, hysteresis, crowdedness and population density, as highlighted in literature, the first hypothesis is formulated:

H1: Demand for domestic summer tourism changes pattern in the direct wake of the Covid-19-pandemic.

There are indications that the relationship between population density and infection rates is weakening as the pandemic proceeds. Florida et al. (2023) point out that during the consecutive phases of the pandemic, small towns and rural regions are just as affected as large areas. This pattern is valid also for the Nordic countries, where Norway, for instance, reports higher per capita infection rates in remote areas for each additional wave of the pandemic (source: FHI). Thus, domestic tourism demand may exhibit different patterns in the mature stages of the pandemic.

With the availability of vaccines and fewer mobility restrictions, the generally persistent tourism demand may evolve back to its pre-pandemic pattern (Zenker & Kock, 2020). Gyimothy et al. (2022) emphasise that with the resurgence of international travel opportunities, it remains uncertain whether domestic destinations can increase their market shares. This leads to the articulation of the second hypothesis:

H2: The change in patterns of demand for domestic summer tourism is temporary.

The validity of the two hypotheses is assessed by estimating the dynamic panel data model for domestic tourism demand in six North European countries. A crucial aspect of this model is to find an appropriate classification of regions. Bohlin et al. (2022), for instance, consider peripheral tourism areas that are more difficult to reach, which also means that they have problems attracting visitors from larger cities. Another recent regional classification is proposed by Silva et al. (2021) who distinguish between urban (cities), coastal, mountain and nature, rural as well as urban mixed regions (cities in combination with rural or mountain and protected areas). However, this definition is not completely useful when crowding is of particular importance, since it does account for density, connectivity and spatial relations with neighbouring regions. The classification also proves to be too rough for countries with long coastal lines close to mountain areas such as Finland, Norway and Sweden. Instead, this study employs the OECD typology of regions that goes beyond administrative and statistical criteria, implying that it also includes the accessibility to cities, with the purpose to capture functional and geographical aspects (Fadic et al., 2019; Falk et al., 2022b). According to this classification, small regions (TL3) are associated with their level of access to cities or functional urban areas (FUA). A functional urban area consists of densely populated cities and of adjacent local units with high levels of commuting (travel-to-work flows) towards the densely populated cities (Fadic et al., 2019). Access to the non-metropolitan areas is defined as the time it takes to reach the nearest urban area, taking into account not only geographical features but also the state of the physical road infrastructure and its condition.

3. Empirical approach

The theory of tourism demand is used to derive the empirical model (Nordström, 2004; Song et al., 2019; Eugenio-Martin & Campos-Soria, 2011). This model is augmented by time-varying regional dummy variables that allow an investigation of the presumptive direct or persistent impacts of the Covid-19 pandemic on domestic overnight stays. Tourism demand depends on economic (real income and prices) as well as non-economic factors such as previous trips, outbreaks of diseases, natural disasters and geopolitical uncertainties (Song et al., 2019; Ghosh, 2022). There are also indications that the income elasticity differs between domestic and international travel (Eugenio-Martin & Campos-Soria, 2011).

Following the literature, a dynamic specification is employed instead of a static model in order to avoid omitted variable bias due to persistence in tourism demand (Song et al., 2019; Ghosh, 2022). The extent to which the regions are affected is measured by regional dummy variables, classified in accordance with Fadic et al. (2019) and interacted with time. Thus, the linear dynamic panel data model for domestic tourism demand at the regional level is specified as follows: (1) $\begin{array}{rl}\ln \left(\mathrm{DN}{S}_{\mathrm{it}}\right)& ={\alpha }_i+\mathrm{\gamma ln}\left(\mathrm{DN}{S}_{\mathrm{it}-1}\right)+{\text{β}}_1\ln \left(C{E}_{\mathrm{ct}}\right)+{\text{β}}_2\ln \left(\mathrm{CPI}{H}_{\mathrm{ct}}\right)+\sum _{k=2018}^{2022}{\text{β}}_{3k}{d}_k\\ & +\sum _{j=\mathrm{MRL}}^{\mathrm{NMRR}}{\lambda }_{j,20}{d}_{j,20}+\sum _{j=\mathrm{MRL}}^{\mathrm{NMRR}}{\lambda }_{j,21}{d}_{j,21}+\sum _{j=\mathrm{MRL}}^{\mathrm{NMRR}}{\lambda }_{j,22}{d}_{j,22}+{ϵ}_{\mathrm{it}},\end{array}$(1) where $\mathrm{DN}{S}_{\mathrm{it}}$ is the number of domestic overnight stays, ln() represents the natural logarithm, i the region (i = 1, 2, … ,76) and c the country (c = 1, … ,6). The summer holiday period (July and August) is indicated by t (t = 2017, … ,2022), $C{E}_{\mathrm{ct}}$ denotes the national household consumption in constant prices for the second quarter of each year and $\mathrm{CPI}{H}_{\mathrm{ct}}$ is the national price index for hotels and restaurants averaged across July and August. Parameters ${\text{β}}_1$ and ${\text{β}}_2$ are the short-term income and price elasticities of domestic tourism demand. Regional fixed effects are captured by ${\alpha }_i,$ which is part of the residual ${ϵ}_{\mathrm{it}}\left(\equiv {\alpha }_i+u_{\mathrm{it}}\right)$. Time dummy variables, $d_k$, capture time-varying factors of domestic tourism demand that are common across all types of regions until 2019. Interaction terms, $d_{j,t}$, between year dummy variables of 2020–2022 and the regional classification dummy variables, ${\lambda }_{j,k}$ (with j = MRL, MRM, NMRS, and NMRR and k = 2020, 2021, 2022) provide information on the presumptive change in demand patterns across the five groups of regions (as defined in Box 1) during these summer periods in comparison with the development of the reference category NMRM. The development in the reference category is captured by the year dummy variables.

To examine whether domestic tourism demand evolves differently across regions as compared to the overall trend, a Wald test on equality of regional coefficients in each summer season is designed as the following using 2021 as an example: $H_0:{\lambda }_{\mathrm{MRL},21}={\lambda }_{\mathrm{MRM},21}$for the two metropolitan regions and $H_0:{\lambda }_{\mathrm{NMRS},21}={\lambda }_{\mathrm{NMRR},21}$ for the two non-metropolitan regions.

Box 1. Typology of regions.

1. Large metropolitan region (MRL), where more than 50 per cent of its population lives in a functional urban area >1.5M inhabitants,

2. Metropolitan region (MRM), where more than 50 per cent of its population lives in a functional urban area >250K inhabitants,

3. Non-metropolitan region with access to metro (NMRM), where more than 50 per cent of its population has access to a metropolitan area >250K inhabitants within a 60-minute drive,

4. Non-metropolitan region with access to small or medium-sized city (NMRS), where more than 50 per cent of its population has access to a small or medium-sized city (or functional urban area) of 50 to 250K inhabitants within a 60-minute drive and

5. Non-metropolitan remote region (NMRR), where 50 per cent of its population does not have access to any functional urban area within a 60-minute drive.

Source: Fadic et al. (2019).

By using the Maximum Likelihood estimator of the dynamic Fixed Effects panel data model developed by Hsiao et al. (2002), the typical Nickell-bias in panel datasets with more observations than time periods is solved (Nickell, 1981). This estimator is also robust to the cross-sectional heterogeneity. Although the specification is estimated in first differences and the variables are assumed to be stationary, the resulting coefficients are usually interpreted as short-run and long-run (semi-) elasticities.

4. Data sources and stylised facts

Timely official panel data on the number of overnight stays in Denmark, Estonia, Finland, Iceland, Norway and Sweden for the 2016–2022 summer seasons (months of July and August) are utilised for the analysis (National Statistical Offices of Denmark, Estonia, Finland Iceland, Norway, Sweden). While all Nordic countries and Estonia have similar reporting systems for official tourism statistics, this is not the case for the two Baltic states of Lithuania and Latvia, despite otherwise cultural and physical closeness. Because of this, the two latter countries cannot be included in the analysis. There are 76 NUTS 3 regions in the dataset: 5 large metropolitans, 14 metropolitans, 8 non-metropolitans – close to metropolitan, 17 non-metropolitans close to medium-sized city and 32 remote regions (Table A2, Online Appendix). Domestic overnight stay data for Estonia, Finland, Norway and Sweden encompass all accommodation establishments (hotels, holiday villages, youth hostels, camping sites as well as commercially arranged private cottages and apartments) while statistics for Denmark are limited to information on hotels with a minimum of forty beds. Accommodation data for Iceland is also restricted to hotels. These differences will be partly captured by the fixed effects panel data regression. Information on final consumption expenditures of households from the OECD is used as an approximation of household income and refers to the second quarter of each year (OECDSTAT n.d.) These data are deflated by the consumer price index (CPI: 11 – Restaurants and hotels) from the same source and time. Since data on household consumption and consumer prices do not have the same periodicity, a rationale is assumed where the development of consumption expenditures the quarter before the main summer holiday months indicates the economic resources available for the trip, while the price level information is most relevant for the peak season itself.

Descriptive statistics (Table 1) reveal that most domestic overnight stays appear in large rather than in very large metropolitan as well as in remote areas (Figure 1). This is partly explained by the small number of regions with very large cities in the countries analysed, while there are many remote regions in the North. Before 2020, the domestic overnight stays follow an uneven pattern, but the metropolitan areas are all growing substantially. To the contrary, both metropolitan regions experience a decline in overnight stays in 2020, while the non-metropolitan ones all exhibit a slight growth. In the summer of 2021, the total number of domestic overnight stays exceeds that of the years before 2020, but there are still variations across regions and the large metropolitan areas do not clearly supersede their pre-pandemic levels (Figure 1). In the summer of 2022, domestic overnight stays are declining compared with the year before. This can be observed for all region types with the exception of the large metropolitan ones. Consumption expenditures of households in constant prices develop weaker during the time period 2020–2022, as compared with earlier years while hotel prices surge.

Figure 1. Demand for domestic summer tourism (domestic overnight stays). Source: Statistics Denmark, Statistics Estonia, Statistics Finland, Statistics Iceland, Statistics Norway, Statistics Sweden, OECDSTAT and own calculations.

Note: The regions are defined as the following: large metropolitan >1.5M inhabitants (MRL), metropolitan >250 K inhabitants (MRM), close to a metropolitan >250 K inhabitants (NMRM), near a city of 50–250 K inhabitants (NMRS) and remote (NMRR).

Table 1. Descriptive statistics of annual growth rates (per cent).

	2016–2019
	Mean	Median	Std. dev.	Minimum	Maximum	Observations
	Dependent variable overnight stays (dlnDNS)
Total	1.9	1.8	9.6	−24.2	53.7	228
MRL	5.2	4.0	7.6	−7.0	20.0	15
MRM	3.1	2.3	5.3	−7.7	14.5	42
NMRM	0.6	1.8	5.6	−12.6	8.6	24
NMRS	2.3	1.8	13.4	−24.2	52.8	51
NMRR	1.0	0.2	9.5	−20.8	53.7	96
Independent variables
dlnCE	2.0	2.6	1.8	−0.4	5.6	228
dlnCPIH	2.7	2.8	1.4	−2.4	6.6	228
	2019–2022
Dependent variable overnight stays (dlnDNS)
Total	6.0	5.0	29.2	−56.9	191.0	228
MRL	5.9	10.4	28.4	−53.2	45.2	15
MRM	5.1	4.4	23.1	−41.4	48.9	42
NMRM	5.0	8.4	17.6	−32.9	31.9	24
NMRS	10.0	5.3	31.1	−39.6	123.5	51
NMRR	4.6	3.2	33.1	−56.9	191.0	96
Independent variables
dlnCE	1.4	4.6	8.3	−12.8	12.9	228
dlnCPIH	4.0	2.6	4.2	−3.7	18.8	228

Notes: DNS represents domestic overnight stays, CPIH the price index for hotels and restaurants and CE is household expenditures in constant prices. Dln() denotes the logarithmic change in the variable as compared to the previous summer period. The regions are defined as the following: large metropolitan >1.5M inhabitants (MRL), metropolitan >250K inhabitants (MRM), close to a metropolitan >250K inhabitants (NMRM), near a city of 50–250K inhabitants (NMRS) and remote (NMRR).

Source: Statistics Denmark (n.d.), Statistics Estonia (n.d.), Statistics Finland (n.d.), Statistics Iceland (n.d.), Statistics Norway (n.d.), Statistics Sweden (n.d.), OECDSTAT and own calculations.

5. Empirical result

Within the overall increase in demand of 11 per cent in the summer of 2020, calculated as (exp(0.103−1)*100), estimations of the dynamic panel data models also reveal that there are marked variations in how the Covid-19 shock affects different regions (Table 2 and Table A1, Specification i, Online Appendix).¹ Thus, in the first summer of the pandemic, there is a clear downturn for metropolitan regions (MRL and MRM), while all non-metropolitan ones (NMRM, NMRS and NMRR) experience a surge in demand. Already in 2021 the pattern is changed, and demand is growing for all regions. Strongest rises are found for the large metropolitan (MRM) regions and those close to a medium-sized city (NMRS). However, the latter effect is only significant at the ten per cent level. In the third summer of the pandemic, domestic tourism demand does no longer differ from the years before 2020.

Table 2. Determinants of demand for domestic summer tourism 2017–2022 (Dynamic QML-FE estimations).

	Interaction terms
	2020			2021			2022
Region	Coeff.		t-stat	Coeff.		t-stat	Coeff.		t-stat
Large metropolitan region (MRL)	−0.388	***	−3.41	0.092		1.57	0.128		1.60
Metropolitan region (MRM)	−0.203	**	−2.48	0.089	**	2.43	0.013		0.19
Region close to medium-sized city (NMRS)	0.073		0.74	0.092	*	1.70	−0.025		−0.34
Remote region (NMRR)	0.080		0.88	0.013		0.39	−0.092		−1.37
Variable	Coeff.		t-stat
lnDNSt-1	0.848	***	11.93
lnCE	0.297		0.79
lnCPIH	−0.161		−0.46
Year_2018 (reference 2017)	0.014		0.55
Year_2019	0.040	*	1.74
Year_2020	0.164	***	3.21
Year_2021	0.155	***	4.17
Year_2022	−0.023		−0.24
Constant	−0.908		−0.21
Wald test of equality of coefficients	P-value
2020 MRL = MRM	0.005
2021 MRL = MRM	0.614
2022 MRL = MRM	0.515
2020 NMRM = NMRR	0.623
2021 NMRM = NMRR	0.398
2022 NMRM = NMRR	0.394
Number of observations	456
Number of regions	76
Pseudo R2	0.87

Notes: Asterisks ***, ** and * denote significance at the 1, 5 and 10 per cent levels. Robust standard errors are employed in the estimations (reported within brackets) conducted using the Stata command xtdpdqml (Kripfganz, 2016) that implements the QML method of Hsiao et al. (2002). The reference category for the regional dummy variables is close to a metropolitan >250 K inhabitants (NMRM). The Wald test investigates equality of the coefficients for the regional dummy variables over summer seasons. Pseudo R-squared is calculated as the square of the correlation between the values predicted by the model and the actual values.

Source: Statistics Denmark (n.d.), Statistics Estonia (n.d.), Statistics Finland (n.d.), Statistics Iceland (n.d.), Statistics Norway (n.d.), Statistics Sweden (n.d.), OECDSTAT and own calculations.

The results mean that none of the two hypotheses can be rejected (regions affected differently, although the change is temporary), implying that there is no imminent evidence of hysteresis.

According to the Wald test, the magnitude of the significant decrease in the two groups of metropolitan regions differ in the summer of 2020 (Table 2). Large metropolitan regions (MRL) are affected by −20 per cent (calculated as exp(0.164-0.388) −1) and the metropolitan regions (MRM) by 3.8 per cent, compared with the base summer 2017. The three non-metropolitan regions (NMRM, NMRS and NMRR) exhibit no significant deviation in their increases neither in 2020 nor in 2021 (p-value > 0.10 for both summers). Already in the summer of 2021, the demand for overnight stays in large metropolitan regions is not significantly different from the reference category (NMRM) as indicated by the Wald test. For the metropolitan regions (MRM), on the other hand, the highest growth across all can be observed (interaction term of 0.089 and a p-value < 0.05). This means that in the summer of 2021, domestic tourism demand surges in all regions, with smaller variations than in summer 2020. In the summer of 2022, the development of domestic tourism is no longer uneven across regions. The Wald tests for 2022 do not reject the hypothesis of equality of the regional dummy coefficients at conventional levels.

Coefficients for household consumption and hotel prices are not significant at conventional levels implying that the change in domestic tourism demand is independent of the pandemic-induced downturn of the economy and the decline or stagnation of hotel prices. This may coincide with suggestions that domestic tourism demand is less income elastic (Eugenio-Martin & Campos-Soria, 2011). However, given that the variables for household consumption and hotel prices represent a higher aggregation level of data, they should be interpreted with caution.

Another possible explanation is the existence of persistence in domestic tourism demand. With a coefficient for the lagged number of domestic overnight stays at 0.848, the variable harbours a high degree of persistence, that is, it is common to return to domestic summer destinations previously visited even during the pandemic. The adjustment rate of domestic overnight stays per year towards the long-term equilibrium is 15 per cent, stressing a high degree of persistence. The so-called half-life, i.e. the time it takes for the process to close the gap between actual domestic overnight stays and the long-term equilibrium 50 per cent, is 4.7 years (calculated as log(0.5)/log(1-0.848)).

Results of this study are consistent with evidence of asymmetric resilience of domestic tourism demand in a group of countries in Northern and Southern Europe as well as in Spanish regions during the first summer of the pandemic (Duro et al., 2022; Falk et al., 2022b). However, as opposed to Schubert and Brida (2009), who reports persistence of a positive shock to tourism demand and Škare et al. (2021) who forecast long-lasting effects from the Covid-19 pandemic, hysteresis in domestic demand for summer tourism in the countries analysed does not occur. Instead, persistence in earlier behaviour is strong and when the different limitations on mobility are released in summer 2022 demand strives towards its pre-pandemic pattern. This pattern confirms the results of the study by Choe et al. (2021) on an earlier epidemic but partly contradicts Mao et al. (2010).

As a robustness check, the regional dummy variables are interacted also for the pre-pandemic period (time dummy variable for 2019). These results indicate that there is no significant uneven development in domestic tourism demand in 2019 (p-value of Wald test > 0.10; not reported). In a second robustness check the static two-way fixed effects model renders misleading results (Table A1, Specification ii, Online Appendix). The reason for this is the magnitude and the strong significance of the lagged dependent variable.

6. Conclusions

This study investigates two aspects of how the Covid-19 pandemic affects domestic tourism demand: (1) If the presumptive impact varies across regions and over time or (2) whether a permanent change (hysteresis) occurs in any of the regions. By doing so, a presumptive shift in domestic summer tourism demand during the separate phases of the pandemic is quantified based on timely official data for five Nordic countries (Denmark, Finland, Iceland, Norway and Sweden) and Estonia, encompassing a total of 76 NUTS-3 regions. The summer period is chosen because in the earlier phases of the pandemic, this is the time with the smallest spread of the virus and fewer restrictions on mobility. Tourism demand is approximated by the number of domestic overnight stays in hotels and similar accommodations in the months of July and August during the years 2016–2022. A main novelty of the study lies in the investigation of whether the consequences are persistent even after the cause of the shock subsides.

The dynamic panel data estimations show that there is a marked downturn in demand for domestic overnight stays in the large metropolitan regions in the summer of 2020, while the non-metropolitan and remote regions gain. The changes range from −20 to 18 per cent in relation to the reference summer 2017. Regions close to metropolitans are also less directly affected in the initial phase of the pandemic. Already in the second summer of the pandemic the pattern of demand is changed. All regions experience a growth as compared with the pre-pandemic level. In the summer of 2022, when most pandemic-related regulations and limitations are released, domestic tourism demand does no longer exhibit any difference as compared to the pre-pandemic level. This means, that persistence in domestic tourism demand appears stronger than any lasting (hysteresis) effects caused by the pandemic. A major implication of the results is that even a severe shock may not necessarily break a new path in patterns or behaviour. After an initial phase of regulations, limitations and general uncertainty, there is a strong strive towards what was known before. Overall, this shows that demand the previous year is the best predictor of the tourism flows even when price fluctuations and economic variability are considered.

There are several implications for tourism research. Firstly, the use of regional data has many advantages. A relatively large number of groups makes it possible to work with shorter time periods and the significant change in domestic tourism demand can only be identified by use of disaggregated data. The shock of the pandemic also emphasises the importance of regional analyses in general, not only for tourism demand. Another conclusion for research is that dynamic panel data models are clearly superior to static ones when persistence is involved and should be the norm for analyses of regional domestic tourism demand. Internationally harmonised official statistics for monthly tourist arrivals and overnight stays would clearly benefit this kind of studies.

Some limitations of the study need to be acknowledged. Only six North European countries are included in the analysis of which all have a negative tourism balance, that is, more persons are normally going abroad than visiting. The coverage of accommodations in the Danish and Icelandic dataset is more limited than that of the other countries, but the dynamic panel data model is expected to control for such time-invariant effects. Certain non-economic factors related to events or summer weather are not controlled for due to lack of information.

Additional studies could include larger groups of countries and bilateral tourism flows in gravity models. In the latter case, substitution effects are possible to identify. Another idea is to apply the empirical framework to establishment data, separating presumptive differences in occupancy across kinds of accommodations. The tourism and travel micro database could also be used to model the travel decision and control for socio-economic characteristics.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data are available from the authors on request.

Notes

1 Due to the high coefficient of the lagged dependent variable, focus is put on the interpretation of the short-term effects and the coefficients of the year dummy variables are interpreted as semi elasticities.