This case study explores the use of railway data to estimate tourism flows between Swiss cantons to strengthen the Tourism Satellite Account (TSA). In Switzerland, the total tourism spend obtained from surveys is distributed among cantons primarily using days of stay. However, as Swiss regions are small, many tourists stay in a hotel in one region but consume tourism services in another region on an excursion, and the location of the daytrip within Switzerland is not available from the survey data. Due to these limitations, excursions are estimated by building a model using data from the national rail provider.
Using railway data to estimate tourism flows between Swiss cantons
Abstract
The main purpose of the Tourism Satellite Account (TSA) in Switzerland is to measure the direct economic effects of this cross-industry sector. The Federal Statistical Office publishes the TSA every three years. In the interim, the annual TSA indicators are published and provide information on the most important key figures of tourism in a reduced form, without the level of detail of a full TSA.
Tourism is of great importance to many regions in Switzerland. To additionally reflect the cantonal development of tourism, the Federal Statistical Office, in collaboration with the State Secretariat for Economic Affairs, also published the annual TSA indicators at the cantonal level in a pilot study, consistent with the concepts and definitions of the system of national accounts.
Regional variation in tourism is high in Switzerland. Some regions generate more than 10% of gross value added directly through tourism, while others generate only 1%. To measure these differences more precisely for quite small units of observation and to better inform local stakeholders, the Federal Statistical Office and the State Secretariat for Economic Affairs needed to expand the set of available data by using more novel datasets.
Calculating annual indicators to the TSA in Switzerland – the need for a novel approach
Tourism activity is aggregated to so-called tourism products. These are, for example, hotel accommodation services or food and beverage serving services. For some products, such as hotels, detailed data are available at a cantonal level that allows computing tourism employment and tourism gross-value added bottom-up from primary data.
For all other products, such as restaurants or transport services, there are no special surveys that allow a bottom-up calculation at cantonal level. These are the products concerned in this report. Instead, the results of the "Annual Indicators of the Tourism Satellite Account" are distributed top-down between the cantons. Firstly, national values are distributed from economic sectors (according to the Swiss equivalent to the European NACE classification - NOGA) to products using an allocation matrix. For example, NOGA 551002 – Hotels, inns and guesthouses without restaurants is associated with the product “Hotels”. Then, the results are scaled down by a factor that corrects for the fact that potentially not all the activity by selected NOGA is touristic (for example, some restaurant services are consumed by the resident population). This correction factor partly uses the “days of stay” as an input (see further below). These steps are identical to the procedure to estimate the annual national indicators to the TSA and similar methods are commonly used in other countries.
The results by tourism products are, in a final step, distributed among cantons using various distribution keys. One of the most important distribution keys are the days of stay that estimate the total number of days tourists spend in a location. This quantity should capture the number of days that tourists consume tourism products in a region. They are thus the sum of overnight stays and one-day visitors. One-day visitors are any tourists that visit a region without an overnight stay. For data reasons it will later be useful to differentiate between tourists that make a daytrip from their permanent place of residence and those who make a daytrip while on an overnight trip in a third location. The latter are here defined as excursions.
Calculating the regional variation of tourism in Switzerland thus requires a precise estimate of the days of stay, both to calculate the tourism correction factors and/or as a distribution key between cantons. Statistics such as the statistics on overnight stays in the hotel industry, the statistics on overnight stays in the non-hotel commercial accommodation, and the statistics on the travel behaviour of the Swiss resident population (for day trips, excursions, and overnight stays with relatives and friends) are the main sources that are directly used to calculate tourist days of stay.
Primary data on hotel and non-hotel (overnight) stays are available in Switzerland. However, as Swiss regions are small, many tourists stay in a hotel in one region (often larger, more urban regions with many hotels), yet consume tourism services in another region (for example, small rural regions) on an excursion. Although the level of tourism is already known for the calculations of the national annual indicators to the TSA, the location of the daytrip within Switzerland is not available from the survey data. Due to these data limitations, excursions are estimated by building a model using data from the national rail provider.
The Federal Swiss Railways survey every connection operated by them or require partner train operators or bus services to conduct a survey for them, to collect data on the ticket used and origin as well as destination of the journey. The survey ensures that every connection (for example, the 09.15 train from Geneva to Zurich) is sampled several times a year. This creates a nearly complete coverage of public transport. The few gaps in the data are mainly for small local connections that are within one canton and thus have no effect on estimating the inter-canton flows of tourists.
The Federal Swiss Railways and the Federal Statistical Office had a good working relationship. The data were shared via a secure governmental data sharing platform that is routinely used by the Federal Statistical Office to ensure data protection standards are met. The data were delivered in machine-readable CSV format. Each interview is one observation available at microlevel. The data concerned only the journey, mode and ticket and not the identity of the traveller.
The railway data includes all users of public transport in Switzerland and is used to obtain a three-dimensional matrix of origin, destination and ticket type. In Switzerland, certain tickets are only available to tourist with proof of an address abroad. Further, some tickets are primarily economical for non-tourists, while other tickets are primarily economical for tourists. To build a model of tourism flows within the country for excursionists, two different samples were constructed.
The first sample consists of tickets and passes that were either bought abroad, directly linked to a tourism activity (for example, a joint museum and train ticket) or bought by tourists with an address abroad. This sample is the strictest sample and has the lowest likelihood of including others than international tourists in the sample. However, some international tourists may be missing from this sample (for example, if they buy a single ticket at a machine or counter).
The second, larger sample starts with all data and excludes tickets aimed at commuters and regular trips. This leaves a sample of all touristic and non-touristic trips. From this sample, data based on the household survey data for the Swiss resident population is subtracted. The error of this sample cannot be precisely estimated. Due to sample and stratification errors present in the household sample, the resulting sample may include data from resident tourists and/or exclude data for international tourists.
In practice, both samples as well as the average between the two samples, are used. The difference in the distribution (not in levels) is small. The published data uses the data of the first sample to minimise the type I error (including others than international tourists).
Estimating excursions
The aim is to calculate a canton-by-canton matrix, where the a,b entry is the number of people-days on an overnight stay in canton a who make a day excursion to canton b. This calculation occurs in three steps.
First, the number of excursions that will take place per overnight stay is simulated.
Secondly, for internal quality assurance, the number of excursions that occur within the canton, where the tourists stay in accommodation is estimated.
Finally, a model that allocates these inter-canton excursions to destination cantons is estimated.
The resulting matrix is applied to a days of stay matrix on overnight stays to allocate days of stay from hotel-heavy cantons to smaller cantons. As a final step, one-day visitors from neighbouring countries are added to the matrix.
Simulating excursions per overnight stay
The number of nights spent on (international) overnight trips, the average length as well as the total number of day trips for Swiss residents, including the number of daytrips from their main residence are known from the direct household surveys. It is assumed that a Swiss resident is, per day, equally likely to make a day trip from the residence as they are to be an excursionist from an overnight stay location. This assumption is specifically suited to the Swiss context and can be tested using the Federal Swiss Railways data. The test is if, conditional on the size of the resident population, the purchase of tourist tickets (for example, a combined museum train ticket) is equally likely, which it is.
With this assumption and the household survey data, the number of excursions for the Swiss resident population can be simulated. The additional restriction that an excursion may only happen if the number of nights per trip is greater than one is imposed.
It is then assumed that the probability of a day trip of a Swiss resident on an overnight stay is equal to that of an international resident on an overnight stay. This assumption is not testable. The argument is that tourists in a destination often have similar interests regardless of residence, and that sorting will take place in the choice of destination. However, on average, it is easier for Swiss residents to reach the excursion location from their permanent residence than it would be for a non-Swiss resident. Hence this assumption could underestimate of the number of excursions.
Estimating the proportion of inter cantonal excursions
The proportion of inter cantonal excursions is estimated in two ways. The preferred approach is to split the number of excursions in each location (from the previous step) using the tickets identified in sample one.
A second approach is to use a gravity model to estimate the number of excursions the non-resident overnight visitors make to a different canton. This is technically redundant but makes it possible to check the data for consistency. The simplifying assumption is made that, after controlling for cantons, potential locations of interest for an excursion are uniform in space. It is assumed that all tourism activity takes place in the geographic centre. The distance from the centre to the borders is measured, and the inverse average distance squared is used as a factor in a linear regression model. The intuition is that visitors are more likely, all else equal, to visit locations that are closer than further away. Hence, conditional on the canton, larger, geographically isolated cantons have more day excursions than small cantons.
Estimating the destination regions of excursions
Finally, the data on origin and destination of the various samples is used to create a matrix, where each entry is the likelihood that an excursion happens between two cantons (with an i,i entry accounting for intra-canton excursions). This probability matrix is then applied to the number of nights by international tourists. The multiplication results in adjustments to the days of stay. Specifically, the days of stay in the origin is subtracted by the number of excursions, which is added to the destination canton.
It is possible to argue that the days of stay should only be adjusted by a fraction of the number of excursions, as a part of the day is spent, for example, at the location of the hotel. However, the days of stay count a day-visitor and an overnight visitor equally, so to ensure conformity of concepts the same approach is employed approach here, where the day of stay is completely allocated to the canton where most of the activity takes place.
Estimating cross-border day tourists
One of the limits of the data are cross-border day tourists. For cross-border tourists, evidence from one-off surveys indicate that many use the car, and for them tickets that are aimed at tourists are often less economical than buying at standard ticket at a machine or counter. Therefore, the data on the levels as provided by the Federal Swiss Railways may be insufficient in quality for publication. Furthermore, due to the varied use of private cars, the Federal Swiss Railways data are not seen to provide enough detail to distribute cross-border excursionists between Swiss cantons. This is instead done by using a gravity model, where the distance from the average location in the canton to the nearest border crossing is measured. The inverse square of this distance is the distribution key that allows the allocation of this tourist activity top-down to cantons.
The Federal Swiss Railways data can be used to test this model. The routes where the train is the most efficient choice are investigated. For these routes, the predictions of the model and the distribution of actual routes taken on trains are similar. As expected, this is not true when all cross-border journeys are included.
The Federal Swiss Railways data are solely used to redistribute excursionists between different cantons, as there is doubt if all (and only) non-resident tourists are captured. The level of excursions is therefore estimated only using the survey on the travel behaviour of Swiss households, which provided high quality estimates. However, the data from Federal Swiss Railways can be used as a second estimate to check for plausibility. Countries that do not have detailed travel surveys could wish to use data on railway usage. The difference in estimates in levels is not very large. This is mainly due to the frequent use of public transport by tourists and the varied offers from public transport providers in Switzerland.
The results show that, while ex-ante not obvious, this more precise method of estimation had little effect on the published results. The effects of inbound and outbound excursionists are symmetrical in Switzerland, likely as annual data averages seasonal effects. Although the total number of excursionists modelled and/or estimated is sizable, the more precise treatment only has a small effect on days of stay. More urban, larger cantons have more outbound excursionists but of a moderate size.
Due to this finding, as well as budgetary constraint, this exercise will not be included in the regular calculations in the near future. That said, if the data are available, it is relevant for other countries to also complete this exercise to ensure that the relatively low impact is also the case in their context.
Data on public transport networks can help estimate the distribution of tourists’ excursions within a country. The Swiss case is particularly well suited due to the high usage of public transport. On routes, where public transport is less popular (cross-border daytrips), the quality of the estimates decreases. This new data can also help test models or provide quality assurance when assumptions are made about the behaviour of resident population relative to tourists. The ability to classify the user of a ticket as tourist or non-tourist is very important for the success of using this data. This case study was facilitated by the efficient interaction with the Federal Swiss Railways, who were able to provide the dataset and had a collaborative mindset. Equally, the legal work on the data protection contract facilitated the scientific interaction.
For further information please contact:
Fabian L Schrey, Scientific Associate, Swiss Federal Statistical Office, info.vgr-cn@bfs.admin.ch
This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area.
© OECD 2025
Attribution 4.0 International (CC BY 4.0)
This work is made available under the Creative Commons Attribution 4.0 International licence. By using this work, you accept to be bound by the terms of this licence (https://creativecommons.org/licenses/by/4.0/).
Attribution – you must cite the work.
Translations – you must cite the original work, identify changes to the original and add the following text: In the event of any discrepancy between the original work and the translation, only the text of original work should be considered valid.
Adaptations – you must cite the original work and add the following text: This is an adaptation of an original work by the OECD. The opinions expressed and arguments employed in this adaptation should not be reported as representing the official views of the OECD or of its Member countries.
Third-party material – the licence does not apply to third-party material in the work. If using such material, you are responsible for obtaining permission from the third party and for any claims of infringement.
You must not use the OECD logo, visual identity or cover image without express permission or suggest the OECD endorses your use of the work.
Any dispute arising under this licence shall be settled by arbitration in accordance with the Permanent Court of Arbitration (PCA) Arbitration Rules 2012. The seat of arbitration shall be Paris (France). The number of arbitrators shall be one.
This document was produced with the financial support of the European Union. The views expressed herein can in no way be taken to reflect the official opinion of the European Union.