doi:10.1111/j.1435-5957.2011.00389.x
Destination choice for relocating firms: A discrete choice
model for the St. Gallen region, Switzerland*
Balz R. Bodenmann1, Kay W. Axhausen1
1
IVT, ETH Zürich, Wolfgang-Pauli-Str. 15, CH-8093 Zürich, Switzerland (e-mail: [email protected],
[email protected])
Received: 21 October 2010 / Accepted: 16 June 2011
Abstract. Because of interdependence between resident population (e.g., as customers or
manpower) and economic activities (e.g., shops or services), relocating firms’ destination choice
plays an important role in comprehensive land use models. However, due to a lack of data at the
micro level enabling the estimation of discrete choice models, little research exists on this topic.
In this paper, data from commercial registers of three Swiss cantons, St. Gallen, Appenzell
Ausserrhoden, and Appenzell Innerrhoden, were analysed. This dataset provides information on
more than 54,000 firms and autonomous plants from 1991 to 2006. Variables such as local taxes,
governmental business friendliness, and accessibilities have been tested in a nested logit (NL)
model for different business sectors. The results show that distance, local taxes, and cantonal
business development strongly influence relocating firms’ destination choice. Furthermore,
significant differences between sectors can be identified. For instance, residents with graduate
degrees have a significant and high impact on business services, but a minimal effect on
businesses in wholesale trade and personal services.
JEL classification: D22, R30, R52
Key words: Firm, destination choice, relocation, St. Gallen, Switzerland
1 Introduction
Spatial planning becomes more and more complex as planners face various interrelated problems: traffic congestion near cities, lack of greenfield sites for new plants, competition between
regions, unbalanced residential and economic growth, and so forth. Therefore, simulations of
urban land use, as well as transport and environmental models, aim to assist voters, politicians,
governments, companies and other decision-makers by assessing various alternatives (Geiger
2007). Thanks to the increasing availability of data and computer capabilities, these models have
*The authors would like to thank the commercial register offices of the Cantons of St. Gallen, Appenzell Ausserrhoden and Appenzell Innerrhoden for providing a comprehensive excerpt of their register. Furthermore, we thank three
anonymous referees for their very helpful comments.
© 2011 the author(s). Papers in Regional Science © 2011 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.
Papers in Regional Science, Volume 91 Number 2 June 2012.
320
B.R. Bodenmann, K.W. Axhausen
been enhanced in recent years (Wegener 2005). Sustainable land use and transport systems
depend particularly strongly on an integrated view, taking into consideration interdependencies
between land use patterns and transport. Modelling such complex systems is the strength of
micro-level models such as UrbanSim (Waddell 2010) or ILUMASS (Beckmann et al. 2007). As
models on a micro level require appropriate assumptions and empirical comparisons, scientific
research on the micro level has been intensified recently in various fields for a better understanding of spatial patterns: traffic behaviour, residential location choice, and others.
Despite this increasing use of models on a micro level and despite a growing supply of
literature, quantitative analysis on relocating firms’ destination choice is still relatively scarce,
mostly due to insufficient data on corporate decision-makers: in this case, companies. Availability of retrospective data is very critical; every year, only about 2 per cent of all companies
migrate across municipal borders (Bodenmann and Axhausen 2008). In addition, various
researchers document large differences between companies by sector affiliation, age and size
(among others, Pellenbarg 2005; Maoh and Kanaroglou 2007). Thus, cross-sector surveys
usually provide an insufficient sample size. Larger samples could be derived from registers, but
these are often subject to data protection rules or are not collected at all.
This paper aims to bridge the research gap and to test the impact of different known driving
factors on destination choice for relocating firms and explore the differences between sectors,
using data recorded by the commercial registers of the Swiss cantons of St. Gallen, Appenzell
Ausserrhoden and Appenzell Innerrhoden. These registers contain information about 54,000
firms and autonomous plants between 1991 and 2006. The examination of these registers with
the Business and Trade Register (BUR) of the Federal Office for Statistics (FOS) allows
extraction of different attributes for the incorporated companies: particularly sector, size, age,
and – for each year – the respective location. This information provides statistically meaningful
results through application of various discrete location choice models.
Following this introductory section, Section 2 presents an overview of recent studies in this
field. Based on these references, some hypotheses regarding factors and their impact on the
spatial behaviour of companies will be developed. Sources and content of data are discussed in
Section 3, focusing on main sources for company (as the decision-maker) data, as well as
variables characterizing municipalities as potential site alternatives. Section 4 starts with a short
introduction to and explanation of the estimation procedure and models used. Then, the estimation results of different models are discussed; one model including all observed companies
and six models for subsets of companies in different sectors. The model selected makes it
possible to explore and quantify the impact of many diverse municipality characteristics on
relocating firms’ destination choices. Section 5 draws conclusions on modelling destination
choice for relocating firms and implications of these results on policies and governmental
regulations.
2 Theoretical background
Obviously, firms do not locate or relocate randomly. According to different (economic) criteria,
they follow behavioural goals such as maximizing profits. Scientists have addressed this topic
since the middle of the nineteenth century. An overview of related theories and scientific work
can be found in Maoh (2005), Moeckel (2006), De Bok (2007) and Bodenmann and Axhausen
(2010).
As the models presented are based on individual companies’ data, this section focuses on
investigations at the micro level. As Maoh and Kanaroglou (2007) state, there are few quantitative studies about mobility and location/relocation choice of firms available. In many other
countries, local and regional databases on firm relocation have been collected and analysed.
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
321
Unfortunately, these studies are usually only carried out by local and regional authorities and
their consultancy companies. Examples include business development agencies’ reports (e.g
Winterthur region; SRW 2010); but results of these studies are seldom found in international
journals and often tend to be biased by political goals. The main problem of quantitative analysis
in this field is collecting enough individual companies’ data to generate valid models. In contrast
to macro level research, for example, analyses of an economy’s growth patterns in a certain
region, research on a micro level depends on precise and un-aggregated data. At macro level,
indicators such as number of employees, figures on import and export, or establishments and
closures of companies can be used to model a region’s economy. These indicators are used
according to requirements and data availability. In contrast, modelling choice decisions on a
micro level must be based on information about decisions made by discrete individuals or units
(called decision-makers).
Recent studies concentrate on data from the Netherlands and Canada (city of Hamilton).
Pellenbarg et al. (2002), as well as Maoh and Kanaroglou (2007), give a good overview of recent
developments in these countries. Occasional studies on a micro level have also been performed
for other geographical regionsfor example, the USA (Romo and Schwartz 1995), India, Brazil
and China (Sridhar and Wan 2007), Great Britain (Devereux et al. 2007), and Switzerland
(Baranzini et al. 2006; Bodenmann and Axhausen 2008).
Studies in the Netherlands, Canada, and Switzerland reveal that different groups of companies have distinct location preferences. In the Netherlands, Pellenbarg (2005) and Van Wissen
and Schutjens (2005) examined databases of the Netherlands Chamber of Commerce. Brouwer
(2004), as well as van Dijk and Pellenbarg (2000) worked with data from sample surveys. These
authors’ findings agree that size and sector affiliation of companies have a strong influence on
their migration behaviour; notably, firms that serve larger markets tend to relocate more often.
Based on a dataset from the Statistics Canada Business Register, Maoh and Kanaroglou (2007)
explored business establishment mobility in the city of Hamilton. In their comprehensive
analyses, they explain that age, size and growth influence the behaviour of companies. For
example, they demonstrate that a company’s growth increases the probability of relocation for
manufacturing companies, but for even bigger firms, the probability of moving decreases again
(van Dijk and Pellenbarg 2000; Maoh and Kanaroglou 2007). In Switzerland, Bodenmann and
Axhausen (2008) demonstrate that, apart from age, size, and sector, the present location (type of
municipality) also influences migration behaviour of companies. Young companies relocate
frequently, especially across longer distances. They are quite often also affected by business
closure. Newly arrived companies also relocate more often – and they also often change
municipalities at the same time. The majority of observed migrations occur between larger
cities.
Van Oort et al. (2007) compared different sectors using micro-data of the national employment register LISA (Landelijk Informatiesysteem Arbeidsplaatsen), combining different
regional registers in the Netherlands. They show in a log-linear model that firms affiliated with
the industrial sector are not much affected by labour force accessibility, while business services
choose municipalities with high accessibility to the labour force. The trade and distribution
sector is most affected by the accessibility variable (road network). Public and personal services
tend to relocate to municipalities with an urban character.
Following Bodenmann and Axhausen (2010), firm location choice influencing variables
characterizing potential sites can be assigned to three basic groups: production factors; business
environment; and governmental environment. A further determinant is the distance between
present location and a potential site. Table 1 gives an overview of the corresponding variables
and their expected signs in a discrete choice model.
First and foremost, production factors are generally routine production costs, for example,
energy, infrastructure, human capital and the required workspace. For infrastructure and taxes,
Papers in Regional Science, Volume 91 Number 2 June 2012.
322
B.R. Bodenmann, K.W. Axhausen
Table 1. Overview of hypotheses on location choice of moving firms
Determinants for destination choice
Production factors
Cost of work space
Availability of work space
Availability of human capital
Business environment
Localization effects
Urbanization effects
Governmental environment
Tax burden
Infrastructure and accessibility
Business friendliness / incentives
Geographical environment
Distance to previous site
Expected sign of the effect
Sectors most affected
+
+
Manufacturing
Manufacturing
+
+
Retail and services
+
+
Services and wholesale trade
-
Personal services/gastronomy
there is a natural overlap with governmental environment; these variables will be discussed later.
Based on different studies and surveys, Bodenmann and Axhausen (2010) highlight the following important production factors: cost and availability of workspace and cost and availability of
workforce (by educational level).
Various studies show that cost of land (real estate or rentals) plays a decisive role in a
company’s selection of a location (Bodenmann and Axhausen 2010). However, Louw (1996)
shows that the issue of land price is critical only in the final phase of the decision-making
process, and thus tends to be overestimated. Based on a comparison of various models, however,
Richardson et al. (1974) mentioned that house and land price composition is, in reality, considerably more complex than assumed in the models. Indeed, land prices can be modelled in two
fundamentally different ways: (i) as an exogenous location factor that is predetermined by the
land and real estate market (cf. Vettiger 1994), and (ii) as an endogenous variable based on
location preferences of various space users (in the sense of von Thünen 1826). Land price can
be modelled using other location factors, possibly making the variable redundant or, at least,
leading to smaller effects. De Bok (2007) estimated discrete choice models at address-level and
tested the influence of rents. Interestingly, the (negative) influence of rent level turned out to be
significant only for service sector firms.
Despite a dearth of studies on the influence of workspace availability, it seems to be a
relatively important factor. In a survey by Sedlacek (1994), 88 per cent of all companies said
they were looking for a new location because they lacked development opportunities at their
current location. Based on various surveys, Pellenbarg (2005) stated that this has been the most
important relocation motivation for decades. Large companies seek enough space to satisfy
foreseeable future business expansion. Maoh and Kanaroglou (2007) also explained the factor
behind growth companies’ higher probability of relocation in Hamilton: scarce floor space. In
general, businesses in the manufacturing sector need large areas (Matter et al. 2008). Therefore,
this sector is expected to be over-proportionately sensitive to availability of workspace.
The labour market (or human capital) plays a central role in most models (van Oort et al.
2007). For various kinds of work, it can be assumed that location of private households has an
influence on the company’s site selection. For example, in a long-term study in the USA, Kim
(2005) demonstrated a connection between population density and density of workplaces.
Güssefeldt and Streit (2000) investigated the general influence of agglomeration effects in
NUTS 2 regions of the European Union. This model shows that variation of a region’s economic
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
323
development can be primarily explained by the labour market (characterized by the share of the
population with higher education and the unemployment rate). Based on a census database of
China in 2001, Hong (2007) studied the influence of firm-specific characteristics on location
behaviour of foreign logistic firms in various Chinese cities. His study indicates that the
relevance of market demand, labour factors and governmental policies differ significantly
among diverse groups of firms. For instance, small firms are more responsive to labour cost than
large companies. In contrast to stand-alone companies, subsidiary companies are less sensitive
to local market demand and labour supply.
Regarding business environment, several studies indicate that agglomeration economies are
the heart of urban processes and their role in explaining firm location is fundamental (e.g.,
Rosenthal and Strange 2003; Maoh and Kanaroglou 2007). We refer to two types of agglomeration economies with localization and urbanization effects. Localization economies result
from an agglomeration of same industry firms (industry cluster) and can reduce costs for
supplies, sales, research and development. Urbanization economies emerge from interdependences between firms of different industries. Buenstorf and Guenther (2007) investigated a
dataset covering German machine tool firms from 1936 to 2002. Their results show that
agglomeration effects have a strong impact on location choice for relocating companies. Both
localization and urbanization effects increase the probability that moving firms relocate into a
given region. Rosenthal and Strange (2003) deduce that localization effects are usually more
important than urbanization effects. In addition, Maoh and Kanaroglou (2007) also demonstrate
decentralization and suburbanization effects for all observed industries. Agglomeration effects
generally tend to be more significant for retail and service organizations. Baranzini et al. (2006)
also verified positive localization effects in Switzerland. Devereux et al. (2007) provide evidence that urbanization effects have a small, but significant, positive effect.
Cities and regions, as well as nations, try to attract companies. In general, governments at
different levels have three main strategies to enhance their attractiveness: reducing tax burden,
providing specific infrastructure and facilitating business (Siebert 2000). Over the last few years,
offering incentives and reducing taxes have proved to be popular, but controversial, policies.
Bondonio and Greenbaum (2007) studied the impact of tax incentives on several aspects of local
economic growth in enterprise zone (EZ) programmes1 in different US states. Their results
provide evidence that EZ incentives do have significant impact. In particular, EZ programmes
are attractive for relocating firms and new firm establishments. In regard to infrastructure,
different studies show the positive influence of leisure, school quality (Gatzweiler et al. 1991)
and traffic infrastructure (Hilber 1999). De Bok (2007), among others, tested different accessibility variables in discrete choice models and demonstrated that accessibility has a positive
effect on a location’s attractiveness for service businesses. Another factor, public authorities’
business friendliness, also has an impact that should not be underestimated. Baranzini et al.
(2006) show that business development by Swiss cantons makes them significantly more
attractive to new and migrating companies. However, grants per se do not have a significant
effect on companies’ decisions. Devereux et al. (2007) investigated what determines where
multi-plant firms choose to set up greenfield plants in Great Britain. Their research focused on
governmental subsidies and reveal that companies are less responsive to governmental subsidies
in regions with few other plants in their industry. The importance of governmental subsidies
grows as the number of plants already established increases. They conclude that, depending on
localization effects, grants have only a small positive influence on location choices.
Various authors verify importance of the distance between the previous site and a potential
new site (e.g., Van Steen 2005; Van Wissen and Schutjens 2005; van Oort et al. 2007; Cabus
1
A substantial majority of US states have established economic revitalization programmes in geographically defined
areas (enterprise zones).
Papers in Regional Science, Volume 91 Number 2 June 2012.
324
B.R. Bodenmann, K.W. Axhausen
et al. 2008). Based on information from the Swiss trade register, Bodenmann and Axhausen
(2008) evaluated relocation distances in Switzerland. The plot of these distances strongly
resembles a negative exponential curve. Stam (2003) explains these findings by analysing the
companies’ networks and their employees. As employees, management included, build up
networks around their work and residence, companies can relocate only within certain limits;
otherwise, they incur a risk that employees seek new jobs in their settled surroundings. Personal
services and gastronomy firms particularly tend to relocate within small distances, since their
customers are usually located in a smaller region.
This paper provides answers about where migrating firms relocate and which municipality
characteristics significantly influence these decisions (by sector). Addressing these research
questions, based on the studies discussed above, this paper tests the following hypotheses (see
Table 1):
Hypothesis 1. The attractiveness of a new site for a relocating company is positively influenced
by the availability of work space, manpower and human capital, as well as localization and
urbanization effects, infrastructure and business friendliness of public authorities.
Hypothesis 2. In contrast, increasing work space, cost, tax burden and distances between the
previous site and a potential site have a clearly negative effect.
Hypothesis 3. There are differences in sectors: manufacturing is particularly sensitive to the
availability of favourable land.
Regarding agglomeration effects, Maoh and Kanaroglou (2007), demonstrate that localization positively affects destination choice of businesses in retail and services. Due to large market
areas, wholesale should highly value proximity to national/international road networks, and
therefore highway connections. Different studies show that the service sector is also highly
affected by accessibility (e.g., De Bok 2007; van Oort et al. 2007). Sectors with strong connection to their site, such as personal services and gastronomy, tend to relocate only over very small
distances.
3 Data
3.1 Companies as agents
Using a specifically designed software program, basic information for businesses located in the
cantons of St. Gallen, Appenzell Ausserrhoden, and Appenzell Innerrhoden was extracted from
the corresponding commercial registers for the years 1991–2006. To do this, several business
characteristics were identified for each calendar year at 31 December: namely, the municipality
of residence, the number of registered persons, and how long the company has been in business.
Based on the respective commercial register excerpts, the development of about 54,600 firms
was traced over a period of 16 years, with the number of registered firms increasing from 20,700
at the end of 1990 to 31,600 by the end of 2006. Commercial registers in Switzerland focus on
the constitution and identification of legal entities; they collect and disclose legally relevant facts
and ensure legal certainty under the mandatory rules of civil law. Indeed, the commercial
registers contain not only firms, but also plants with certain autonomy from the parent company.
Even though business demographic events like formations, closures, and migrations are
recorded in the commercial registers, a certain ambiguity reigns. For example, the difference
between genuine establishments and closings and market entries and exits (e.g., based on
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
325
Fig. 1. Number of migrations observed between the case study area and other Swiss districts (1991–2006)
mergers) is quite difficult to identify. Assignments were based on remarks in the registers. Based
on this information, 8 per cent of all registrations and 13 per cent of all de-registrations were
based on mergers and acquisitions. As Hutter (2005) points out, the share of ‘invalid’ formations
and closings could also be significantly larger. In location choice, mergers and acquisitions data
interpretation is also ambiguous, particularly because no conclusions can be drawn as to whether
a company will remain at its existing location. However, on average, only 0.6 per cent of
businesses are potentially affected by this ambiguity per year.
Figure 1 shows the case study area and number of migrations observed between the case
study area and other Swiss districts. Most migrations observed take place within the case study
area; migrations decrease substantially with increasing distance from the original location. In
the following, estimated models include all observed migrations between municipalities –
including migrations in and out of the case study area.
In addition, sector identification (NOGA Codes) and size indication (in size ranges) were
taken from the business and trade register (BUR) of the Federal Office for Statistics. Using
Bürgle’s methodology (2006), 10 business sectors are examined: agriculture and mining; manufacturing; construction; wholesale trade; retail trade; gastronomy and hotels; transport and
communication; service and finance; health and educational service; and remaining business
sectors.
Using data from the BUR, sector affiliation for each company observed can be traced back
without interruption to the year 2003. Since changes in specific sector affiliation are very
unlikely, sector affiliation of the year 2003 has also been used for the period 1991 to 2002. Firms
without information were allocated to individual sectors based on the company’s name and
business goals. Unfortunately, company size is not recorded in the commercial registers; this
information is only available for data made available by BUR for 2003–2006.
Table 2 provides an overview of the cumulated annual number of firms and migrations
observed in the time period 1991–2006. More than one third of all companies observed are in
Papers in Regional Science, Volume 91 Number 2 June 2012.
326
B.R. Bodenmann, K.W. Axhausen
Table 2. Cumulated annual number of firms and migrations observed
Total
Migrations
Migration rate (%)
Sector affiliation
Service and finance
Wholesale trade
Transport and communication
Health and educational service
Manufacturing
Retail trade
Construction
Gastronomy and hotels
Remaining sectors
Agriculture and mining
140,468
52,389
12,061
8,672
59,381
44,293
45,960
23,419
114
5,706
3,168
1,162
202
145
801
574
561
266
1
47
2.26
2.22
1.67
1.67
1.35
1.30
1.22
1.14
0.88
0.82
Legal form of the companies
Holding companies
Joint stock companies
Sole proprietorships, partnerships
Exempted from tax
1,125
240,708
149,436
1,194
57
4,963
1,899
8
5.06
2.06
1.27
0.67
All companies
392,463
6,927
1.77
Note: Data from the commercial register of the cantons of St. Gallen and both Appenzell 1991–2006.
the service and finance sector. Migration in the sectors of agriculture and mining, health and
educational services, and remaining sectors is extremely rare. A company’s legal structure plays
an important role in investigations of tax incentives’ effects. Most companies (61% of all
observed entities) are joint stock companies, paying appropriate taxes for legal persons. The
majority of remaining companies are sole proprietorships and partnerships (38%), and generally
pay the much higher tax rates for individuals. Few companies (less than 3‰) profit from holding
privilege – an extremely low tax rate applicable to companies meeting very specific regulatory
requirements. Evidently, most holding companies are affiliated with the service and finance
sector.
On average, only 1.77 per cent of all companies migrate in any one year, meaning that only
a 1 per cent of the non-movers random sample has been used for the models. The number of
observations thus decreases from 392,463 to 10,728, which also reduces processing time for
calculations from more than one week to less than three hours. For detailed descriptive analyses
of the dataset, the reader is referred to Bodenmann and Axhausen (2008).
3.2 Municipalities as alternatives
Alternatives in the discrete choice models presented are municipalities, characterized by a large
set of variables discussed in Section 2. The alternatives dataset contains only Swiss municipalities that experienced at least one migration between 1991 and 2006 (584 municipalities). Most
of the data used are available for all involved municipalities and the whole period of 16 years.
Two sets of variables are available only for the three Swiss cantons observed: degree of land use
in building zones and duration of the approval process for building licence applications. As this
paper focuses on relocating firms’ destination choices, the proposed variables are limited to pull
factors – push and keep factors are not taken into consideration. Note, however, that logit models
consider only differences between the alternatives, namely, between current and possible others.
Table 3 provides an overview of the variables tested.
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
327
Table 3. Descriptive variables of the municipalities
Factors
Unit
Arithmeticaverage
Standard deviation
Factors of production
Land price for commerce and industryb
Residuals of land price for commerce and industryb
Land price for residential useb
Residuals of land price for residential useb
Degree of land use in building zonesb
[CHF/m2]
[CHF/m2]
[CHF/m2]
[CHF/m2]
[pax/acre]
174.658
11.271
446.338
7.189
0.456
129.759
49.844
209.821
100.342
0.124
Unemploymentb
Population with graduate degreesb
[%]
[%]
0.026
0.055
0.010
0.027
Business environment
Large or intermediate citya
Rate of employees within the same sectorb
Index of diversity of sectorsb
[0,1]
[%]
[sectors]
0.045
0.062
16.908
(0.206)
0.054
3.061
Governmental environment
Tax burden for partnershipsb
Tax burden for joint stock companiesb
Tax burden for holding companiesb
[CHF]
[CHF]
[CHF]
107,722
247,198
1,905
15,453
30,923
1,121
Municipality with a motorway connectiona
Municipality with a rail stationa
[0,1]
[0,1]
Accessibility of employeesb
[]
b
Duration of building licence application process
Cantonal business developmentb
[weeks]
[]
Geographical aspects
Distance to the previous site
[km]
Notes:
a
0.421
0.616
(0.494)
(0.486)
11,929
10,972
4.965
3.172
1.640
0.873
122.228
79.088
dummy variable, b value standardized in the model.
Production factors are modelled by land price for commerce and industry, land price for
residential use, degree of land use in building zones, unemployment rate, and share of economically active population with graduate degrees. Data on average land prices per square-metre for
case study area municipalities since 1990 is based on a 2008 survey of local building authorities.
For municipalities outside this perimeter, data was taken from Immo-Monitoring of Wüest &
Partner (2006). Land prices from both sources are based on recorded transactions as well as
expert estimates. Following the example of Guevara and Ben-Akiva (2009), land prices in the
case study area municipalities have been estimated by multilinear regression, using all variables
of the discrete choice model characterizing municipalities as regressors. In addition to the
original values from Immo-Monitoring, resulting residuals have been added to discrete choice
models as control variables, with the aim of quantifying that part of land price not explained by
discrete choice model variables. Degree of land use in building zones is derived from number of
inhabitants and employees per hectare in designated building zones. As stated, this information
covers only the case study area (status 2007). The labour market is modelled by both the
unemployment rate and the rate of economically active population with graduate degrees
provided by the Swiss Federal Statistical Office (FSO 2008). Since (potential) employees of a
company do not have to reside in the same municipality as the company, these two variables are
calculated as a weighted sum, including all municipalities within a radius of 30 km, employing
a negative exponential function.2
2
The function used is similar to the calculation of accessibility (see e.g. Tschopp 2007; Löchl, 2010): xi = Sj Wj ¥ eb¥d,
with i = municipality of interest; j = municipalities within a certain radius r around i, Wj = number of (relevant) persons
in municipality j, d = distance between i and j [km], b = 0.10, r = 30 km.
Papers in Regional Science, Volume 91 Number 2 June 2012.
328
B.R. Bodenmann, K.W. Axhausen
Business environment indicators include: share of employees within the same sector (localization effects) and index of sector diversity (urbanization effects). The rate of employees within
the same sector is, similar to the two variables representing labour market, calculated as a
weighted sum. Because localization effects are not assumed to be as far-reaching as commuting
distances, the chosen function reflects a much tighter radius around the relevant municipality.3
To represent localization effects, diversity of work places in different sectors on NOGA-code
level 2 was calculated for each municipality (Baumgärtner 2003; Hoffmann 2006). Data on
number of employees and work places by sector comes from the Swiss census of enterprises
(FSO 2008). Since cities play an essential role in firmographics events (Bodenmann and
Axhausen 2008), a dummy variable for large and intermediate cities has been introduced.
Following the typologies of municipalities as set out by the FSO, this involves all agglomeration
centres with at least 45,000 inhabitants (Schuler et al. 2005).
Several variables representing governmental environment have been tested: tax burden for
different legal business structures, infrastructure (i.e., access to motorways and railways),
accessibility of employees, duration of building licence applications approval process, and an
index for cantonal business development. Information about tax liability since 1994 is available
(for an overview, see Swiss Federal Tax Administration 2007). Data on motorways and rail
stations in municipalities – as well as this accessibility variable – were provided by the Institute
for Transport Planning and Systems (IVT) of ETH Zürich (Tschopp 2007; Fröhlich 2008). This
set of variables also includes the accessibility of human resources.4 Overall, this indicator could
also represent the general business environment; duration of the approval process for building
licence application could represent municipalities’ business friendliness. This information is
based on a 2008 survey of local building authorities and is only available for municipalities
within the case study area. In contrast, actual resource allocation (funds and personnel) to
cantonal business development is available for all Swiss cantons, but only on cantonal and not
on municipal level. This data is based on an index developed and surveyed by Credit Suisse
Economic Research (Credit Suisse 2003).
Distances are an important factor in the calculation of several variables, as well as for the
distance between actual and potential sites. For all calculations, shortest road network distances
were used (based on Fröhlich 2008).
4 Model
4.1 Estimation procedure
In these discrete choice models, companies face 120 alternatives each year between 1991 and
2006: all 114 municipalities in the study area plus 6 municipalities randomly selected from the
rest of Switzerland. The set of alternatives reflects the decision-making companies’ point of
view; for each observation, the first alternative represents the company’s current location. This
allows a quasi two-stage hierarchical approach to modelling the decision-making process, as
illustrated in Figure 2. The first stage determines each company’s probability of moving during
the year. The second stage determines the location choice for companies moving. This structure
is modelled in a nested logit (NL) model with two nests (McFadden 1978): Nest 1 ‘Stay’ has one
alternative (the present location) and nest 2 ‘Move’ contains 119 selectable alternatives.
3
The function is the same as explained in footnote 3; with the following modifications: b = 0.50, r = 10 km.
Ai = Sm Sj Oj ¥ eb¥t, with Ai = accessibility in municipality i; j = municipalities in Switzerland and surrounding
Europe, m = means of transport (car, public transport), Oj = number of opportunities in j (in this case employees), i =
travel time (generalized costs) with modus m between i and j [min.], b = 0.20.
4
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
329
Fig. 2. Hierarchical structure of the NL model of firm relocation decision
In this paper, we present only a brief description of the NL model; for further information,
see Train (2003). Discrete choice models belong to the random utility models family (RUM). In
RUMs, a decision-maker n chooses between a set of J alternatives. Choosing an alternative j, the
decision-maker obtains a certain profit called utility Unj. with j = 1, . . . , J. It is assumed that the
decision-maker chooses the alternative providing the greatest utility. However, the utility itself
cannot be measured directly. Instead, based on attributes of the alternatives xnj, as well as
attributes of the decision-maker sn, a representative utility Vnj = V(xnj, sn) is specified. As there are
unobserved aspects of utility, an error term enj is introduced and the utility function decomposed
into Unj = Vnj + enj.
The probability that decision-maker n chooses alternative i out of J alternatives is given by:
Pni = Prob(U ni > U nj ∀j ≠ i )
Pni = Prob(Vni + ε ni > Vnj + ε nj ∀j ≠ i )
The distribution of choice probabilities depends on assumptions for the error terms. As the
NL model belongs to the generalized extreme value (GEV) models, the error term enj is
distributed as a univariate extreme value (McFadden 1978). But – in contrast to multinominal
logit (MNL) models – an NL model assumes that alternatives within the same nest are correlated. Therefore, each nest k is associated to a structural (logsum) parameter lk. This parameter
measures the degree of independence between the unobserved components enj " j of the utility
function for the respective nest’s alternatives.
In a MNL model without nests, the choice probabilities are given by:
Pni =
eVni
∑ j eVnj
In a NL model, we assume the set of alternatives j is partitioned into K not overlapping
subsets called nests B1, B2 . . . BK. For any two alternatives j and m in the same nest Bk, the
unobserved portion of utility enj and enm correlate. Whereas for any two alternatives in different
nests, for example, j ∈ Bk and m ∈ Bᐉ with k ⫽ ᐉ, the error terms are still uncorrelated. The
choice probability for alternative i ∈ Bk is calculated as follows:
(∑
=
∑ (∑
eVni
Pni
λk
j∈Bk
K
=1
j∈B
e
e
Vnj λk
Vnj λ
)
)
λ k −1
λ −1
Since only 2 per cent of all companies relocate to another municipality, the set of nonmoving companies observed has been sampled. Like the nests mentioned above, the total
number of companies in the business register has been divided in two categories: movers and
non-movers. Companies moving have all been included, whereas for the non-movers a randomly
Papers in Regional Science, Volume 91 Number 2 June 2012.
330
B.R. Bodenmann, K.W. Axhausen
Table 4. Overview of the parts of the utility-functions and nests
Alternative
Part of the utility-function for alternative
Parts of the utility-functions
Advantage of not migrating
Advantage of proximity
Location related factors
Nests
Nest of the previous site
Nest of other municipalities
Selection Bias
Block of the previous site
Notes:
a
a and j in α × e
− ϕ × di→ j
{
1
2
3
Constant
–
Set of factorsb
–
Distancea
Set of factorsb
–
Distancea
Set of factorsb
–
Distancea
Set of factorsb
Nest A
–
–
Nest Bc
–
Nest Bc
–
Nest Bc
–
–
Sel. Bias
, distance di,j in km,
b
see Table 3 (Descriptive variables),
120
–
c
fixed.
chosen sample of 1 per cent has been used. However, this approach may cause a selection bias,
which must be taken into account. In the special case of endogenous stratified sampling (ESS),
also called choice-based sampling, Bierlaire et al. (2008) demonstrate that this potential selection bias can be corrected by estimating additional parameters. These parameters play a role
similar to alternative specific constants, and are designed to absorb the potential bias due to
(choice-based) sampling. Indeed, the proposed parameter wi behaves like a shift to the
alternative-specific constant for alternative i.
Table 4 gives an overview of components for calculating the utility-function in this paper.
The representative utility Vj for the alternatives j = 1, 2, . . . , 120 is composed of three parts: (i)
advantage of not migrating to another municipality; (ii) advantage of proximity, measured by
distance from the previous location to the potential location; and (iii) location related factors.
The advantage of not migrating is represented by the alternative-specific constant ASCi for
alternative i = 1 (the present location). In contrast, the advantage of proximity concerns all other
alternatives j = 2 . . . 120. Since the distribution of migrating companies’ distances observed
follows a negative exponential curve, distance di,j between the present location i and alternative
j is correspondingly modelled by a ¥ e-j¥d with two estimated parameters a and j. The location
related factors mentioned in Table 3 influence the utility of all alternatives j = 1, 2 . . . 120. For
each of these factors v = 1, 2, . . . 18 one parameter bk is estimated. The only exception is dummy
variable large or intermediate city. As mentioned in Section 2, the influence of cities is quite
complex and differs between alternative 1 (previous location) and all other alternatives (new
locations). For this reason, two different parameters are estimated: b18,1 for previous locations
and b18,2 for new locations.
As shown in the first part of this section, the NL model differentiates between two nests: nest
A ‘Stay’ consists of alternative 1 and nest B ‘Move’ consists of all other alternatives. In the
model, only the nest A parameter is estimated, as the parameter for nest B has been fixed to 1.
To correct a potential selection bias for alternative 1, the corresponding parameter of Bierlaire
et al. (2008) has been estimated.
All estimates were performed with biogeme 1.9 (Bierlaire 2003, 2010). This freeware
package is specifically designed for research in the context of discrete choice models and –
particularly relevant to this paper – and is able to cope with selection bias. Due to the highly
complex calculations, C code for Feasible Sequential Quadratic Programming (CFSQP) has
been used as optimization algorithm. This is a nonlinear solver developed by Lawrence, Zhou,
and Tits (see Lawrence et al. 1997). For spatial analysis and data preparation, the GIS-Software
GeoMedia Professional 6.1 were used (Intergraph 2010).
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
331
4.2 Results for all companies
Table 5 shows the results of a model including all companies observed. The variables have been
standardized (apart from dummy variables) and are now comparable. In this model, with 6,900
moving and 3,800 non-moving companies representing 385,500 in the specified municipality,
McFadden’s adjusted rho-square is 0.392 (for details, see Bierlaire 2010). This determination
coefficient is biased by the number of companies staying in the same municipality. In a
corresponding model including only companies moving, the adjusted rho-square results in a
much lower value of 0.262. Apart from the constant representing the previous site’s attractiveness, results are comparable to estimates presented in this paper.
As expected, distance plays an essential role. On the one hand, attractiveness of the previous
site is, compared to all other variables, extremely high. As nearly two-thirds of companies
Table 5. Estimation of utility parameters (all sectors)
General information
Number of observations
Null log-likelihood
Final log-likelihood
Adjusted rho-square
Estimated utility parameters
Geographical aspects
Previous site (Constant)b
Distance to the previous site:
Phi in function: alpha ¥ e^(phi ¥ distance)
Alpha in function: alpha ¥ e^(phi ¥ distance)
Factors of production
Land price for commerce and industryf
Residuals of land price for commerce and industryf
Land price for residential usef
Residuals of land price for residential usef
Degree of land use in building zonesf
Rate of unemploymentf
Rate of economically active population with graduate degreef
Business environment
Previous site is in a large or intermediate citye
Alternative is a large or intermediate citye
Rate of employees within the same sectorf
Index of diversity in different sectorsf
Governmental environment
Tax burden for partnershipsf
Tax burden for joint stock companiesf
Tax burden for holding companiesf
Municipality with a motorway connectione
Municipality with a rail statione
Accessibility to employeesf
Duration of the approval process for building licence applicationf
Cantonal business developmentf
Model parameters
Nest 1: previous sitec
Nest 2: remaining alternativesc
Value
10,728
-51,360
-31,238
0.392
Value sign.
4.809
-0.076a
0.606a
0.001
-0.007a
-0.014a
-0.011a
-0.025a
0.012
0.056a
-0.065
0.194a
0.016a
0.060a
-0.041a
-0.106a
-0.287a
0.049a
0.061a
0.033a
-0.021a
0.119a
1.000d
7.520a
Notes: a Significant according to t-test, b Sum of estimated constants and shift, c Biogeme estimates 1/l d Fixed to a
value of 1.00, e Dummy variable f Standardized values.
Papers in Regional Science, Volume 91 Number 2 June 2012.
332
B.R. Bodenmann, K.W. Axhausen
moving, chose a new site within the previous municipality, this holds – even in a model
including only movers. The constant representing the previous site5 depends not only on
characteristics of the companies concerned, but also on exogenous factors. Other models with
the same dataset show that this constant is positively correlated with a municipality or region’s
economic situation: for example, gross domestic product. On the other hand, a site’s attractiveness decreases with increasing distance. Results for the negative exponential function regarding
distances estimate 0.28 utility of a site within a distance of 10 km.6 Hence, in comparison, this
parameter shows a very high value. However, beyond a distance of 60 km, it decreases to an
almost negligible value of 0.01.7 These results, in general, correspond to the findings of Stam
(2003); Van Steen (2005); and Van Wissen and Schutjens (2005).
Analysing production factors, all parameters have the expected sign. Estimated impact of
land price corresponds to De Bok’s results (2007) and other considerations discussed in Section
2: small and (generally) negative. Impact of land price for commerce and industry is near zero
and insignificant. In contrast, residuals, representing the part not explained by the variables in
this model, are significantly negative and large. Therefore, high land prices only have a negative
impact if the price cannot be balanced by other positive impacts for companies (e.g., high
accessibility). Land prices for residential use have a stronger negative impact. This could be
caused by a crowding-out effect; in municipalities with high quality residential land, commercial use tends to be replaced by residential use (in terms of the Thünen rings; von Thünen 1826,
1966). The generally weak effect of land price is also in line with Louw’s findings (1996). The
negative effect of degree of land use shows that availability of work space has a significant
positive impact. Devereux et al. (2007), as well as Maoh and Kanaroglou (2007) observed the
same effects, interpreting their results in the same way.
Scrutinizing business environment variables, localization (sector specialization) and urbanization effects (sector diversity) both show positive effects. Compared to other parameters,
localization effects seem to play a relatively minor role in location choice. Interestingly, these
results contradict those of Rosenthal and Strange (2003); in the models presented, urbanization
effects are generally stronger than localization effects, possibly because Rosenthal and Strange
use a more appropriate disaggregation of sectors. On the other hand, the results reconfirm those
of Devereux et al. (2007). The impact of cities is very strong as well as heterogeneous; cities
tend to have a push effect for companies already located there, but they are very attractive as a
potential alternative. Interestingly pull factors have a much stronger impact than push factors.
The model tests different governmental behaviour variables. As expected, taxes have a
significant negative impact on migrating companies – holding companies and joint stock
companies respond strongly to low tax rates. Tax plays a less important role for partnerships, as
they often revolve around one person, generally living at the company’s location. Another
relatively strong effect can be attributed to cantonal business development. Baranzini et al.
(2006) obtained similar results using the same variable in their location choice model at the
Swiss cantonal level. Long waits for building licence application approvals are significantly
negative, but small. As this variable stands for a municipality’s business unfriendliness, this
result was expected. However, compared to the cantonal business development function, this
variable seems to have a very small impact; possibly, because data on duration of the approval
process for building license application was available only for municipalities in the case study
5
In fact, the constant is the sum of three (partly estimated) parts: the estimated attribute-specific constant ASC1, the
estimator to correct a potential selection bias wstay, and the shift representing the cases sorted out in the sampling process
(Bierlaire et al. 2008). Therefore no t-test has been calculated. Indeed, tests with similar nested logit models showed that
the constant ASC1 is significant.
6
Partial utility regarding distance with estimated parameters (see Table 5) and a distance of 10km (d): Vdist = a ¥
ej¥d = 0.606 ¥ e-0.076¥10 = 0.28.
7
10 km and 60 km represent approximately the arithmetic average of observed moving distances minus and plus half
of the standard deviation respectively (observed average: 36.0 km; standard deviation 46.7 km).
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
333
area. Finally the variables representing infrastructure also have a substantially positive impact
on a municipality’s attractiveness.
Different models have been estimated to compare the robustness of results. They show that
different sampling of alternatives and companies (staying in the same municipality) tend to
influence the level of estimated parameters. But generally, the ranking and the sign of the
parameters remain unchanged. A comparison between the results presented in this paper and
ordinary NL and MNL models reveal only minor differences with other NL models, while
results of a MNL model are only comparable in ranking and sign. This is not surprising, given
the estimated parameter for nest 2 (alternatives to the previous alternative) indicating a high
correlation in this nest.
4.3 Results by sector
Table 6 provides an overview of estimated utility parameters for six different sectors. As the
sector of finance and services includes more than 50 per cent of all companies observed and as
the comparison of different models show that this sector is very heterogeneous, we present the
results for two extremely unequal subgroups of this sector: business services,8 and public and
personal services,9 instead of the whole service sector. Other subgroups – for example the
finance sector– generally get results between these two extremes. Holding companies are all
assigned to the business service sector. Due to better comparison results from the different
sectors, holding companies have been omitted in the following models. Apart from business
services and wholesale trade, the models reach a better adjusted rho-square than a model
including all sectors, as discussed in the section above. However, according to the t-test, several
parameters are not significant. For estimated parameters’ significance, the business services
model provides the best results. As this sector includes still more than 20 per cent of all observed
companies, this is no surprise. However, the adjusted rho-square is relatively low – an indication
that companies in this sector are still very heterogeneous. In contrast, companies in the public
and personal service, as well as gastronomy sectors seem to be quite homogenous, resulting in
a higher adjusted rho-square.
A comparison between the different numerical geographical results is quite difficult. In a
diagram, differences show up more clearly. Figure 3 shows the disadvantage of moving – by
distance – for different sectors, compared to the alternative of staying put. The level of a zero
utility therefore represents the most likely alternative, namely, the previous site (alternative 1).
All other alternatives have a distance-related handicap caused by the effort of moving.10 This
handicap (disadvantage) is represented by negative utility with a variance between 3.70 and
6.90, depending on sector and distance. The utility decreases with increasing distance. Because
the utility function for distance uses a negative exponential weighting, the graph is relatively
steep within small distances and becomes flatter with increasing distances. Differences between
alternatives at distances exceeding 50 km are small.
Two sectors heavily weight the moving disadvantage; public and personal service, as well as
gastronomy and hotels. Companies in these sectors often depend on their location (e.g., a
restaurant at a location with a special view) or serve a small market area (e.g., hairdressers or
8
Business services (NOGA 74) includes professional, scientific and technical services, as well as administrative and
support services.
9
Public and personal services includes waste collection and waste management services (NOGA 90), arts, entertainment and recreation (92), services furnished by membership organizations (91), and other personal services (93)
10
Calculated as the utility of geographical proximity modelled by a ¥ e-j¥d (with d = distance, a and j = estimated
parameters) minus the utility of not moving at all as modelled by the alternative-specific constant for alternative 1
(ASC1): Ud = (a ¥ e-j¥d) - ASC1
Papers in Regional Science, Volume 91 Number 2 June 2012.
Papers in Regional Science, Volume 91 Number 2 June 2012.
Notes:
a
-0.270
0.109b
0.009
0.030b
-0.033b
-0.071b
0.045b
0.048b
0.018b
-0.017b
0.083b
1.000a
13.000
-0.345b
0.126b
0.022b
0.075b
-0.041b
-0.095b
0.025b
0.052b
0.030b
-0.012b
0.136b
1.000a
7.670b
1.000a
14.400b
-0.031b
-0.030b
0.016b
0.032b
0.017b
-0.009b
0.065b
-0.046
0.074b
0.006
0.026b
-0.003
-0.006
0.000
-0.004
-0.003
-0.011
0.028b
4.739
-0.074b
0.331b
Value
1,001
-4,792
-2,741
0.427
Retail trade
1.000a
4.610
0.008
-0.204
0.033
0.179
0.028
-0.041
0.215
-0.218
0.366
0.041
0.045
-0.012
-0.027
-0.003
-0.017
-0.018
-0.149
0.160
5.826
-0.061b
0.987
Value
495
-2,370
-1,313
0.444
Gastron. hotels
1.000a
5.920b
-0.073b
-0.152b
0.064b
0.069b
0.040b
-0.027b
0.157b
0.316b
0.287b
0.024b
0.086b
-0.001
-0.017b
-0.012b
-0.010b
-0.032b
-0.003
0.086b
4.462
-0.068b
0.752b
Value
2,102
-10,063
-6,486
0.355
Businessservices
1.000a
2.160b
-0.077
-0.230
0.149
0.295b
0.171b
-0.070
0.345b
0.118
0.721b
-0.015
0.076
0.034
0.045
-0.076
-0.034
-0.094
0.149
0.066
6.943
-0.099b
2.050b
Value
413
-1,977
-1,042
0.470
Personal services
Fixed value, b Significant according to t-test, c Sum of estimated constants and shift, d Biogeme estimates 1/l. Detailed information for parameters see Table 5.
0.000
-0.003
-0.011b
-0.008b
-0.015b
0.043
0.009b
-0.001
-0.001
-0.015b
-0.013b
-0.022b
-0.009
0.037b
Value
4.288
-0.080b
0.352b
Value
Utility parameters
1,699
-8,134
-5,051
0.378
Wholes. trade
5.099
-0.081b
0.633b
1,390
-6,655
-3,854
0.420
Observations
Null log-likelihood
Final log-likelihood
Adjusted rho-square
Geographical aspects
Prev. site (Const)c
Distance: phi
Distance: alpha
Factors of production
Land price commerce
– residuals
Land price residential
– residuals
Degree of land use
Unemployment
Pop. with degree
Business environment
Prev. site in a city
Alternative is a city
Empl. in same sector
Index of diversity
Governmental environment
Tax partnerships
Tax joint stock comp.
Motorway connection
Rail station
Accessibility
Process build.licence
Business develop.
Model parameters
Nest: previous sited
Nest: remaining alt.d
Manufacturing
General information
Table 6. Estimation of utility parameters by sectors
334
B.R. Bodenmann, K.W. Axhausen
Destination choice for relocating firms
335
Fig. 3. Disadvantage of moving distance by sectors
fitness centres). Bodenmann and Axhausen (2008) show that these companies seldom migrate,
but do have a higher probability of closure. In contrast, companies in the wholesale trade
and business services sector value moving disadvantage moderately. For these companies,
distance also plays a less significant role in their location choice as well, possibly because
companies involved are often small and serve large geographical markets. In general, the
importance of negative distance effects is in line with results of van Oort et al. (2007). As
expected, businesses in personal services and gastronomy are considerably more sensitive to
increasing distances.
For production factors, utility parameters usually show the expected sign. The only exceptions are land price for commerce and industry, and the unemployment rate; but these estimates
could be explained by the need to seek locations with very high accessibility (e.g., in centres).
This assumption is reinforced by the high (positive) impact of a motorway connection, rail
station, accessibility, and location in a city. The unemployment rate was introduced to the model
to mirror availability of short-term manpower. But since unemployment also indicates economic
problems in the respective region, this variable turns out to have a negative effect for most
sectors.
In comparison, companies in the retail sector tend to be highly affected by land prices.
However, the results are not significant, certainly because companies in this sector often migrate
by closing at the previous location and registering again at the prospective site. Interestingly,
companies in the retail sector are only marginally affected by residential use land price; thus,
this sector shows no crowding-out effect (due to a high residential use quality). In line with this
result, this sector is also unaffected by the degree of land use in building zones. In contrast,
manufacturing and business services sectors only respond very significantly to high quality for
residential use. In fact, the expected significant impact of degree of land use on manufacturing
firm’s location decisions did not materialize.
Results for large and intermediate cities’ business environment differ between sectors. For
both service sectors, centrality is outstandingly important. They move far less often out of a city
Papers in Regional Science, Volume 91 Number 2 June 2012.
336
B.R. Bodenmann, K.W. Axhausen
(positive parameter for circumstance that previous site is a city) and often chose new sites in
cities. In contrast, all other sectors tend to leave large and intermediate cities. Confirming
expectations, manufacturing sector companies tend to abandon cities. This is also in line with
Maoh and Kanaroglou’s results (2007), showing similar behavioural patterns in Hamilton. For
public and personal services, this is also in line with the van Oort et al.’s (2007) findings.
However, this study shows a contradictory result for business services, perhaps caused by the
different set of variables. Van Oort et al. (2007) use one variable – population density – to
characterize urbanity. In this study, three variables have been used: two variables representing
cities and land use degree representing land availability. Agglomeration effects are very marked
in manufacturing, gastronomy and hotels, and business services sectors. In contrast, for companies in the public and personal service sector, numerous other companies in the same sector
are a disadvantage. Contrary to the findings of Maoh and Kanaroglou (2007), localization effects
do not have a strongly positive influence on the wholesale trade sector. Urbanization effects are
most highly valued by companies in the business services sector. In fact, urbanization effects
conditioning may not be appropriate. A comparison with Rosenthal and Strange (2003) indicates
that distinctions between sectors have to be more detailed, with evidence that agglomeration
effects should be calculated at walking distance.
In general, estimated governmental environment parameters are comparable to the model
that includes all observed companies. Most sectors appreciate the negative tax burden impact for
joint stock companies more than the impact of tax burden for partnerships. Only retail trade is
slightly affected by tax burden for joint stock companies. To summarize, it can be stated that
sectors with low value creation also tend to be less affected by tax burden for joint stock
companies. Accessibility (motorways and railway stations) is especially important for public
and personal service sector. Indicators for business friendliness of municipalities also have the
expected values and signs. Only retail trade is virtually unaffected by these variables. Van Oort
et al. (2007) also show that generally accessibility variables have an important positive impact.
Their high parameters for manufacturing and business services are broadly in line with the
results presented here. Indeed, the models of van Oort et al. show a quite low influence of
accessibility on personal services. This result may be explained by the very strong urbanity
influence in van Oort et al.’s model.
5 Conclusions
This paper examines municipality characteristics that significantly influence decisions on destination choice for relocating firms and is based on data from commercial registers of three
Swiss cantons, St. Gallen, Appenzell Ausserrhoden, and Appenzell Innerrhoden. The dataset
provides information on more than 54,000 firms and autonomous plants from 1991 to 2006.
Diverse variables, characterizing potential sites as alternatives, were tested in a nested logit (NL)
model. Table 7 shows estimated parameter ranking in different sectors, sorted according to
parameter relevance in a model including companies from all sectors. In general, the most
important factors in companies’ location decisions are: cities, cantonal business development
and tax burden. For potential sites in cities, the study summarizes advantages and disadvantages
of these locations. This result corresponds to the effects predicted by the innovative milieu
approach and other urban agglomeration paradigms (e.g., Hunecke 2003; Florida 2005). The
different valuations of a site in a city suggest that companies in the manufacturing sector of tend
to leave cities. In contrast, companies in retail trade, public and personal services sectors
predominately tend to choose new locations in cities. Tax burden and cantonal business development parameters indicate the very positive effect of governmental business friendliness
although various accessibility indicators also play a strong role in most of the models. In
Papers in Regional Science, Volume 91 Number 2 June 2012.
3a
2a
4a
1a
6a
5a
8a
10a
7a
9a
11a
15a
12a
13a
16
14a
17
18
1a
2a
3a
4
5a
6a
7a
8a
9a
10a
11a
12a
13a
14a
15
16a
17a
18
Alternative is a city
Cantonal business development
Tax burden for joint stock companies
Previous site is in a city
Municipality with a rail station
Index of diversity in sectors of trade
Population with graduate degree
Motorway connection
Tax burden for partnerships
Accessibility to employees
Degree of land use in building zones
Process for building licence
Employees within the same sector
Land price for residential use
Rate of unemployment
Residual land price for residential use
Residual land price for commerce
Land price for commerce
Notes: a Significant parameter according to t-test.
b
Model including all companies observed, apart from holding companies.
Manufacturing
Allb
Parameter
2a
3a
4a
1
5a
9a
14a
6a
8a
10a
12a
11a
14
13a
7
16a
17
18
Wholesale trade
1a
2a
6a
3
4a
8a
7a
10a
5a
9a
16
12a
13
18
11
15
13
16
Retail trade
Table 7. Ranking of utility parameters overall and by sector
1
3
4
2
5
8
6
11
17
12
14
9
9
18
7
15
13
16
Gastro hotels
2a
3a
4a
1a
8a
5a
6a
9a
7a
10a
11a
12a
13a
15a
17
16a
14a
18
Bus.ser-vices
1a
2a
4
8
3a
11
14
6
10
5a
9
13
18
11
6
16
15
16
Pers.ser-vices
Destination choice for relocating firms
337
Papers in Regional Science, Volume 91 Number 2 June 2012.
338
B.R. Bodenmann, K.W. Axhausen
general, a railway station in a potential municipality has a stronger positive impact than a
motorway connection. Interestingly, the various land price parameters have only minor effects.
Land price for residential use is most important, indicating a crowding-out effect between
residential and business use.
In general, these results are in line with the expected results discussed in Section 2. There
are, however, exceptions on the modelled sectors level: land price effects are quite small
(consistent with Louw 1996), and for the manufacturing sector also. Additionally, the manufacturing sector certainly tends to choose municipalities with a lower degree of land use.
However, this does not happen significantly more than in other sectors. Localization effects also
have only moderate effects on retail trade. With expected accessibility results, only public and
personal services sectors show a clear above-average preference for municipalities with high
accessibility and motorway connection as well as railway stations. Not surprisingly, this sector
has a clear tendency to migrate in direction of centres.
These results confirm that governments and politicians have several options for influencing
regional site competition for companies: business friendliness, taxes, and, with a smaller
effect, accessibility. Regarding business friendliness, (cantonal) business development is most
visible for companies and therefore has a large impact. However, Devereux et al. (2007)
showed that grants do not have a strong effect on companies’ decisions. As a consequence,
business development is more successful in supporting companies, for example, during the
process of formation or migration, as well as dissemination of information about potential new
sites. All over the world, low taxes are a common instrument to attract companies and individuals. The models confirm that this is an effective option, although at least in Switzerland
over the last 15 years, taxes in most cantons decreased significantly, making it difficult to
further lower corporate taxes substantially. In contrast to the first two alternatives, accessibility
lost importance in Switzerland over the last decades (Tschopp 2007), probably because differences in accessibility between municipalities and regions diminished considerably in this
period. Therefore, from a governmental point of view, the cost-benefit ratio of such projects
lost attractiveness.
Overall, these results are in line with other similar studies on a micro level. Predictably, this
also holds true for Baranzini et al.’s (2006) study in Switzerland. There are some important
differences, for example, tax burden for partnerships and sole proprietorships turned out to have
a positive effect in Baranzini et al.’s model. One reason for this ‘unintuitive’ result could be the
model structure with cantons as alternatives. Because tax rates for partnerships vary between
municipalities, the models presented here (with municipalities as alternatives) are more complete and lead to more comprehensive results. The Baranzini et al. study is enhanced by
consideration of ‘quality of life’ variables. However, those estimated parameters are not significant for relocating firms. In the Netherlands, van Oort et al. (2007) distinguish models for
different sectors. Their models also argue that cities play an important role for relocating firms.
In contrast to the results presented above, in the Netherlands accessibility variables play the
most important (attractive) role for all sectors. In Switzerland, this holds true only for the sector
of public and personal services. A reason for this difference could be the introduction of a
variable representing cities.
A comparison of different models like Baranzini et al. (2006), van Oort et al. (2007), etc.,
confirm that the variable set characterizing municipalities (as alternatives) has an important
influence on estimated parameters. Comparing the effect of individual variables is considerably
more difficult. As far as employee modelling, we must also caution that the results in this paper
apply only to firms and autonomous plants. Non-autonomous plants – and their employees – are
often not mentioned in commercial registers. This paper focuses on the destination choice of
relocating firms, but most tested variables cover only pull factors. A model covering the whole
migration process – including the decision whether to move or not – must include push as well
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
339
as keep factors. This is a very interesting wider project that emerged as a consequence of our
research.
References
Baranzini A, Ramirez J, Ugarte Romero C (2006) Les déterminants du choix de (dé-) localisation des entreprises en
suisse. Cahiers de recherche, HES-SO/HEG-GE/C-–06/11/1–CH, Centre de Recherche Appliquée en Gestion
(CRAG), Haute école de gestion de Genève, Genève
Baumgärtner S (2003) Warum Messung und Bewertung biologischer Vielfalt nicht unabhängig voneinander möglich
sind. In: Weimann J, Hoffmann A, Hoffmann S (eds) Messung und ökonomische Bewertung von Biodiversität:
Mission impossible? Metropolis-Verlag, Marburg
Beckmann K, Brüggemann U, Gräfe J, Huber F, Meiners H Mieth P, et al. (2007) ILUMASS integrated land-use
modelling and transportation system simulation: Endbericht, Spiekermann & Wegener Stadt- und Regionalforschung, Zentrum für angewandte Informatik der Universität zu Köln, Institut für Stadtbauwesen und Stadtverkehr
RWTH Aachen, Otto-Friedrich-Universität Bamberg Institut für Theoretische Psychologie, Deutsches Zentrum für
Luft- u. Raumfahrt e.V. in der Helmholtz-Gemeinschaft Institut für Verkehrsforschung, LUIS – Lehr- und Forschungsgebiet Umweltverträgliche Infrastrukturplanung, Dortmund
Bierlaire M (2003) BIOGEME: A free package for the estimation of discrete choice models. Paper presented at the 3rd
Swiss Transportation Research Conference, Ascona, Switzerland, March 2003
Bierlaire M (2010) Estimation of discrete choice models with BIOGEME 1.9. URL: www.biogeme.epfl.ch
Bierlaire M, Bolduc D, McFadden D (2008) The estimation of generalized extreme value models from choice-based
samples. Transportation Research Part B, Methodological 42: 381–394
Bodenmann BR, Axhausen KW (2008) Schweizer Unternehmen: Quo vaditis? Firmendemographische Trends am
Beispiel des Wirtschaftsraums St. Gallen. Raumforschung und Raumordnung 66: 318–332
Bodenmann BR, Axhausen KW (2010) Synthesis report on the state of the art on firmographics. SustainCity Working
Paper, 2.3, IVT, ETH Zürich
Bondonio D, Greenbaum RT (2007) Do local tax incentives affect economic growth? What mean impacts miss in the
analysis of enterprise zone policies. Regional Science and Urban Economics 37: 121–136
Brouwer AE (2004) The inert firm: Why old firms show a stickiness to their location. Paper presented at the 44th ERSA
Congress, Porto, August 2004
Buenstorf G, Guenther C (2007) No place like home? Location choice and firm survival after forced relocation in the
German machine tool industry. Jena Economic Research Paper 2007–053, Max Planck Institute of Economics, Jena
Bürgle, M (2006) Modelle der Standortwahl für Arbeitsplätze im Grossraum Zürich zur Verwendung in UrbanSim.
Arbeitsberichte Polyprojekt Zukunft urbane Kulturlandschaften, 8, NSL, ETH Zürich, Zürich
Cabus P, Horemans E, Vanhaverbeke W (2008) Vestigingsgedrag van bedrijven in Vlaanderen: Een analyse in functie
van het ruimtelijk economisch beleid. STOIO Studies 2008 (4), Steunpunt voor Ondernemen en Internationaal
Ondernemen, Leuven/Gent
Credit Suisse (2003) Kantonale Wirtschaftsförderung: Ein erster Vergleich, Spotlight 5 May. Credit Suisse Economic
Research, Zürich
De Bok M (2007) Infrastructure and firm dynamics. Dissertation, TRAIL Research School, Delft
Devereux MP, Griffith R, Simpson H (2007) Firm location decisions, regional grants and agglomeration externalities.
Journal of Public Economics 91: 413–435
Florida R (2005) Cities and the creative class. Routledge, New York
Fröhlich P (2008) Änderungen der Intensitäten im Arbeitspendlerverkehr von 1970 bis 2000. Dissertation, ETH Zürich
Gatzweiler HP, Irmen E, Janich H (1991) Regionale infrastrukturausstattung. Forschungen zur Raumentwicklung 20,
Bundesforschungsanstalt für Landeskunde und Raumordnung (BFLR), Bonn.
Geiger M (2007) Ein Simulator für die Raumplanung. Tec21 125: 24–29
Guevara CA, Ben-Akiva M (2009) Addressing endogeneity in discrete choice models: Assessing control-function and
latent-variable methods. Working Paper Series, TSI-SOTUR–09–03, MIT Portugal, Transportation Systems, Porto
Salvo
Güssefeldt J, Streit C (2000) Disparitäten regionalwirtschaftlicher Entwicklung in der EU. CeGE-Discussion Paper, 5,
Centrum für Globalisierung und Europäisierung der Wirtschaft, Georg-August-Universität, Göttingen
Hilber C (1999) Standortattraktivität, Liegenschaftspreise und Mieten, WWZ-Forschungsbericht 1999 (4),
Wirtschaftswissenschaftliches Zentrum WWZ, Universität Basel
Hoffmann S (2006) Biodiversität: Ein wissenschaftliches Konstrukt auf politischem Prüfstand. FEMM Working Paper
Series, 06029, Faculty of Economics and Management Magdeburg
Hong J (2007) Firm-specific effects on location decisions of foreign direct investment in China’s logistics industry.
Regional Studies 41: 673–683
Papers in Regional Science, Volume 91 Number 2 June 2012.
340
B.R. Bodenmann, K.W. Axhausen
Hunecke H (2003) Produktionsfaktor Wissen: Untersuchung des Zusammenhangs Wischen Wissen und Standort von
Unternehmen. Aachener Reihe Mensch und Technik, 45, Wissenschaftsverlag Mainz, Aachen
Hutter T (2005) Neugründungen und Überlebensraten von Unternehmen: Der Kanton St. Gallen im inter- und intrakantonalen Vergleich 1996 bis 2002. Statistik aktuell, 7, Fachstelle für Statistik Kanton St.Gallen, St. Gallen
Intergraph (2010) GeoMedia professional user guide. Intergraph Corporation, Huntsville, AL
Kim S (2005) The rise and decline of US urban densities. Washington University, St. Louis, URL: http://soks.wustl.edu/
papers.html
Lawrence CT, Zhou JL, Tits AL (1997) User’s guide for CFSQP version 2.5: A C code for solving (large scale)
constrained nonlinear (minimax) optimization problems, generating iterates satisfying all inequality constraints.
Technical Report, TR-94-16r1, Institute for Systems Research, University of Maryland, College Park
Löchl M (2010) Application of spatial analysis methods for understanding geographic variation of prices, demand and
market success. Dissertation. ETH Zurich, Zurich
Louw E (1996) Kantoorgebouw en vestigingsplaats: Een geografisch onderzoek naar de rol van huisvesting bij
locatiebeslissingen van kantoorhoudende organisaties. Dissertation, Delftse Universitaire Pers, Delft
Maoh HF (2005) Modeling firm demography in urban areas with an application to Hamilton, Ontario: Towards an
agent-based microsimulation model. Ph.D. dissertation, McMaster University
Maoh HF, Kanaroglou PS (2007) Business establishment mobility behavior in urban areas: A microanalytical model for
the city of Hamilton in Ontario, Canada. Journal of Geographical Systems 9: 229–252
Matter D, Fahrländer S, Fuchs S, Heye C, Unternährer T, Weilenmann B (2008) Bauzonen Schweiz: Wie viele Bauzonen
braucht die Schweiz? Bundesamt für Raumentwicklung ARE, Ittigen
McFadden D (1978) Modelling the choice of residential location. In: Karlquist A, Lundqvist L, Snickars F, Weibull J
(eds) Spatial interaction theory and residential location. North-Holland, Amsterdam
Moeckel R (2006) Business location decisions and urban sprawl: A microsimulation of business relocation and
firmography. Dissertation, Universität Dortmund
Pellenbarg PH (2005) Firm migration in the Netherlands. Paper presented at the 45th ERSA Congress, Amsterdam,
August
Pellenbarg PH, van Wissen LJG, van Dijk J (2002) Firm migration. In: McCann P (eds) Indusrial Location Economics.
Edward Elgar Publishing, Cheltenham
Richardson HW, Vipond J, Furbey RA (1974) Determinants of urban house prices. Urban Studies 11: 189–199
Romo FP, Schwartz M (1995) The structural embeddedness of business decisions: The migration of manufacturing
plants in New York State 1960–1985. American Sociological Review 60: 874–907
Rosenthal SS, Strange WC (2003) Geography, industrial organization and agglomeration. Review of Economics and
Statistics 85: 377–393
Schuler M, Dessemontet P, Joye D (2005) Die Raumgliederungen der Schweiz. Bundesamt für Statistik (BFS),
Neuenburg
Sedlacek P (1994) Wirtschaftsgeographie: Eine Einführung. Wissenschaftliche Buchgesellschaft, Darmstadt
Siebert H (2000) The paradigm of locational competition. Kieler Diskussionsbeiträge, 367, Institut für Weltwirtschaft,
Kiel
Sridhar KS, Wan G (2007) Firm location choice in cities: Evidence from China, India, and Brasil. China Economic
Review 21: 113–122
Stam E (2003) Why butterflies don’t leave: Locational evolution of evolving enterprises. Faculty of Spatial Science,
Utrecht.
Standortförderung Region Winterthur (SRW) (2010) Jahresbericht 2009. Standortförderung Region Winterthur,
Winterthur.
Swiss Federal Statistical Office (2008) STATINF: Statistical information system. URL: www.statweb.admin.ch
Swiss Federal Tax Administration (2007) Steuerinformationen. URL: http://www.estv.admin.ch
Train KE (2003) Discrete choice methods with simulation. Cambridge University Press, Cambridge
Tschopp M (2007) Verkehrsinfrastruktur und räunliche Entwicklung in der Schweiz 1950–2000. Dissertation, Universität Zürich, Zürich
Van Dijk J, Pellenbarg PH (2000) Firm relocation decisions in the Netherlands: An ordered logit approach. Papers in
Regional Science 79: 191–219
Van Oort F, Ponds R, van Vliet J, van Amsterdam H, Declerck S, Knoben J, et al. (2007) Achtergronden: Verhuizing van
bedrijven en groei van werkgelegenheid. Nai Uitgevers, Rotterdam
Van Steen PJM (2005) Bedrijvendynamiek onder het vergrootglas. In: Pellenbarg PH, van Steen PJM, van Wissen LJG
(eds) Ruimtelijke aspecten van de bedrijvendynamiek in Nederland. Van Gorcum, Assen
van Wissen L, Schutjens V (2005) Geographical scale and the role of firm migration in spatial economic dynamics.
Paper presented at the 45th ERSA Congress, Amsterdam, August
Vettiger H (1994) Lebensverhältnisse und Lebensqualität in städtischen Räumen der Schweiz: Möglichkeiten der
Anwendung objektiver und subjektiver Raumbeobachtungsindikatoren. Dissertation, Hochschule St.Gallen
Papers in Regional Science, Volume 91 Number 2 June 2012.
Destination choice for relocating firms
341
Von Thünen JH (1826) The isolated state in relation to agriculture and economics, or studies on the influence exerted
by grain prices, the wealth of the land and the taxes on agriculture. Perthes, Hamburg
Von Thünen JH (1966) Von Thunen’s isolated state. Hall P (ed), translated by Wartenberg CM. Pergamon, Oxford
Waddell P (2010) UrbanSim. URL: http://www.urbansim.org
Wegener M (2005) Integrated land-use transport modelling: Progress around the globe. Paper presented at the 4th
Oregon Symposium on Integrated Land-Use Transport Models, Portland OR, November
Wüest and Partner (2006) Facts & figures: Herbst, Immo-Monitoring. Verlag Wüest and Partner, Zürich
Papers in Regional Science, Volume 91 Number 2 June 2012.
doi:10.1111/j.1435-5957.2012.00389.x
Debido a la interdependencia entre la población residente (p.ej. como clientes o mano de obra)
y las actividades económicas (p.ej. comercios o servicios), la elección del destino en el traslado
de una empresa juega un papel importante en modelos exhaustivos de uso del suelo. Sin
embargo, debido a una falta de datos a nivel microeconómico que permitiesen estimar modelos
de elección discreta, apenas existen estudios sobre este tema. En este artículo se han analizado
datos de registros comerciales de tres cantones suizos: San Galo (St. Gallen), Appenzell Rodas
Exteriores (Ausserrhoden), y Appenzell Rodas Interiores (Innerrhoden). Este conjunto de datos
proporciona información acerca de más de 54.000 empresas y factorías autónomas desde 1991
a 2006. Se ha empleado un modelo logit anidado (nested logit, NL) para diferentes sectores
empresariales, utilizando variables tales como impuestos locales, grado de benevolencia gubernamental con las empresas, y accesibilidades. Los resultados muestran que la distancia, los
impuestos locales, y el desarrollo empresarial cantonal influyen poderosamente en la elección
del destino en el traslado de una empresa. Adicionalmente, es posible identificar diferencias
significativas entre sectores. Por ejemplo, los residentes con licenciaturas tienen un elevado y
significativo impacto en los servicios empresariales, pero un efecto mínimo en empresas de
comercio al por mayor y de servicios a individuos.
© 2012 the author(s). Papers in Regional Science © 2012 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.
Papers in Regional Science, Volume 91 Number 2 June 2012.