Effects of business improvement districts on firm performance, property values and urban safety: an empirical study of five small to medium-sized Swedish towns

ABSTRACT Business improvement districts (BIDs) have emerged as possible solutions for the revitalisation of declining urban areas. We investigate the effects of a voluntary Swedish BID programme on firm performance, urban safety and property values in five small to medium-sized Swedish towns, and find that it increased labour productivity for incumbent firms within the BID by about 7%, mainly through an increase in revenues. However, the positive effect of the BID programme on firm performance is transitory, decreasing sharply during the third year after implementation and then becoming insignificant. As for urban safety and property values, the results are mostly insignificant.


INTRODUCTION
Many city centres are struggling with long periods of economic decline.Urban sprawl and the decentralisation of retail to the urban periphery are often discussed as particularly decisive for this development, resulting in the migration of customers, retail establishments and other businesses from downtowns to outlying suburban areas (Peel et al., 2009).Moreover, many local governments have not been able to meet the challenges facing their cities, providing insufficient basic services such as street cleaning and sanitation, maintenance for public spaces, and safety (Hoyt & Gopal-Agge, 2007;Jansen, 2017).To avoid extensive downturns, property and business owners have been compelled to provide these services themselves (Hogg et al., 2003).
One initiative that has received widespread attention is the implementation of business improvement districts (BIDs), which are privately managed and publicly sanctioned organisations that supplement public services in city centres or derelict urban neighbourhoods.In the Swedish context there are two types of BIDs.The first has its background in town centre management (TCM), and is motivated by city retailers having to compete with external shopping centres.The other is neighbourhood business improvement districts (NBIDs) formed by property owners in collaboration with policymakers and local authorities to improve conditions in derelict residential areas (Stalevska & Kusevski, 2018).The introduction of NBIDs in larger Swedish cities has been investigated in several previous studies (e.g., Bohman & Jingryd, 2015;Hertting et al., 2022;Kronkvist & Ivert, 2020;Kusevski et al., 2023;Stalevska & Kusevski, 2018;Valli & Hammami, 2020), while the introduction of BIDs in small to midsized Swedish towns has not.
BIDs are often grounded in the following three elements: (1) a well-defined geographical area; (2) a public-private partnership between local property and business owners, on the one hand, and the local government, on the other; and (3) the assessment of a compulsory tax on local property and business owners, which constitutes the base for the funding of place-based services and is regulated by law (Caruso & Weber, 2006;Hoyt & Gopal-Agge, 2007).
BID programmes have been argued to be beneficial for the perceived attractiveness of the designated areas (Sutton, 2014), and a plethora of studies have taken up the task of investigating the effectiveness of BID programmes.Few of these studies, however, have attempted to investigate the effects of BID programmes on firm performance (e.g., Sutton, 2014).Interest has instead been directed towards measuring the impact of BIDs on crime levels (e.g., Clutter et al., 2019;MacDonald et al., 2010MacDonald et al., , 2013;;Mello, 2018) or residential and commercial property values (e.g., Brooks & Strange, 2011;Ellen et al., 2007;Hirao, 2021;Miller, 2013).
We contribute to the literature by analysing the impact of a BID programme implemented in 2015 in five Swedish small to medium-sized towns on firm performance, urban safety and property values.In contrast to most other countries, where tax contributions from business and property owners are compulsory for BID programmes, contributions in Sweden are voluntary and the implementation process is not regulated by law.This study is thus the first attempt to evaluate the effects of BID programmes that are voluntary, thereby providing evidence about whether voluntary BID programmes are sufficient for improving firm performance, reducing crime and increasing property values.
We find that the implementation of the voluntary BID programme in Sweden increased the labour productivity of incumbent firms located within the BID by, on average, 7.62%.This effect is mostly due to an increase in firm revenues (11.65%), while we observe small changes in the number of employees.The estimated effect of the BID programme on revenues is positive and highly significant shortly after the implementation of the programme (12.40% in 2015 and 13.74% in 2016), but decreases to 5.92% in 2017 (significant at the 10% level) and to zero in 2018.The effect of the BID programme on the number of employees is statistically significant only in the first year after the BID programme was implemented.We find no effects on firm performance outside the geographical boundaries of the BIDs.Hence, the voluntary BID programme was able to improve the performance of firms located within the BIDs, but the effects were only transitory.
Furthermore, we find few significant differences in crime levels between the BIDs and the control districts.The difference is, on average, negative for all three years (2016-18), but statistically significant only in 2018 with 4.94% fewer crimes reported in the BIDs compared with in the control areas.Crime levels are also higher just outside the BIDs than in the control towns for all three years, with a statistically significant difference in the year after the BID programme was implemented (+3.43%).This could then be seen as an indication that improvements in one area increase crime in adjacent areas, a result previously reported by Porreca (2023) when studying the gentrification of neighbourhoods in Philadelphia during the period 2011-20.However, since this part of our study only has data from the period after the introduction of the BIDs, the results should be interpreted with caution, and more research is warranted.Finally, we find no statistically significant effects of the BID programme on property values within or outside the designated BIDs.

THE HISTORY AND ROLE OF BIDs IN SWEDEN
BIDs originated in North America as a reaction to the trend toward retail decentralisation during the 1970s-80s (Stalevska & Kusevski, 2018).In these BIDs, retailers and real estate owners introduced cooperative efforts to revitalise often rundown city centres by providing service and improvements to streets and parks beyond what the local government could provide, financed by a compulsory tax on businesses within the BID areas.The reasons for implementing BIDs were often attributed to failures of past policies and practices of the state (Ward, 2007).
As in the United States, many European city centres faced competition from external shopping centres.However, the response was not to create BIDs, but rather to create TCM schemes based on voluntary membership and financing (Stalevska & Kusevski, 2018).These TCMs were in most cases formed by local retailers and other businesses in collaboration with the municipalities, through either projects or a specific limited liability company set up to run the TCM scheme (Edlund & Westin, 2009).The efforts to increase city attractiveness include street-level cleaning and maintenance, the organisation of events and other promotional activities, using often rather small budgets (Stalevska & Kusevski, 2018).
BIDs that were more like their North American counterparts were first introduced in Europe in the beginning of this century.In Sweden, interest and knowledge regarding BIDs spread during this period through Swedish politicians and other stakeholders visiting well-established BIDs in other countries, through renowned international speakers promoting BIDs at Swedish conferences, and through practitioners and politicians attending international conferences with a focus on BID-related topics (Cook & Ward, 2012).
A Swedish research project called 'The Good City' was implemented between 2005 and 2010.Edlund and Westin (2009) drew attention to the possibility of creating a Swedish BID model when evaluating the project.They organised a well-attended conference in Stockholm in October 2010 on the topic of a Swedish BID model, with international and Swedish experts.The audience consisted mainly of public and private officials, including city and transport planners, consultants, and architects, working at the local, regional or national levels with TCM and related issues (Cook & Ward, 2012).
The Swedish Association of Towns and Cities 1 decided to design a Swedish BID model after the conference.In 2015, their BID model was implemented in five Swedish small to midsized towns, which also are the ones included in this study. 2This BID programme is intended to increase the attractiveness of the city centre, and is implemented through a 18-month process, involving taking the BID area through seven steps (introduction; workshop; analysis, case study and study visits; business development plan; coaching; results; and conclusions) related to five focus areas (brand, supply, place, accessibility and safety).
Compared with a traditional Swedish TCM scheme, the bid scheme involves a wider set of actors, has a more structured and focused process of implementation, and in most cases a larger budget.However, both membership and financing are voluntary.Compared with the BIDs discussed by Ward (2007, p. 667), which were instigated due to the 'failures of the past policies and practices of the state', these BID schemes are motivated by city centre retailers fearing competition from external shopping centres, something that is also clear from the business development plans made by the towns.
The Swedish Association of Towns and Cities BID scheme includes measures to augment the safety and aesthetics of retail areas (e.g., improvements in the physical form of retail areas), create new meeting places (e.g., parks or playgrounds), support retail-related events (e.g., markets and showrooms), and widen the range of services provided in these districts (e.g., provide free Wi-Fi in public spaces).
The BID programme offer strategies for improving the attractiveness of the designated areas in the hope of increasing their customer range and thus, eventually, their customer base (Sutton, 2014).In addition to the provision of increased shopping values, an essential contributor to the attractiveness of a district is a high level of safety.By including measures meant to repair the signs of urban disorder and improve security, BID programmes are expected to make neighbourhoods less attractive for offenders (Hanish & Guerra, 2000).Finally, the presence of amenities such as good retailing and a high level of personal security may also induce property prices to rise to offset the higher level of utility that these amenities then generate (Glaeser, 2008).

PREVIOUS STUDIES ON BIDS AND BID EFFECTIVENESS
Although the body of literature about the functional mechanisms of BID programmes and their effectiveness is quite vast, early studies have mostly focused on defining BIDs, describing the BID implementation process and proposed measures, and intuitively debating the potential benefits and concerns related to BID implementation (e.g., Michel & Stein, 2015;Mitchell, 2001;Morçöl & Wolf, 2010).Several studies have employed before-andafter or trend analysis, cost-benefit analysis, or comparisons between BIDs and non-BIDs to assess the effectiveness of BID programmes.Such methodological approaches are, however, unlikely to differentiate the effect of BIDs from the effects of other extraneous variables, simultaneous events, or local and regional economic trends (Caruso & Weber, 2006;Mitchell, 2001;Reenstra-Bryant, 2010).These studies are, however, valuable for reviewing the wide palette of indicators that could be used to assess BID success, such as customer flows, firm revenue and employment, the mix of stores and services, average rents and property values/prices, vacancy rates, and crime levels in the BIDs and their surrounding areas (Reenstra-Bryant, 2010).
A few studies have used more robust quantitative methodologies to distinguish the impact of BID programmes.These are summarised in Table A1 in Appendix A in the supplemental data online and are categorised based on whether they investigate the impact of BIDs on firm performance, crime levels or property values.
Few of these studies have attempted to robustly isolate the effects of BID programmes on firm performance.One rare exception is Sutton (2014), who investigates the impact of BIDs established between 2002 and 2008 in New York City on the economic performance of incumbent retailers.Sutton finds that the BID programmes had no significant impact on retailers' revenues, or the number of employees, compared with retailers in statistically comparable non-BID census tracts.However, Sutton finds a decline in revenues and number of employees of incumbent firms located in small community BIDs that are often characterised by a narrow scope of services and a weak economic environment.Sutton's results indicate that the effects in larger BIDs are positive, though only modestly significant for revenues and not statistically significant for employment.Sutton then argues that the negative effects in small BIDs are due to stronger competition effects, while the slightly positive effects in larger BIDs are due to agglomeration economies.
As is true of many other of the studies listed in Table A1 in Appendix A in the supplemental data online, Sutton's (2014) research is focused on the effects of BIDs in New York City, which are often larger and benefit from resource levels surpassing those of many other BIDs established in the United States or internationally.These BID programmes are also legally binding in character, which contrasts with the BID model implemented in Sweden.Methodologically, Sutton restricts her analysis to retail firms and builds her discussion on the effects generated from the increase in independent retailers in the BIDs.To tackle the endogeneity issues related to the non-random establishment of BIDs, Sutton uses propensity score matching to find adequate control establishments.However, as pointed out by Greenstone et al. (2010), such a strategy implies that the adoption of BID programmes can be correctly modelled by observable characteristics, while in most cases, many important characteristics are generally unknown and unobserved by the researcher.
Research measuring the effects of BIDs on crime levels has mostly focused on urban spaces located within the boundaries of the United States, but the results are far from unanimous.Several studies have confirmed reductions in crime rates.MacDonald et al. (2010), for example, find a decrease of 12% in the incidence of robberies and of 8% in the incidence of violent crimes following the implementation of BID programmes in Los Angeles.Based on a sample of 4327 police departments spread across the United States, Mello (2018) finds that BID cities hiring one more police officer decrease violent crime by 1.3% and property crime by 0.8%.This result is confirmed by Piza et al. (2019), who find that opening a police station in a BID leads to a decrease in burglary and motor vehicle theft in that district.However, Clutter et al. (2019) instead find that the expected robberies per foot of street length increase by 59% in BIDs, in line with arguments that a higher density of firms and customers in an area increases the number of 'desirable targets' and thus in the incidence of crime (Cohen & Felson, 1979).Han et al. (2017) show that such effects may also be heterogeneous over time, as they record stronger negative impacts on crime directly after the implementation of the BID.
The estimated changes in property values due to BID programmes are not undisputedly positive either.Ellen et al. (2007) use a difference-in-differences approach with a hedonic price model to estimate the effects of 44 BIDs in New York City over the period 1974-2003, finding that commercial property values within the BIDs appreciated by 15.7-31.2%,while residential values appreciated by 12.4% following the implementation of the BID programmes.They also find that only large BIDs have a significant impact, a result confirmed by other studies (e.g., Gross, 2005).The positive effects of BIDs are also identified by Jansen (2017), who estimates increases of £68,000-105,000 3 for properties located both within the BIDs and in adjacent neighbourhoods.Hanson (2017) instead reports that the price of properties located either within BIDs or within 500 m of their boundaries decreased by 17.18%.Using a difference-indifferences specification, Hirao (2021) also finds negative effects of 10-42% for the value of residential properties after the implementation of BID programmes in Westminster, UK.
Several studies have also indicated that the impacts of BID programmes are concentrated within the BIDs.Mello (2018) finds that hiring one more police officer in a BID city results in little change in neighbouring cities.However, other studies have indicated that the effects of BID programmes do extend beyond the boundaries of the BID.Clutter et al. (2019) report that the increase in expected robberies per foot of street length that they found within the BID also extends to neighbouring areas, although the effect exhibits a sharp distance decay of 3.5% per street block outside the boundaries of the BID.
In the case of property values, several studies (e.g., Ellen et al., 2007) have pinpointed the lack of spillovers or the presence of sharp distance-decaying effects (Hanson, 2017;Miller, 2013).Other studies have indicated that there are spillover effects on property values: both those that are in the same direction as the effects within the BIDs (Jansen, 2017) and those that are in the opposite direction after 'crossing' over the BID borders (Hirao, 2021).
It should be noted that much of the literature reviewed in this section relates to BID programmes implemented in settings quite different from ours.Many studies are from large North American or UK cites, while we study small to midsized Swedish towns.The legal structure surrounding the implementation of the BID programmes under study is also different in the United States and Europe, with participation and funding being voluntary in Sweden.Finally, the measures taken within different BID programmes also differ, not only between countries but also between different BID initiatives within countries.As such, comparisons between countries or different types of BID programmes should be done with caution.

Selecting control group towns
We investigate the impact of a voluntary Swedish BID programme, which was implemented in five small to medium-sized towns (Filipstad, Ludvika, Orsa, Rättvik and Torsby; henceforth called 'treated' towns) in 2015 under the supervision of the Swedish Association of Towns and Cities, on firm performance, property values and urban safety.Each of the five is a small monocentric town (with a population between 5000 and 15,000 inhabitants); their economies are based on natural resources and focus mainly on the production of manufactured goods and electrical power, as well as on tourism.These towns were not chosen at random but were part of an initiative by the Swedish Association of Towns and Cities to implement their so-called 'Swedish BID model', with the same steps and timeline for implementation specifically in small to midsized Swedish towns.The programme focuses on enhanced security, the removal of graffiti, street cleaning and sanitation, improvements in lighting, placemaking (e.g., facade beautification, the maintenance of public sidewalks, streetscaping and landscaping), the construction of new meeting spaces, and the facilitation of cooperation between public and private stakeholders.The geography of the BID and the possible spillover areas are illustrated in Figure 1 for one of the treated towns, Rättvik.
Several previous studies have used matching techniques to address the selection bias inherent in the adoption of BID programmes (Brooks, 2008;Brooks & Strange, 2011;Piza et al., 2019;Sutton, 2014).Matching techniques reduce selection bias by identifying counterfactuals based on control groups constructed based on information observed before treatment, for example, firm density, firm revenue, assessed property value, population levels and/or population density (Daunfeldt et al., 2017;Dehejia & Wahba, 1999, 2002;Hanson & Rohlin, 2018;Sutton, 2014).As pointed out by Greenstone et al. (2010), such a strategy implies that the adoption of a BID programme can be correctly modelled by the observable characteristics of the towns.However, the formation of BIDs is not always a transparent process, and the decision to implement BID programmes is thus not always based on measurable indicators.
To address this empirical problem, we use seven other small to midsize towns that adopted the same BID programme as the treated towns, but after the conclusion of our study period, as controls for the five treated towns that implemented the BID programme in 2015.The control cities are Borlänge, Enköping, Grängesberg, Lidköping, Simrishamn, Sollerön and Svärdsjö, each of which implemented the BID programme in either 2019 or 2020.These control cities are also situated in the Mid-South part of Sweden and are small and medium-sized monocentric towns with populations between 1000 and Effects of business improvement districts: an empirical study of five small to medium-sized Swedish towns 555 25,000 inhabitants, except for Borlänge, which has 45,000 inhabitants.The towns in the treated and control groups are similar in population density with fewer than 90 inhabitants/km 2 , as well as regarding economic structure, all relying heavily on natural resources, the production of manufactured goods and electrical power, and tourism.The population statistics also show stagnant or even negative population growth trends in both treated and control group towns. 4 Looking at economic indicators, we find that BID and control towns are quite similar.For example, all towns exhibit high unemployment rates (10.4-16.0% in the BID towns and 10.0-15.5% in the control towns, as compared with 7.6% at the national level); the share of selfemployed fell between 5.5% and 11.3% in the BID cities and between 5.6% and 15.5% in the control cities (with a national average of 9.2%); and the monthly average income of the population is between SEK24,000 and SEK27,500 in the BID towns and between SEK27,000 and SEK30,000 in the control cities (SEK37,100 nationally).Other socio-economic indicators, such as annual crime rates, also indicate that these cities have similar problemswhile the crime rates per 100,000 inhabitants are between 6900 and 10,200 in the BID cities, they are between 7000 and 11,700 in the control cities.This can be compared with the national average crime rate of 14,200 crimes, and of between 14,800 and 21,200 in the three largest cities in Sweden (Stockholm, Gothenburg and Malmö).
Most importantly, however, is that both the entry and control-group towns were deemed in need of BID programmes within a six-year period, indicating that these locations are not only similar in observables but also in the unobservable determinants of establishing BID programmes.Similar methods to find counterfactuals has been previously used by, for example, Brooks and Strange (2011), 5 who choose as controls areas which they refer to as 'almost BIDs', that is, areas that adopt the BID programme at the end of the period chosen for analysis, had their BID programme revoked or considered adopting the BID programme but never officially started the process (see also Brooks, 2008).However, the focus on small monocentric towns, implementing the specific BID model of the Swedish Association of Towns and Cities, also means that inference out of sample should be done with care.Larger cities, with other types of BID programmes being implemented, could be facing other outcomes due to scale, geography and economic context.

Model specification
To find out how the implementation of the BID programme impacted incumbent firms within the treated towns, we estimate a difference-in-differences model.That is, we compare the levels of the outcome variables within and outside the BIDs in the treated towns after treatment with those before treatment and with the levels of the outcome variables in the control towns during the whole period of analysis.
A standard difference-in-differences regression model in which we have two treated areas in the BID towns (one inside the BID, TR in , and one outside, TR out , which may possibly experience spillover effects) can be written as: ( where TR in is an indicator equal to 1 for all observations within the BID in the treated towns; TR out is an indicator variable equal to 1 for all observations located within the treated towns but outside the BID; TP t is an indicator variable equal to 1 for the years after the introduction of the BIDs; and (TR in × TP t ) is thus equal to 1 for observations located inside the BID in the treated towns in the years after the introduction of the BID programme; while (TR out × TP t ) has a similar interpretation for observations outside the BID in the treated towns.The parameter estimates b 3 and b 4 thus measure the impact of the introduction of the BID programme for observations located within and outside the BID, respectively.Finally, 1 it is a random disturbance term assumed to have zero mean and constant variance.
A potential drawback of the specification in equation ( 1) is that it controls for heterogeneity only at the treatment group level via the TR in and TR out indicator variables and thus ignores potential heterogeneity at lower levels.Recent applications of difference-in-differences analysis (e.g., Arcidiacono et al., 2020) have thus often also controlled for heterogeneity at the level of the observational units under study (i.e., firms, real estate properties).We follow this approach when data are available, 6 and our preferred regression model can thus be written as follows: where Y it is the outcome variable in our analysis (i.e., labour productivity, real revenues, number of employees, number of crimes, real estate values); b i is an observational unit fixed effect; and b t is a year-specific fixed effect.The terms (TR out × TP t ), (TR out × TP t ) and 1 it have the same interpretation as above.Note also that the log transformation of the outcome variable (ln Y it ) has the benefit of making the parameter estimates related to the effect of the BID programme interpretable in percentage terms after using the formula 100 Finally, the effect of BID programmes is not necessarily linear over time, and to test the hypothesis of non-linear effects, we also estimate our model using the following specification: b l , we thus estimate one effect parameter for each of the years 2015-18 regarding both those observational units located within the BID and those located outside the BID but within the treated towns.

Data and descriptive statistics
Part four of the implementation process for the BID programmes require the towns to create a business development plan.The goals included in these business development plans vary slightly from town to town, but most of them are focused on three main themes: a better business climate that attracts and retains firms in the area, a more attractive environment that attracts residents and visitors, and a decrease in blight. 7As such, we choose to focus on the performance of the firms in the area, property values and urban safety as our outcome variables.Following Özçelik (2020), we measure changes in firm performance in terms of labour productivity and decompose this measure into changes in real revenues and changes in the number of employees.These variables are available to us because all limited liability firms in Sweden are required to report such information annually to the Swedish Companies Registration Office (CRO).We use a database compiled by Bisnode, which has gathered this information from the CRO.The dataset includes, among other variables, revenues, number of employees, location and industry classification. 8The panel structure of the data enables the estimation of our preferred models as described in equations ( 2) and (3).
We also study how the implementation of the BID programme has affected urban safety by measuring the number of crime reports for 250 different offences under Swedish law.Measuring crime and how different policy tools affect the number of crimes in small to midsized towns is difficult due to the relatively few crimes committed, and especially when these small towns are also divided into different areas (BID and spillover areas). 9Crimes are measured on the town, area, type of crime and year level.In the estimations we compare the number of crimes in the BID areas and the spillover areas with the number of crimes in the control towns. 10As such, the number of crimes is not related to population.We chose this approach because population in these towns does not change much during the short period of analysis, and we control for such developments by the inclusion Effects of business improvement districts: an empirical study of five small to medium-sized Swedish towns 557 REGIONAL STUDIES of area-and year-specific fixed effects in our empirical model.Data on the number of reported offences for the years 2016-18 were provided by the Swedish Police for both the treated and the control towns.The first available year for the number of crimes is thus 2016, which implies that we are not able to use a difference-in-differences model in this case.Instead, we use a model similar to that described in equation ( 1), with the addition of a crime-category fixed effect.This approach allows us to compare the number of crimes committed within the BIDs and in the potential spillover areas in the treated towns to crime rates in the control towns while controlling for possible heterogeneity in the types of crimes.Since data are available only for the period after the establishment of the BIDs, the comparison in this case is a straightforward analysis of cross-sectional differences in the number of crimes in the different areas (BID and spillover areas versus control towns), and the results should therefore not be interpreted as causal effects of the implementation of the BID programme.
We are also interested in how the introduction of BIDs affects property values.However, a practical problem one faces when trying to measure how the introduction of BID areas has affected property values is that we need data on property values measured with such a geographical accuracy that we can identify what objects are within the BID and spillover areas.This excludes housing price indexes as they are only made on the municipal or town level.Ideally, we would like to be able to follow individual properties over time using market values, that is, property sales prices.However, there are simply too few sales each year, and the objects sold are also too heterogeneous, in these towns to be of any practical use.We therefore use data on property values from real estate valuations assessed for all Swedish properties for taxation purposes. 11These are made regularly on all properties, irrespective if they have been sold or not, thereby increasing the number of observations in our data.In addition, we can geocode each property on the address level, minimising the risk of properties being categorised as belonging to the wrong area.
These valuations are set by the Swedish Tax Authority and equal 75% of the property's estimated market value two years before the taxation event.In addition to selfreported data from owners on restorations and other actions that could affect the market value of the properties, the tax authority uses data on actual sales in the area, as well as their own valuations, to obtain a final estimate of the property's market value.Assessments of the properties in our analysis took place in 2012 and 2018, implying that we have access, for each property, to only one observation before and one after the introduction of the BID programme.This lack of observations makes the use of our preferred model impossible, and we instead rely on the standard differences-in-differences model presented in equation ( 1), but with the addition of municipality-specific fixed effects to account for possible heterogeneity in local real estate markets.Descriptive statistics for our five outcome variables at the beginning and the end of our study period (i.e., in 2012 and 2018) are presented in Table B1 in Appendix B in the supplemental data online.
We attempt to isolate the effects of the BID programme from the effects of other factors by means of a difference-in-differences analysis.This methodological approach assumes that the development of our outcome variables in the control towns accurately reflects the development of these variables in the treated towns in the absence of treatment.This is impossible to test, but we follow the standard approach to verifying this assumption by checking that the development of the outcome variables in the control towns is similar to that of the outcome variables in the treated towns in the period before treatment.We then assume that the trends in the treated towns would have continued unaltered after the treatment point if the BIDs had not been implemented (Hanson & Rohlin, 2018).
The parallel trend assumption is confirmed by the data in Table C1 in Appendix C in the supplemental data online, which presents the dependent variables in the firm data regressions year by year together with their associated 95% confidence intervals.The data show no clear trends in any of the outcome variables before treatment, and the confidence intervals for the different areas overlap in most cases.This fact suggests that the outcome variables in the treated and control towns were developing similarly before treatment, indicating that the selected towns are likely to be valid counterfactuals for the treated towns.Moreover, our treated and control towns are located at a considerable geographical distance from each other, which suggests that the introduction of the BIDs affected only the treated towns and that there were no spillovers into the controls (Hanson & Rohlin, 2018).

ESTIMATION RESULTS
For the outcome variables related to firm performance, the impact of the introduction of the BID areas is assessed using a difference-in-differences regression model similar to that used by Arcidiacono et al. (2020), presented in equation ( 2).This specification allows us to compare firms in the treatment groups with firms in the control group.The first hypothesis to be tested is whether the development of productivity, revenues and employment significantly differs between the groups of firms (treated and spillover as compared with controls), which would be indicated by statistically significant b 4 and b 5 coefficient estimates as presented in equation (2).Our identifying assumption when estimating the impact of BIDs on firm performance is thus that the treatment variables related to the introduction of the BIDs are uncorrelated with the error term of the regression equation, conditional on the firmand timespecific fixed effects in equation ( 2).
The results from estimating equation ( 2) are presented in Table 1 (model 1), showing that the implementation of the BID programme significantly increases the labour productivity of firms located within the boundaries of the BIDs by, on average, 7.62%.The increase in labour productivity for firms located in BIDs is due to the considerable increase in real revenues of 11.65%, concurrent with a much lower (but still significant at the 10% level) increase in the number of employees by 3.74%.
The year-by-year estimations (equation 3) show that the effect on productivity was lower in the first year after implementation (an increase of 6.23%, statistically significant at the 10% level) than in the second year (an increase in productivity of 11.04%, significant at the 1% level).Thereafter, the effect on productivity becomes insignificant, which suggests that the BID programme yields mainly transitory effects for a shorter period after it is implemented.
The estimation of the year-by-year treatment effects on real revenues after the implementation of the BID programme (Table 1, model 2) suggests that the impact is positive and significant immediately after implementation -12.40% in 2015 and 13.74% in 2016but that it decreases to 5.92% in 2017 (still significant at the 10% level).In 2018, however, we no longer find any statistically significant effect of the implementation of the BID programme on firms' real revenues.
Overall, we find no statistically significant effects of BIDs on the number of employees in the treated towns.In the year-by-year estimations, we find that the implementation of the BID programme increased the number of employees in 2015 by 5.64% in the BIDs, which might indicate that firms temporarily hire additional personnel to implement the changes required by the BID programme.
Turning to the impact of the BID programme on crime, we find that BIDs have 2.66% fewer crimes than the control towns and that this difference is not statistically significant at conventional levels (model 1 in Table 1).The year-byyear model (model 2 in Table 1) reveals that the difference is negative for all three years (2016-18), with a difference of −4.94%, which is statistically significant at the 10% level, in 2018.Note, however, that data on crimes are available only from 2016 onwards, which means that we are not able to perform a difference-in-differences analysis.Because of this, the results should not be interpreted as estimates of the causal effects of the BID programme, but rather as an indication of how crime rates in BIDs and the areas adjacent to them differ from the crime rates in the control towns.
Finally, we turn to the estimation results regarding the effects of the BID programme on property values.Since data for the property value assessments are only available for two years, our estimation method must be adjusted accordingly.In this case, our estimation method is more closely related to the standard difference-in-differences model presented in equation ( 1), but with the addition of municipality-specific fixed effects to account for heterogeneity in local real estate markets.Our identifying assumption when estimating the impact of BIDs on property values is thus that the treatment variables related to the introduction of the BIDs are uncorrelated with the error term of the regression equation, conditional on the municipality specific fixed effects.Estimated under this Effects of business improvement districts: an empirical study of five small to medium-sized Swedish towns 559 assumption, we find no statistically significant effects on property values within the BIDs.The results from analysing the impact of BIDs on adjacent areas are reported in Table 2.We find no statistically significant effects for any of our firm performance variables or for property values when estimating our most basic model (model 1 in Table 2).This is also the case for the year-by-year results (model 2 in Table 2), except for a 3.87% increase in the number of employees in 2017 (significant at the 10% level).However, we find some indications that the number of crimes in the spillover areas is higher than that in our control towns when measuring the average effect over all study years (model 1 in Table 2: 3.43% higher, significant at the 10% level).The yearby-year estimation shows that although more crime is found in the spillover areas across all years, the only statistically significant result is during the year when the BID was established (model 1 in Table 2: 3.43% higher, significant at the 5% level).
Last, we find no statistically significant effects on property values outside the BIDs.Thus, apart from crime rates, our results show no signs of spillover effects to adjacent areas in the treated towns.This seems to imply that the impact of the BID programme is mostly restricted to the implementation area, and any positive effects seem to be largely transitory in nature.

DISCUSSION AND CONCLUSIONS
We have investigated the effects of a voluntary Swedish BID programme on firm performance, urban safety and property values, finding that the BID programme increased labour productivity of the incumbent firms located within the boundaries of the BID by 7.62%.This effect was driven by a considerable increase in real revenues, combined with a smaller increase in the number of employees.Our results thus imply that voluntary BID programmes can be effective for improving the performance of incumbent firms within the BID, suggesting that regulatory action might not be needed to achieve the desired effects on firm performance.
However, the positive effect of the BID programme on firm performance is largely transitory, suggesting that voluntary BIDs mainly yield transitory effects for a short period after their implementation.There are numerous explanations for this result.First, the novelty of and thus the interest in BID projects might wear off over time.The decaying employment effect may reflect the shortterm investments that are usually made during the first 18 months of a project when it is implemented under the management of the Swedish Association of Towns and Cities.Another interpretation is that the voluntary element of the BID programme leads to freeriding behaviour, reducing the effectiveness of the programme.As we are not able to distinguish among these competing hypotheses, the efficiency of voluntary BID programmes constitutes an interesting avenue for further research.
We furthermore found that the BIDs had 2.66% fewer crimes than the control towns, but this difference is not statistically significant at conventional levels.The model in which the year-by-year differences are studied shows that the differences are negative in all three years (2016-18) but statistically significant only in 2018 (−4.94%).These results should be interpreted with caution due to the lack of pre-treatment data, but it should be noted that the previous literature (e.g., Cook & MacDonald, 2011) has also reported a lag between the implementation of BID programmes and reductions in crime rates.Finally, we found no statistically significant effects of the BID programme on property values.This result is surprising since theoretical studies and the bulk of the empirical research (e.g., Ellen et al., 2007;Miller, 2013;Jansen, 2017;Hanson, 2017) indicate positive effects.There are, however, some possible explanations for these differences in outcomes.First, previous studies have focused on very large cities, for example, New York, Los Angeles and Washington, DC, while we investigated the effect in much smaller towns.Second, the properties analysed in our study are a mix of commercial and residential properties, and at least for the commercial real estate market, property values are closely connected to revenues.As we find that the introduction of BIDs had only short-term effects on revenues, these effects are likely too small to affect real estate values.Third, the lack year-to-year measures of property values for individual properties is a data limitation that might make it difficult to obtain statistically significant results.Access to such data would provide an interesting avenue for future research.
Regarding the impact of BIDs on areas adjacent to them, we find no statistically significant effects on firm performance or property values, except for a 3.87% increase in firm employment in 2017.This is likely not an effect of the implementation of the BID programme two years earlier but rather due to some coincidence that affected the treated towns rather than the controls at that point in time.Several previous studies have also indicated a lack of spillover effects (e.g., Ellen et al., 2007;Gross, 2005;Jansen, 2017) or the presence of sharp distance-decay effects (Hanson, 2017;Miller, 2013).
Turning to crime rates, we found that the level of crime is higher in the spillover areas than in the control towns when measured as the average over all years under study.The year-by-year estimation showed that although a higher level of crime was found for all years, the only statistically significant result was found during the year when the BID programme was implemented (+3.43%).This might indicate that crime is pushed outside the BID boundaries following the implementation of safety and security measures within the BID.
Our results could be seen as an indication that gentrification of one area increases crime in adjacent areas (Porreca, 2023).The only study directly measuring the impact of BIDs on crime in Sweden is Kronkvist and Ivert (2020).They found a significant reduction in crime in the affected area, combined with spatial diffusion of benefits rather than displacement of crime.However, since we lack pretreatment data in our study, we advise caution when interpreting our results.It should also be noted that Cook andMacDonald (2011), MacDonald et al. (2013) and Mello (2018) do not identify any statistically significant effects on crime levels outside the BIDs.The lack of pre-BID data concerning reported crimes is a key limitation of our study, and these results should therefore only be regarded as an indication that BIDs in Sweden might displace crime to adjacent areas.More research, using more detailed datasets following the development of reported crimes in BID and spillover areas is warranted.
There is also a possibility that the introduction of BIDs could lead to commercial gentrification, a process where the actions taken within the BID scheme lead to increased attractiveness of the city centre that in turn increases revenues for firms located there.However, the increased attractiveness then also makes it possible for property owners to increase rents, which could potentially displace the least profitable firms from the BID areas.Such a process has, for example, been documented by Hertting et al. (2022) for a BID programme in Gothenburg, Sweden.
The study of entry and exit due to the introduction of BIDs has been outside the scope of our study.The reason is that the process of firm displacement would likely take place over much longer time periods than we have data for.Our annual report data are furthermore not well-suited to study entry and exit because the only way of identifying such events is by recording when a firm organisation number first appears in or leaves the dataset.There are, however, other reasons for this than a formal entry or exit of a firm.If, for example, the firm is sold, it is often the case that the firm is registered with a new organisation number.
Still, based on the results presented above, we consider the risk of commercial gentrification due to the introduction of the BIDs under study to be low.Our findings only show transitory effects on revenues and productivity of the affected firms, and we find no effect on property values which would likely be the case if property owners could increase rents in any significant manner.However, if one could find suitable, detailed data of firm entry and exit and covering a longer time period, the issue of commercial gentrification due to the introduction of BIDs in small to midsized towns is another interesting question that warrants more research.
We conclude that voluntary BIDs in small to midsized Swedish towns seem to be effective to some extent.However, the focus on small monocentric towns, implementing the specific BID model of the Swedish Association of Towns and Cities, also means that inference out of the sample should be done with care.Larger cities, with other types of BID programmes being implemented, could be facing other outcomes due to scale, geography and economic context.Furthermore, the short-lived impact of these schemes leaves open discussions about the size of the investments, freeriding and the limited capacity of these towns to work with a tool that requires extensive resources.Mihaescu et al. (2022), for example, found considerable heterogeneity in the proportion of businesses that were willing to contribute to the formation and costs of the BID.They also found that the most important contributors to the BID financing were the municipalities, which can be compared with other countries where the private sector in most cases take full responsibility for the financing through obligatory taxes directly after the initial stage of the process (Hoyt & Gopal-Agge, 2007).
Another particularity of the voluntary Swedish model according to Mihaescu et al. (2022) is that most of the financing was spent on administrative activities related to establishing the initial partnerships between the stakeholders, while only a small share of the resources went to actual projects aimed at improving the physical environment of the area or to safety-related measures.However, most BID managers nevertheless declared that they were pleased with the results of the BID programme, and about half of the surveyed managers also stated that they did not want any new legislation to regulate the functioning of BIDs in Sweden.

Figure 1 .
Figure 1.Rättvik business improvement district (BID) and its spillover area, with the location of incumbent firms.
where D YEAR t refers to indicator variables for observations in2015, 2016, 2017 and 2018.

Table 1 .
Effects of the business improvement district (BID) programme on labour productivity, real revenues, employment, real estate valuations and crime within the treated BIDs.

Table 2 .
Spillover effects of the business improvement district (BID) programme on labour productivity, real revenues, employment, real estate valuations and crime