Should I stay or should I go? An analysis of Portugal’s intracountry migration

University of Minho School of Economics and Management Francisco Figueira Ribeiro Should I stay or should I go? An analysis of Portugal’s intracountry migration. april 2025 University of Minho School of Economics and Management Francisco Figueira Ribeiro Should I stay or should I go? An analysis of Portugal’s intracountry migration. Master’s Dissertation Master in Economics Dissertation supervised by Professora Doutora Linda Veiga april 2025 Copyright and Terms of Use for Third Party Work This dissertation reports on academic work that can be used by third parties as long as the internationally accepted standards and good practices are respected concerning copyright and related rights. This work can thereafter be used under the terms established in the license below. Readers needing authorization conditions not provided for in the indicated licensing should contact the author through the RepositóriUM of the University of Minho. License granted to users of this work: CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/ i Acknowledgements The completion of this dissertation was made possible thanks to the support and contribution of many people, to whom I express my deepest gratitude. I would firstly like to extend my gratitude to Professor Linda Veiga, for the invaluable guidance, support, knowledge and opportunities along the development of this dissertation. Thank you for the patience and motivation. I am also grateful to all the teachers I have had the privilege of learning from. A special thank you to my friends, particularly Pedro Machado and Francisco Cabeleira, with whom I have shared this academic journey with since day one. Your support, insightful discussions, and friendship over the last five years have meant the world to me. Lastly, my deepest appreciation goes to my parents and sisters for their endless love, patience and encouragement, allowing me to pursue my studies. Your unwavering support has made this journey possible, and for that, I am forever grateful. ii Statement of Integrity I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. University of Minho, Braga, april 2025 Francisco Figueira Ribeiro iii Abstract Population flows have led to an uneven distribution of Portugal’s population and certain municipalities and regions have become more developed than others. This study investigates which factors influence the attractiveness of a municipality. The empirical analysis uses a panel data set of Portuguese mainland municipalities from 2007 to 2023, estimating fixed effects and random effects regressions. Municipal net migration serves as dependent variable, while local government fiscal policy, socioeconomic and amenity variables as independent variables. The analysis found that while certain socioeconomic factors influence migration patterns, mostly those regarding the labour market, local government’s fiscal policy and amenities did not exhibit a statistically significant impact. This may suggest there is a local governments’ lack of influence due to the high degree of centralization of state activities in Portugal. Keywords Internal Migration; Net Migration; Portuguese Municipalities; Tiebout Model. iv Resumo Os fluxos populacionais conduziram a uma distribuição desigual da população em Portugal e alguns municípios e regiões tornaram-se mais desenvolvidos do que outros. Este estudo investiga quais os factores que influenciam a atratividade de um município. A análise empírica utiliza um conjunto de dados em painel dos municípios de Portugal continental de 2007 a 2023, estimando regressões de efeitos fixos e de efeitos aleatórios. O saldo migratório municipal serve como variável dependente, enquanto as variáveis socioeconómicas, política fiscal do governo local e comodidades são variáveis independentes. A análise concluiu que, embora alguns fatores socioeconómicos influenciem o saldo migratório, principalmente os relacionados com o mercado de trabalho, a política fiscal e as comodidades locais não tiveram um impacto estatisticamente significativo. Isto pode indicar a pouca capacidade de influência dos governos locais dado o elevado grau de centralização das atividades do Estado em Portugal. Palavras-chave Migração Interna; Saldo Migratório; Munícipios Portugueses; Modelo de Tiebout. v Contents 1 Introduction 1 2 Theoretical Framework 3 2.1 Migration....................................... 3 2.2 TieboutHypothesis.................................. 7 2.3 EmpiricalStudies................................... 8 3 The Portuguese Case 16 3.1 InstitutionalFramework................................ 16 3.2 General socio-economic context . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 3.3 Local government policy action to attract population . . . . . . . . . . . . . . . . . . 24 3.3.1 Municipality of São João da Pesqueira . . . . . . . . . . . . . . . . . . . . 25 4 Data and Descriptive Statistics 31 4.1 Data and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 5 Empirical Framework 35 5.1 EstimationStrategy.................................. 35 5.2 Baselinemodel.................................... 36 6 Empirical Results 38 6.1 Baselinemodel.................................... 38 6.2 Five-yearperiodmodel ................................ 42 7 Conclusion 45 Bibliography 47 vi 2 Theoretical Framework The prospects of a region or place have always been closely tied to its demographic structure. Demographics play a decisive role in shaping decisions made by central governments and businesses regarding resource allocation. This is why migration flows are so important for a region’s economic, social and amenity development, in this line regional policy has been considered vital to attract migrants, through the promotion of a region or place (Rodríguez-Pose and Ketterer, 2012; Niedomysl, 2004). To prevent major population shifts, local governments must ensure their territories have good levels of economic activity, businesses, infrastructures, services and amenities to stabilize their population levels and improve the territory’s attractiveness. While amenities have been considered to assume a relevant role in the appeal of places in the United States, the role of economic factors has been predominant in Europe (Rodríguez-Pose and Ketterer, 2012; Faggian et al., 2012). While local governments through public goods, services, tax and expenditure have also been considered relevant in influencing location decisions (Friedman, 1981; Day, 1992; Westerlund and Wyzan, 1995; Dahlberg et al., 2012). From a microeconomic perspective, households/individuals have always been considered to maximise their utility under a budget constraint, taking into account both economic and non-economic aspects. While economic aspects often include expected income variables such as wages, unemployment rates and local tax rates. Non-economic include, natural and man-made amenities and public services (Faggian et al., 2012). The following two sections present a theoretical background, on local governments and migration and the third section on the main empirical studies. 2.1 Migration The study of migration by economists has been mainly split into individual choice models and models of spatial equilibrium and optimal dynamic migration (Jia et al., 2023). Individual choice models started to consider the individual’s decision to migrate under a cost-benefit framework, people choose to move where their future stream of benefits (mostly wages) is maximised. Still, they incur costs of travelling, finding work, 3 learning a new language and culture, and emotional attachment (Sjaastad, 1962). Incorporating these analytically and treating migration as an investment (similar to human capital investment), the individual decides to move from location ito jif the earnings and cost differentials over space are maximised and positive (Sjaastad, 1962). Let the present value of earnings stream in place jless than those in ibe: n ∑ t=1 (Ejt −Eit) (1 + r)t,(2.1) where r is the discount rate and E the earnings in each location. And the present value of net costs between these two places is: n ∑ t=1 (Cjt −Cit) (1 + r)t,(2.2) C stands for costs in each location. Then the present value of the investment in migration from ito jis: PVij = n ∑ t=1 Ejt −Eit (1 + r)t− n ∑ t=1 Cjt −Cit (1 + r)t.(2.3) Kennan and Walker (2011) extended individual choice models through a structural dynamic model, focusing on the relationship between income prospects and migration decisions. Individual migration decisions are based on an optimal search process, reacting to wage differentials across locations, by maximizing the expected lifetime income net of moving costs (Jia et al., 2023). Illustrated in the next equations, there are Jlocations, lbeing a vector of recent locations and l0the current location, ωa vector with information on wage and utility in each location, state vector xconsisting of l,ω, and age. ζjbeing a random variable independent and identically distributed across locations, periods and individuals and independent of state vector (Jia et al., 2023). Accounting for recent locations is the vector l, accounting for wages wis a vector of wages and utility information across locations. For individuals whose home location is hthe flow payoff is: ˜uh(x, j) = uh(x, j) + ζj,(2.4) where uh(x, j) = α0w(ℓ0, ω) + K ∑ k=1 αkYk(ℓ0) + αHχ(ℓ0=h) + ξ(ℓ0, ω)−∆τ(x, j).(2.5) The first term w(ℓ0, ω)in equation 2.5 represents wage in the current location (l0), the second term Yk(l0)consists of non-pecuniary variables (amenities) and the third the preference for the native location χ(ℓ0=h). Then ξis a random permanent component in each location, which is learned when the 4 location is visited. The term ∆τrepresents the moving cost from l0to ljand a unexplained component ζj, capturing shocks to preferences or moving costs (Jia et al., 2023). Lastly moving costs (2.6) are function of distance D(ℓ0, j)from the current location to j, and χ(j∈A(l0)) to represent a the set of locations adjacent to l0, and a previously known location χ(j=l1), age aand population size ηj (Kennan and Walker, 2011). ∆τ(x, j) = (γ0τ+γ1D(ℓ0, j)−γ2χ(j∈A(ℓ0)) −γ3χ(j=ℓ1) + γ4a−γ5ηj)χ(j=ℓ0).(2.6) Individuals maximise the expected present value of lifetime utility by moving from the current location, with moving costs, and accumulating utility in the next location, depending on the shocks to which they are exposed and the knowledge they have about opportunities elsewhere (Kennan and Walker, 2011). Beyond individual choice models, Tiebout (1956) introduces a framework assuming no moving costs and complete information, resulting in a free movement of consumers toward the community that best satisfies their preferences. Building on the concept of migration, Harris and Todaro (1970) propose a two-sector model of rural-urban migration, where the urban minimum wage exceeds rural earnings, and migration flows take place. This process will eventually lead to equilibrium, as migration continues until both wages are equal (Jia et al., 2023). Expanding the discussion to spatial equilibrium Roback (1982) presents a model with homogeneous and perfectly mobile workers where indirect utility depends on nominal wages, housing costs and local amenities. The model assumes a fixed land supply, due to limited housing elasticity and implies that any shock to the local economy is fully capitalized in land prices, leaving the utility of the worker unaffected. For instance, a positive productivity shock raises nominal wages and demand for housing, workers will consume less housing due to higher prices, which then facilitates in-migration. Moreover, areas with greater amenities attract more migrants, which lowers wages and raises prices of local goods, services and land (Jia et al., 2023). To account for dynamic migration patterns, Coen-Pirani (2010) develops a general equilibrium model of gross and net migration across locations, considering the unobserved heterogeneity of individuals in migration decisions. Gross migration flows are driven by idiosyncratic match shocks, between workers and locations, and net migration flows due to persistent productivity shocks. Workers are drawn to locations experiencing positive productivity shocks and discover their idiosyncratic match upon migration (Jia et al., 2023). Based on the work of Roback (1982); Beeson and Eberts (1989); Rappaport (2004); Faggian et al. (2012), Rodríguez-Pose and Ketterer (2012) present a simple locational choice spatial equilibrium model to analyse the attractiveness of different territories. In which net migration is modelled as being determined by firms’ and households’ reactions to differences in productivity and utility attributes across the territory 5 (Rodríguez-Pose and Ketterer, 2012). Firms and households are assumed to be mobile and their location preferences depend on profit and utility maximization across different places. Shaping the behaviour of firms, the net present returns between any region jand iare: 1 (1 + d)t∫∞ t Πj t≥1 (1 + d)t∫∞ t Πi t.(3) Companies are expected to maximize current and expected profits (π) across different locations discounted by d. Profits depend on wages, rental costs of land and exogenous natural or socioeconomic features of each place. Firms are located in areas with higher profits (j) until the long run, where current and expected profits are equal across all possible locations. The location decision made by companies will shape the economic characteristics and attributes across the territory (Rodríguez-Pose and Ketterer, 2012). On the other hand, the place-specific utility derived by each individual will depend on the consumption of goods, non-traded housing services and non-economic place-based natural or (man-made) cultural amenities, giving rise to net present utility (Rodríguez-Pose and Ketterer, 2012): Vi=1 (1 + d)t∫∞ t Vi t(Gi t, Di t, Zi t),(4) Gi t,Di tand Zi trepresent the consumption of goods, housing and amenities respectively. A lifetime budget constraint (Equation 5) dictates the mover decision taking into account rental and housing prices pi t, average wages wi tand the probability of becoming employed in any given place ei t(Rodríguez-Pose and Ketterer, 2012). 1 (1 + d)t∫∞ t(Gi t+pi tDi t)≤1 (1 + d)t∫∞ t wiei t.(5) With the maximisation of Equation (5) under the budget constraint, the following indirect utility function is achieved, framing regional net migration as the structural outcome of a set of factors (Rodríguez-Pose and Ketterer, 2012): Un(pn, Zn, wn, en, Sn, Nn), n :i, j. (6) The indirect utility is positively related to natural and sociocultural or general amenities Zn, household income wn, the probability of finding a job en, and although not considered in the empirical part of this work, the territory embedded socioeconomic regional features Sn, presence of migrant communities Nn and housing costs pn(Rodríguez-Pose and Ketterer, 2012). The migration decision will be triggered by the indirect utility difference between region iand region j(∆Uji =Uj−Ui), and the possible psychological and pecuniary utility costs of moving Cij. The individual moves from ito jif the utility differential between the two places minus the costs of moving 6 is higher than zero (∆Uij −Cij >0), meaning that the benefits of moving to joutweigh the costs (Rodríguez-Pose and Ketterer, 2012). Hence, according to (Ferguson et al., 2007,p.82) people ”vote with their feet”, one can assess the attractiveness of a region by analysing the in and out-flows of economic agents (Rodríguez-Pose and Ketterer, 2012). The population stock in region iat time t0is expressed as: Pi t0=Pi t1+Mij t1−Mji t1+di t1−bi t1(7) in which Pi t1stands for individuals who did not move between t0and t1,Mij t1the migrants moving away from region ito any region j,Mji t1migrants coming from any region jto region i, lastly di t1and bi t1 which are deaths and births respectively Rodríguez-Pose and Ketterer (2012). The population change in region iover the period t0and t1is (Rodríguez-Pose and Ketterer, 2012): Pi t1−Pi t0=Mji t1+bi t1−Mij t1−di t1(8) Rearranging and standardising equation 8 by the population stock at t0, an expression of the net migration rate of the region iis achieved, which indirectly depends on the utility differential across different places (Rodríguez-Pose and Ketterer, 2012). Migi t1=Mij t1−Mij t0 Pi t0 =(Pi t1−Pi t0−bi t1+di t1) Pi t0 .(9) 2.2 Tiebout Hypothesis The main problem in determining the optimal supply of a public good is the mechanism by which consumer voters register their preferences. From the government’s perspective, it must find the population’s preferences and tax them accordingly. If this happens, public revenues and expenditures reflect consumer-voter preferences, but consumers can understate their preferences and free-ride (avoid tax). Hence no local mechanism is put in place, to ensure the expenditure on public goods is at a proper level. If a resident is about to move to another jurisdiction, what variables will he consider relevant to conduct his move? It depends on his preferences, if he has children, certainly he is moving to a municipality with a considerable level of expenditure on schools. However, if he is retired, that would not be important as he would prefer expenditure on nursing homes and health services. Therefore, the supply and quality of infrastructure and services, parks, police stations, firefighter stations, roads, and parking lots, will enter an individual’s decision (Tiebout, 1956). Following the argument on local expenditures, Tiebout (1956) develops a Local Government Model, yielding a solution for the level of expenditures for public goods based on the following assumptions: (1) 7 consumer voters are fully mobile and will move to that community where their preference patterns, which are set, are best satisfied; (2) consumer voters are assumed to have full knowledge of differences among revenue and expenditure patterns and to react to these differences; (3) there is a large number of communities in which the consumer-voters may choose to live; (4) restrictions due to employment opportunities are not considered; (5) the public services supplied exhibit no external economies or diseconomies between communities; (6) for every pattern of community services set by a city manager, who follows the preferences of older residents of the community, there is an optimal community size (the number of residents for which the bundle of services can be produced at the lowest average cost); (7) communities below the optimum size seek to attract new residents to lower average costs, those above do the opposite, and those at the optimum try to keep their population constant. Given the full mobility assumption and the remaining, flows take place between communities that are not at the optimal community size. Each city manager has a demand for nlocal public goods, m−1 other city managers go into a national market and bid for adequate units of public service of each kind. The demand for each public good nwill be the sum of the demand for all communities m. This demand will approximate the demand that represents the true preferences of consumer voters (as if they revealed their true preferences) (Tiebout, 1956). However, the model is considered to fall short of the optimum, because if a given community is at the optimum population level, and some consumer-voter considers it the best preference fit, he/she cannot move there and must look for a perfect substitute. If that substitute isn’t available, the optimal preference position is not reached, and ultimately the solution will approximate the optimum. Two additional points are highlighted; firstly, changes in the costs of one public service will cause changes in the quantity produced and secondly, the costs of moving between communities should be recognised. (Tiebout, 1956). While some assumptions may not be realistic, Tiebout stated “The consumer-voter may be viewed as picking the community which best satisfies his preference pattern for public goods” (Tiebout, 1956, p. 418). Hence, Local governments being responsible for supplying public goods and services, and collecting the respective taxes, may be able to affect the consumer-voter community choice (migration flows). 2.3 Empirical Studies Beyond the theoretical sphere, extensive empirical literature addresses migration, Table 1 presents a summary of empirical findings on migration flows and community selection, that make use of aggregate data analyses. While Table 2, household-level analyses. In general these studies consider the role of socioeconomic, local government and amenity variables. 8 Commonly used variables to represent job market opportunities include unemployment, employment rates, and distribution of employment across sectors. Economic opportunities, Gross Domestic Product (GDP) per capita, average wages, and the shares of highly educated population/workers, a skilled labour force, can help to attract businesses and companies to a given community, making it more attractive to potential migrants. Highly skilled workers can also contribute positively to a community through knowledge spillovers, which can increase the general productivity of a place Rodríguez-Pose and Ketterer (2012). Amenities also play a significant role in shaping the attractiveness of places, contributing to their overall convenience, comfort, and enjoyment. Natural amenities such as maximum, minimum, and average temperatures, precipitation, cloudiness, coastal communities, and natural landscapes are frequently highlighted ((Graves, 1976, 1980; Rodríguez-Pose and Ketterer, 2012; Ketterer and Rodríguez-Pose, 2015)). Similarly, man-made amenities such as theatres, museums, housing and access to transportation (Sasser, 2010; Rodríguez-Pose and Ketterer, 2012; Dahlberg et al., 2012; Buch et al., 2014). Conversely, factors such as crime, traffic, low environmental quality, pollution, and high population density tend to decrease a region’s desirability (Buch et al., 2014; Banzhaf and Walsh, 2008; Germani et al., 2021). Local governments can also influence migration flows through public goods, services, and local taxes, as explored through the lens of community choice and migration. Most studies have employed discrete choice models using household data (Friedman, 1981; Quigley, 1985; Westerlund and Wyzan, 1995; Bayoh et al., 2006; Dahlberg et al., 2012). However, results have been mixed: while Friedman (1981) and Bayoh et al. (2006) found that public services attract new residents (albeit with small effects in the former and large in the latter), Quigley (1985) reported a negative impact. These differences may stem from differences in household characteristics and estimation approaches (Dahlberg et al., 2012). Other studies have used aggregate data, primarily finding a link between local government services, expenditure, and net migration (Cebula, 2024; Cebula and Clark, 2013; Cebula and Nair-Reichert, 2012; Brueckner and Št’astná, 2020; Yu et al., 2019; Day, 1992). 9 Table 1: Empirical Studies on Migration (aggregate data) Reference Dependent Variable Independent Variables (Sign) Methodology Sample Cebula (2024) gross in-migration rate (-)unemployment,(-)cost of living,(- )population density,(+)median family income,(+)average January temperature,(+)tax freedom indices. Panel 2SLS State level USA [20102017] Germani et al. (2021) net migration rate (-)unemployment,(+)income,(+)education,(- )air pollution,(+)value of real estate properties,(+)number of firms,(+)infrastructures. OLS and 2SLS Italy provincial-level data [2011-2015] Brueckner and Št’astná (2020) log difference between shares of population migrating and not migrating (+)prior interregional commuting,(-)job seekers per job vacancy/college graduates per job vacancy,(+)high educated population,(+)Govt Spending on transportation,culture and sports,(-)Govt Spending on industry and housing support,(-)population density. 2SLS Regional level Czech Republic [2011-2015] Continued on the next page 10 (Table 1 – Continued.) Reference Dependent Variable Independent Variables (Sign) Methodology Sample Yu et al. (2019) net migration rate (+)GDPpc,(- )unemployment,(+)education,(+)buses,(+)roads,(- )temperature,(-)humidity. Bootstrap linear regression Chinese city-level data [2010] Barreira et al. (2017) % population change between censuses (-)unemployment,(+)employment share(secondary and tertiary sector),(+)average maximum temperatures,(+)higher proportion of middle-aged vacant houses. Random Effects Portuguese city-level [1991-2011] Ketterer and Rodríguez-Pose (2015) net migration rate (-)unemployment,(-)agricultural share,(+)quality of government. Fixed effects and GMM EU NUTS-2 regions [19952009] Buch et al. (2014) net migration rate (of workers) (+)employment growth, (+)tourist stays,(-)crime rate,(+)recreation areas,(+)average flat size. Pooled OLS and Fixed Effects German cities [20002007] Continued on the next page 11 (Table 1 – Continued.) Reference Dependent Variable Independent Variables (Sign) Methodology Sample Cebula and Clark (2013) net migration rate (+)employment growth,(+)median family income,(-)cost of living,(+)mean january temperature,(-)population density,(-)state personal income tax,(- )property tax,(+) per pupil education expenditure,(+)medicaid expenditure per recipient. 2SLS State level USA [2000 and 2008] Cebula and NairReichert (2012) net migration (-)unemployment,(-)cost of living,(+)median family income,(+)employment growth,(-)state income and local property tax,(+)per pupil education expenditure. OLS and 2SLS State level USA [july2000july2008] Rodríguez-Pose and Ketterer (2012) net migration rate (+)GDPpc,(-)unemployment,(- )share of young,(+)social filter index1,(+)recreation and tourism,(+)January temperatures. Fixed effects, IV and GMM EU Regions [1990-2006] Continued on the next page 1accounts for territory innovation enhancing features, built using regional educational attainments and the composition of productive resources 12 Figure 1: Population Density and Percentage of Population Above 75 Years (a) Population Density (b) Percentage of population above 75 years Data source : National Institute of Statistics (INE) 19 Beyond the phenomenon of population ageing, the decline in economic activity is also becoming increasingly prevalent, evidenced by a reduction in employment opportunities and business presence, diminished investment levels, a stagnation in innovation, ageing infrastructure, shortage of human capital, and low productivity. These factors reflect a broader inability of these regions to generate value added (Figure 2) and to remain competitive both nationally and globally (Unidade de Missão para a Valorização do Interior, República Portuguesa, 2017). Figure 2: Gross Value Added (millions of euros) 2023 Data source : INE All these asymmetries led to substantial variations in the distribution of population. Figure 3a is proof that citizens have been moving towards urbanised areas, mostly across the coastline. In relative terms the municipalities of Mafra, Montijo, Moita and Albufeira have had a significant population growth. While Arruda dos Vinhos, Benavente, Palmela, Sesimbra, Cascais, Lagos, Portimão and Loulé exhibit more moderate population growth, characterized by lower growth rates. On the other hand, rural regions either closer to the coastline or in the hinterland, have been expe20 riencing sharp and significant population decreases, some exceptions include the interior district capital cities, which experienced slight growth, reflected in small positive growth rates. Figure 3: Population and employment variation at the municipal level (a) Municipal population variation 2001-2021 (b) Municipal employment variation 2001-2021 Data source : INE From 2001 to 2021 (Figure 3b), the variation in municipal employment aligns with what was expected. Urban municipalities located along the coastline exhibited a greater percentage change in employment. The most pronounced positive growth rates were observed in the municipalities of Torres Vedras, Mafra, Arruda dos vinhos, Sobral de Monte Agraço, Sesimbra, Alcochete benavente, Montijo, Palmela, Odemira and Aljezur. An analysis of employment variation across economic activity sectors between 2001 and 2021 reveals a significant shift in job opportunities, moving away from the primary and secondary sectors toward a predominance in the tertiary sector (Figure4 and A.1). Primary sector employment grew significantly in relative terms in the municipality of Odemira, with more moderate increases observed in Lousã and 21 Sardoal. In contrast, the vast majority of the remaining municipalities experienced a decline in primary sector employment. While employment growth in the secondary sector was negative in the majority of municipalities. Tertiary sector employment growth was positive across most of the territory. However, several municipalities in Alto and Baixo Alentejo registered declines in this sector. Unsurprisingly, these sub-regions also experienced population decreases, as illustrated in figure 3a. Figure 4: Secondary and tertiary sector employment variation at the municipal level (a) Municipal secondary sector employment variation 2001-2021 (b) Municipal tertiary sector employment variation 20012021 Data source : INE Overall, economic activity experienced significant growth and expansion in the coastal and surrounding municipalities of the districts of Braga, Porto, Lisbon, Setúbal and Faro (figures 2,3b and 4). This development contributed to the increased attractiveness of these regions, leading to a concentration of population, human capital, and higher wages. 22 Figure 5: Percentage of population from another municipality. Data source : Sales Index Regarding migration flows, the movement of population between municipalities is illustrated in 5. In 2001, just over two percent of the population migrated between municipalities, with this percentage showing a slight increase by 2011. Furthermore, in the most recent census, the registered percentage of citizens from another municipality reached nearly seven percent. One can conclude that inter-municipal relocation increased over the period from 2001 to 2021; however, this dynamic was not uniform, as several municipalities did not register net positive migration flows. Those belonging to the districts of Leiria, Lisbon, Setúbal and Faro were, in general, more attractive exhibiting positive average net migration. Similarly, municipalities surrounding Porto—such as Vila do Conde, Matosinhos, Maia, and Vila Nova de Gaia—as well as Aveiro, Oliveira do Bairro, and the district capitals municipalities Braga, Viseu, and Bragança also recorded positive average net migration. 23 Figure 6: Average net migration [2001-2021] Data source : INE 3.3 Local government policy action to attract population Municipalities facing population decline and economic stagnation have not remained passive. In response to these challenges, many have adopted measures aimed at attracting both new residents and businesses to their territories. One of the simplest instruments at their disposal is the setting of local tax rates. Adjustments to the Personal Income Tax and Local Property Tax rates can serve as incentives for individuals. In 2024, according to Associação Portuguesa de Famílias Numerosas (2025) 274 municipalities adopted a special Local Property Tax regime targeting families. This measure allows for a reduction in the tax amount based on the number of dependents: 30 euros for one dependent, 70 euros for two, and 140 euros for three.1 1https://www.apfn.com.pt/IMI2025.php 24 Further incentives are specifically directed at younger populations. For instance, in Arcos de Valdevez, young individuals purchasing a property intended solely as their permanent residence are granted exemptions from both the Local Property Tax and the Property Transfer Tax. The Local Property Tax exemption exemption is valid for an initial period of three years and may be renewed for an additional two years. 2 Some municipalities provide birth and adoption allowances, often conditional on the funds being spent within the local economy. Notable examples include Chaves, Alcobaça, Montemor-o-Velho and São João da Pesqueira , which have implemented these incentives as part of broader strategies to promote demographic sustainability.3 In the field of education, municipalities are offering a range of supports: annual school vouchers redeemable in the local economy, free or subsidized school transport and scholarships for students in public higher education, as well as merit-based awards for top-performing 12th grade students. 4 In addition to attracting new residents, fostering local economic development through business activity and investment has become a strategic priority. However, the structural characteristics of many interior Portuguese regions may pose significant challenges to private sector investment. To address these challenges, municipalities have implemented a range of incentives aimed at attracting business investment. These include reductions or exemptions in the local corporate income tax surcharge, often determined by factors such as business volume, economic activity sector, company size and jobs. As well as local property tax minimum rates, reduction or exemption. Additional support measures include the provision of infrastructures and industrial lots, microcredit programs for entrepreneurs, and the simplification of bureaucratic processes. 5 3.3.1 Municipality of São João da Pesqueira São João da Pesqueira is located in the interior of Portugal, in the northern part of the Viseu district, within the Douro Demarcated Region. Like many other interior municipalities, it has faced a consistent population decline over the past two decades. The selection of São João da Pesqueira is driven not only by personal affiliations but also by its historical and cultural significance. Recognized as the oldest municipality in Portugal, it is situated within 2https://www.cmav.pt/pages/2604 3https://www.chaves.pt/pages/966 https://portaldomunicipe.cm-alcobaca.pt/50697/incentivo-a-natalidade https://www.sjpesqueira.pt/p/apoioanatalidadeeadocao https://www.cm-montemorvelho.pt/index.php/ incentivo-a-natalidade 4https://www.sjpesqueira.pt/servicos/educacao https://www.cm-carrazedadeansiaes.pt/servicos/educacao/ bolsas-de-estudo https://cm-feira.pt/-/ano-letivo-2024-2025-vale-oferta-de-material-escolar 5https://www.sjpesqueira.pt/noticia-17/informacao-apoios-incentivos https://www.cm-nelas.pt/investir/apoios/ incentivos-do-municipio/ https://cm-proencanova.pt/1030/incentivos-ao-investimento 25 the culturally rich Douro wine region, which is presently experiencing a period of notable socioeconomic adversity. Similar to other interior municipalities, it is undergoing sustained demographic and economic decline, driven in part by its geographic isolation from major economic hubs and exacerbated by limited accessibility, due to underdeveloped road infrastructure and challenging topography. Nevertheless, tourism has been experiencing growth, as in other regions of Portugal. While it may offer an opportunity to counteract out-migration by creating employment, it remains uncertain whether it can ensure long-term demographic stability. São João da Pesqueira recorder negative net migration rates from 1991 to 2019. However, from 2019 onwards, this trend reversed, with the municipality registering positive, albeit modest, net migration rates (Figure 7). Despite this shift, the overall population trend remained negative: the population declined from 9581 in 1991 to 6772 in 2023. This decline is primarily driven by persistent negative demographic change, with deaths consistently outnumbering births, and by the long-term effects of outward migration. A breakdown of the population by age groups provides a more nuanced understanding of demographic dynamics. Between 1991 and 2023 the number of residents aged 65 and over increased steadily, while younger age groups, 14 years old and under experienced a sharp decline, reflecting a sustained drop in birth rates. The working-age population (25-64) remained relatively stable at around 4000 until 2011. However, from 2012 onwards, a downward trend emerged, likely associated with the effects of the sovereign debt crisis, stabilizing at approximately 3500 by 2019. The municipality experienced a significant decrease in its youngest age groups, driven by outmigration and persistently low birth rates. Meanwhile, those who remained have contributed to the growth of the older population segment. In 1991, the two youngest age groups accounted for 40% of the total population; by 2023 their share had fallen to 20%. Conversely, the proportion of the oldest age group increased from 16% in 1991 to 29% in 2023. This demographic shift puts at risk the municipality’s capacity to replenish its population, further exacerbating the overall decline. Economically, the municipality has traditionally relied on pprimary sector activities, particularly wine and olive oil production. However, employment in this sector has declined, reflecting the impact of mechanization. At the same time, employment in the tertiary sector has grown modestly, likely due to the growth of tourism and agricultural support industries, this growth has not been enough to compensate the job losses in the primary sector since 2001. Employment in the secondary sector has also experienced a slight decline over the same period. Over the last three decades, agricultural activities have faced increasing competition at both the global and European Union levels. Small family-owned businesses struggled to remain competitive and ultimately 26 Figure 7: Net migration (São João da Pesqueira) Data source : INE exited the market, contributing to the employment decline in the primary sector, as shown in Figure 9. This may have also been the case for other municipalities heavily reliant on agriculture, particularly in the context of globalization and market liberalization. On March 21st, a semi-structured interview was conducted via email with the current mayor of São João da Pesqueira, Manuel Natário Cordeiro. The interview was focused on the municipality’s strategies to stimulate economic activity and attract new residents. 1. Has the transfer of new powers to the municipality under Law 50/2018 been favourable? Has it provided the municipality with new tools to attract population? The mayor responded that Law 50/2018 has contributed to a greater proximity and interest in local development. He emphasized that, in the area of Education, although the municipality spends more than it receives from the central government, this commitment to education will eventually pay off. The areas of Health and Social Action have also benefited, with municipalities better positioned to improve and expand their services 2. Given the challenges posed by population decline in the interior of Portugal, what initiatives has the municipality promoted to retain young people and attract families? Have these initiatives been successful? The mayor emphasized that the municipality is committed to creating conditions that ensure the 27 Figure 8: Population by age groups (São João da Pesqueira) Data source : INE Figure 9: Employment by sector (São João da Pesqueira) Data source : INE 28 5 Empirical Framework 5.1 Estimation Strategy The estimation strategy consists of a panel data econometric analysis. A panel data set is characterised by repeated observations of the same units (Portuguese mainland municipalities) across time. Several advantages can be exploited when using panel data, one is avoiding/overcoming the bias caused by unobserved heterogeneity. Yit =β0+ k ∑ j=1 βjXjit +ϵit (10) This equation represents a panel data econometric model in which explanatory variables are contained in the vector Xjit for municipality i, year t, and jfor each variable. The coefficient βmeasures the effect of each variable on the dependent variable (Y), and the disturbance term εit. However, if Xjit variables are not successful enough in explaining Y there may be unobserved effects, usually added to the model as αi. To tackle unobserved effects, two methods can be used, Fixed Effects or Random Effects each method handles the unobserved effect differently. The fixed effects model allows for a correlation between αiand independent variables to exist, it can be applied using several methods. One of which is the Least Squares Dummy variable (LSDV) method. This assumes that the unobserved effect αiis a parameter to be estimated for each i, incorporating each individual unobserved effect into the model, by including a dummy variable for each i, to avoid the dummy variable trap the constant is dropped. Other methods, such as the First Differences or Within Groups are also used, the latter being the chosen fixed effects method by stata. On the other hand, the random effects model treats the unobserved effect αias being a random variable, relying firstly on the condition that unobserved variables are drawn from a given distribution. Allowing αito be included in a compound disturbance term uit =αi+εit. And the second condition is that αimust be uncorrelated with observed variables Xjit. To choose between random effects and fixed effects the Hausman test is used. If the null hypothesis 35 is rejected the key assumption of the random effects is violated (αimust be uncorrelated with observed variables Xjit) and fixed effects is the appropriate specification. If we fail to reject the null hypothesis, random and fixed effects are valid, but fixed effects will be inefficient. 5.2 Baseline model The baseline empirical model for net migration is presented in the following equation: nmrit =αsocioeconomicit−1+βlocalgovernmentit−1+γamenityit−1+µi+νt+εit (11) i= 1, . . . , 278; t= 2006, . . . , 20231 where nmr it is the net migration rate for municipality i in year t. Independent variables are distributed among three vectors for notation purposes and lagged one period due to endogeneity issues (RodríguezPose and Ketterer, 2012). Being the socioeconomic and local government vectors of most interest, and amenity as controls. The first vector including socioeconomic variables (economici,t−1), is composed of five independent variables, to account for the socioeconomic conditions of each municipality: 1. avgsalary : Average salary in each municipality, at 2022 prices; 2. UIbenef : Unemployment insurance beneficiaries to capture the local labour market impact; 3. SIIbenef : The beneficiaries of social integration income, to measure the poverty level of municipalities; 4. HEworkers : Percentage of workers with higher education, as these may contribute positively to the community; 5. mvhousing : Median value of housing at 2022 prices. The second vector (localgovernmenti,t−1) includes expenditure, debt and local taxes, that are within the reach of the municipality’s executive: 1. LPT : Local property tax; 2. pPIT : Local personal income tax participation; 3. TotalExp : Logged per capita total expenditure at 2022 prices; 1The period covered changes across specifications. 36 4. DebtMun : Logged per capita debt, at 2022 prices; 5. EnvExp : Logged per capita expenditure in environment protection, at 2022 prices; 6. CultExp : Logged per capita expenditure in culture, at 2022 prices. The third vector (amenityi,t−1), measures the appeal of municipalities from a non-pecuniary perspective, amenities which contribute to convenience and quality of life. 1. hotelguests : Places with more tourism activity may be more pleasant and appealing, while on the contrary they can contribute to decreasing housing supply, this variable measures the number of hotel guests in the percentage of the total population; 2. tcrim : Places with a higher level of crime activity may be less attractive. Finally, µiand νtaccount for municipal fixed effects and time fixed effects, and ϵit the error term. The average salary avgsalary , local property tax LPT , expenditure on environment protection EnvExp and culture CultExp , crime rate tcrim and median value of housing mvhousing turned out to have high VIF values, indicating the presence of multicollinearity. To address it, the decision was to drop these variables, and the logged average salary was replaced by annual average salary growth. The initial baseline model is estimated lagging independent variables by one year, additionally alternative models are estimated lagging independent variables by two and three years. To capture longer term effects, additional models are estimated with variables computed in five, four and three year period averages. Before proceeding to discuss empirical results, a analysis regarding the Hausman test was performed, both the baseline and period models presented a p-value lower than 0.05, leading to the rejection of the null hypothesis and indicating that fixed effects are the appropriate specification. 37 6 Empirical Results 6.1 Baseline model The results of the baseline model are present in Table 6. As expected, wage growth exhibits a positive coefficient but never achieving statistical significance. Local labour market conditions, specially the percentage of unemployment insurance beneficiaries, has a negative and significant impact on the dependent variable in most model specifications, in the fourth column a one percentage point increase in unemployment insurance beneficiaries is associated with a 0.0408 percentage point decrease of the net migration rate. The percentage of employed individuals with high education for the last three columns has the expected positive sign and a negative sign for the first column, never reaching statistical significance. Similarly, the percentage of beneficiaries of social integration income generally displays the expected negative sign across most specifications but fails to attain statistical significance. The percentage of hotel guests exhibits mixed results, with no statistical significance. Regarding local government variables, neither the personal income tax nor debt, both of which exhibit positive coefficients, demonstrate a significant effect on net migration. Likewise, total expenditure per capita which has a negative coefficient, does not achieve statistical significance. The attractiveness of a municipality may not be solely influenced by the prior year conditions, but those from two or three years ago. To account for this, additional models were estimated using two-year lagged and three-year lagged independent variables in the Tables 7 and 8. The negative and significant effect of labour market conditions persists ( UIbenef ) for the majority of the columns as in Table 6. Additionally poorer municipalities (with a higher percentage of SIIbenef ) are less attractive, significant at the 5% level in the first column of Table 7. Moreover, salary growth exhibits a positive and significant effect on net migration in the first, second and fourth column of Table 7 and the fourth column of Table 8 at the 5% significance level. The percentage of hotel guests has a negative and significant impact on the attractiveness of municipalities in the first and second column of Table 7, and a positive effect in the third column of Table 8. 38 Table 6: Baseline model regressions (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowtht−10.00354 0.00189 0.000620 0.00204 (0.00421) (0.00288) (0.00252) (0.00210) HEworkerst−1-0.0307 0.00466 0.00577 0.00980 (0.0284) (0.0219) (0.0196) (0.0176) UIbeneft−1-0.0450 -0.0473** -0.0511*** -0.0408*** (0.0281) (0.0184) (0.0151) (0.0150) SIIbeneft−1-0.0776 0.00838 -0.00983 -0.00602 (0.0484) (0.0244) (0.0207) (0.0172) hotelguestst−15.50e-05 -1.52e-05 8.33e-05 (0.000197) (0.000162) (0.000177) pPITt−10.0327 0.0236 (0.0219) (0.0169) logT otExpt−1-0.0198 (0.0881) logDebtMunt−10.0255 (0.0740) Constant 0.412 0.281 0.528*** 0.437*** (0.329) (0.205) (0.142) (0.130) Observations 1,865 2,725 3,483 4,443 R-squared 0.562 0.666 0.662 0.662 No. of municipalities 278 278 278 278 Adj. R20.558 0.663 0.660 0.661 Notes: All regressions include municipal and year-fixed effects, and cover 278 mainland Portugal municipalities. First column covers the period from 2015 to 2022, second column from 2009 to 2023, third and fourth column from 2007 to 2023. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1. 39 Table 7: Baseline model regressions two year lags (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowtht−20.00818** 0.00697** 0.00483* 0.00447** (0.00392) (0.00294) (0.00258) (0.00226) HEworkerst−2-0.000744 0.00396 0.00364 0.00197 (0.0272) (0.0257) (0.0225) (0.0197) UIbeneft−2-0.0381 -0.0377** -0.0484*** -0.0401*** (0.0253) (0.0174) (0.0127) (0.0133) SIIbeneft−2-0.109** 0.00717 -0.0104 -0.000523 (0.0476) (0.0235) (0.0193) (0.0158) hotelguestst−2-0.000953*** -0.000420** -0.000192 (0.000228) (0.000202) (0.000248) pPITt−20.0290 0.0319* (0.0227) (0.0175) logT otExpt−20.0548 (0.101) logDebtMunt−2-0.0610 (0.0832) Constant 0.470 0.112 0.416*** 0.297** (0.300) (0.203) (0.137) (0.121) Observations 1,865 2,466 3,205 4,165 R-squared 0.581 0.657 0.664 0.670 No. of municipalities 278 278 278 278 Adj. R20.577 0.655 0.662 0.669 Notes: All regressions include municipal and year-fixed effects. All models cover 278 mainland Portugal municipalities except the fourth. First column covers the period from 2016 to 2023, second column from 2010 to 2023, third and fourth column from 2008 to 2023. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1. 40 Table 8: Baseline model regressions three year lags (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowtht−30.00618 0.00608 0.00398 0.00569** (0.00536) (0.00376) (0.00293) (0.00257) HEworkerst−30.0248 -0.00421 -0.00115 -0.00724 (0.0247) (0.0252) (0.0214) (0.0197) UIbeneft−3-0.0620** -0.0347* -0.0471*** -0.0366*** (0.0265) (0.0181) (0.0114) (0.0126) SIIbeneft−3-0.0753 0.0298 0.00612 0.0141 (0.0550) (0.0245) (0.0185) (0.0144) hotelguestst−3-4.30e-07 0.000189 0.000275** (0.000109) (0.000121) (0.000132) pPITt−30.0109 0.0205 (0.0294) (0.0183) logT otExpt−30.156* (0.0836) logDebtMunt−3-0.0556 (0.0956) Constant 0.621** -0.260 0.377*** 0.294*** (0.308) (0.193) (0.116) (0.110) Observations 1,629 2,211 2,934 3,888 R-squared 0.514 0.657 0.670 0.682 No. of municipalities 278 278 278 278 Adj. R20.510 0.655 0.668 0.681 Notes: All regressions include municipal and year-fixed effects. All models cover 278 mainland Portugal municipalities. First column covers the period from 2017 to 2023, second column from 2011 to 2023, third and fourth column from 2009 to 2023. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1. 41 Overall, strong economic prospects in the short term (1 to 3 years prior), such as wage growth enhance the attractiveness of municipalities. However the effect is only significant when considering second and third-order lags, suggesting that individuals may not base migration decisions on immediate wage growth changes as they may not not know them, but rather on wage growth from two or three years prior. Labour market conditions are also a consistent predictor, always presenting a negative coefficient, while hotel guests exhibit mixed results. In short it is possible to conclude that net migration is not solely influenced by the previous year conditions, but shaped by factors over at least the past three years. However, not every variable reached a significant result, this may indicate that the short term is not enough to predict the net migration of just a single year. The process of migration is continuous and information that individuals consider may not only be based on the preceding year (1st 2nd or 3rd order lags), but on an average of previous years, that captures a wider view of the development of a place. An increase in the number of unemployed may not only affect the next year net migration but the following years, the same applies for salary growth that may not only contribute to increasing net migration for a single year but several. Hence additional regressions were estimated using five, four and three year periods with averaged variables. 6.2 Five-year period model In order to capture long run effects of independent variables on net migration, fixed effects models with 5 year-period averaged variables were estimated. The effects of salary growth are in line with the previous results, having a positive coefficient for all the columns, but failing to achieve statistical significance at least at the 5% level. The percentage of unemployment insurance beneficiaries, has a negative coefficient significant at the 1% and 5% level, for the last three columns. The percentage of hotel guests has a positive and significant coefficient and lastly total expenditures of municipalities has a negative and significant coefficient. This may possibly mean that on a longer term perspective municipalities with better job market conditions, and amenities are more attractive. While those where the local government spends more are less attractive, as the higher spending may require higher local taxes. In addition to the previous, additional regressions were estimated using 4 and 3 year periods in the appendix tables B.1 and B.2. For the four year period models, the percentage of unemployment insurance beneficiaries keeps the negative coefficient and significance for all columns, salary growth has a positive coefficient, but never reaching statistical significance. In the case of the percentage of hotel guests, presenting a positive and significant coefficient for all the columns. The logged municipal total expenditure, presents a negative and significant coefficient, lower than the five year period model. Contrary to the 42 Table 9: Five-year period regressions (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowth 0.0100 0.0131 0.0158 0.0223* (0.0155) (0.0131) (0.0118) (0.0118) Heworkers -0.0477* -0.01000 -0.0145 -0.00978 (0.0285) (0.0218) (0.0208) (0.0200) UIbenef -0.0547* -0.0738*** -0.0542** -0.0462** (0.0309) (0.0219) (0.0226) (0.0201) SIIbenef 0.00990 0.0248 0.00634 0.00704 (0.0379) (0.0242) (0.0241) (0.0220) hotelguests 0.000400** 0.000451*** 0.000424*** (0.000175) (0.000170) (0.000155) pPIT 0.0203 0.00539 (0.0310) (0.0221) logT otExp -0.617*** (0.235) logDebtMun 0.0360 (0.104) Constant 0.224 0.494** 0.422** 0.370** (0.330) (0.219) (0.187) (0.164) Observations 718 877 971 1,112 R-squared 0.801 0.751 0.742 0.735 No. of municipalities 278 278 278 278 Adj. R20.798 0.749 0.740 0.734 Notes: All regressions include municipal and period-fixed effects, and cover 278 mainland Portugal municipalities. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1. 43 expected the percentage of workers with high education presents a negative coefficient significant at the 5% level for the first column. Despite being in line with the previous tables, table B.2 only achieved significant results for the unemployment insurance beneficiaries (second and third column), the percentage of hotel guests for all columns, and the percentage of highly educated workers, again having a negative and significant coefficient in the first column. The labour market variable, unemployment insurance beneficiaries, was the most consistent across the different periods used. Similarly, the percentage of hotel guests, hotel activity seems to contribute positively to the attractiveness of municipalities, in a longer term perspective, contrary with what was found in some of the baseline lagged models. The municipal total expenditure has a negative and significant coefficient for the 2 longest periods considered. Which may support the possibility that high spending may lead to higher local taxes, making the municipality less attractive, or that municipalities that are facing population decline are increasing expenses to counteract it. 44 A Appendix - Figures Figure A.1: Municipal primary sector employment variation 2001-2021 51 B Appendix - Tables Table B.1: Four-year period regressions (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowth 0.0192 0.00708 0.0158 0.0159* (0.0132) (0.0133) (0.0101) (0.00878) Heworkers -0.0571** -0.0211 -0.00896 0.00150 (0.0270) (0.0223) (0.0185) (0.0180) UIbenef -0.0620** -0.0698*** -0.0531*** -0.0387** (0.0281) (0.0222) (0.0159) (0.0168) SIIbenef 0.0178 0.0221 0.00118 -0.00789 (0.0427) (0.0246) (0.0202) (0.0189) hotelguests 0.000412** 0.000455*** 0.000424*** (0.000202) (0.000165) (0.000163) pPIT 0.0193 0.00338 (0.0279) (0.0221) logT otExp -0.454** (0.202) logDebtMun 0.0247 (0.0992) Constant 0.254 0.370 0.454*** 0.396*** (0.302) (0.234) (0.140) (0.139) Observations 806 974 1,222 1,390 R-squared 0.813 0.779 0.733 0.710 No. of municipalities 278 278 278 278 Adj. R20.810 0.777 0.731 0.708 Notes: All regressions include municipal and period-fixed effects, and cover 278 mainland Portugal municipalities. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1. 52 Table B.2: Three-year period regressions (1) (2) (3) (4) Variables nmr nmr nmr nmr salarygrowth -0.000350 -0.000739 0.00387 0.00778 (0.0111) (0.00641) (0.00538) (0.00561) Heworkers -0.0859** 0.00976 0.00829 0.0160 (0.0384) (0.0180) (0.0166) (0.0166) UIbenef -0.0471* -0.0585** -0.0421** -0.0290* (0.0283) (0.0236) (0.0179) (0.0170) SIIbenef 0.00353 0.0253 0.00628 -0.00307 (0.0473) (0.0206) (0.0185) (0.0170) hotelguests 0.000655*** 0.000408** 0.000372** (0.000215) (0.000168) (0.000144) pPIT -0.0111 0.00623 (0.0255) (0.0171) logT otExp -0.128 (0.154) logDebtMun 0.0477 (0.0894) Constant 0.364 0.323 0.335** 0.282** (0.346) (0.203) (0.148) (0.136) Observations 810 1,377 1,681 1,946 R-squared 0.781 0.786 0.755 0.736 No. of municipalities 278 278 278 278 Adj. R20.778 0.785 0.753 0.735 Notes: All regressions include municipal and period-fixed effects, and cover 278 mainland Portugal municipalities. Robust standard errors clustered by municipality in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1.