Do welfare benefits deter migration? Jordi Jofre-Monseny ∗ Universitat de Barcelona & Institut d’Economia de Barcelona (IEB) May 2012 Abstract: The aim of this paper is to estimate the effects of welfare benefits on migration by using a national policy providing higher unemployment protection in the lagging Spanish regions of Andalucía and Extremadura. The regional implementation of the policy generates a geographic discontinuity in the treatment status of municipalities. Focusing on municipalities that are sufficiently close to (but at different sides of) the border enables us to identify the effect of unemployment benefits on population growth, out- and in-migration rates and on unemployment. Our results indicate that higher unemployment benefits reduced the rate at which rural municipalities lost population between 1981 and 1991, mainly through a reduction in out-migration. We also find that in-migration and unemployment rates are higher on the welfare ‘generous’ side of the border. The effects found are large and mostly driven by the behavior of the socioeconomic group with the lowest educational level. Key words: Welfare benefits, migration, unemployment. JEL Codes: H73, J6, R23. Contact Address: Dpt. d’Economia Pública, Economia Política i Economia Espanyola, Diagonal 690, Torre 4, Planta 2, (08034) Barcelona, e-mail: jordi.jofre@ub.edu ∗ I thank Gilles Duranton and Matthew Turner for useful comments as well as participants at the Urban and Real State Day Workshop held in April 2012 at the Department of Economics of the University of Toronto. Funding from grants ECO2010-16934 and 2009SGR102 is gratefully acknowledged. 1 1. Introduction There is ample evidence that one unintended consequence of unemployment benefits is increased unemployment. For instance, Katz and Meyer (1990) and Bover et al (2002) document a spike in the probability to leave unemployment at the time unemployment benefits are about to expire for the US and Spain, respectively. Card and Levine (2000) conclude that a temporary extension of unemployment benefits in New Jersey caused longer spells of unemployment. Unemployment benefits might distort migration decisions, too. First, they can limit the net gains of migrating to areas with better labor market conditions. Second, geographical variation in welfare generosity can attract welfare-prone individuals to areas offering higher benefits. In this paper, we analyze the effects on migration of a national policy providing more generous unemployment benefits in the Spanish regions of Andalucía and Extremadura. The regional implementation of the policy, coupled with data on mobility at the municipality level, enables us to credibly estimate the effects of welfare benefits on mobility. A substantial literature, including Levine and Zimmerman (1999), Gelbach (2004) and McKinnish (2005, 2007), has studied the mobility effects of US interstate differentials in welfare generosity. ADFC (Aid to Families with Dependant Children) is the main program providing variation in welfare generosity across US states. Since this program is aimed at single poor mothers with children, this literature deals with female mobility. Levine and Zimmerman (1999) use the National Longitudinal Survey of Youth to regress the individual probability to move-out from a State as function of the State’s ADFC generosity. This paper adopts a ‘differences-indifferences’ approach by comparing poor single mothers with children to poor single mothers without children. The results found are economically small and statistically insignificant. With Census data from the 1980 and 1990, Gelbach (2004) also performs a ‘differences-in-differences’ analysis by comparing single mothers that are high-school dropouts to various suitable control groups, noting that migration gains in terms of welfare payments are a function children’s age. For 1980 data, the author finds that the probability to move out decreases with welfare generosity more for the treated than for the control group. The results for 1990 are less conclusive. McKinnish (2005) focuses on short-distance moves at US State borders. If geographic distance reflects mobility costs, out-migration from ‘low’ ADFC States should be higher in border than in interior counties. This pattern is confirmed by the data although the implied effects are small. In a follow-up paper, McKinnish (2007) combines this ‘border’ approach with the ‘differences-indifferences’ analysis pursued in Gelbach (2004). The results of this ‘triple-difference’ model confirm that ADFC causes a (modest) reduction in out-migration. 2 Non-US evidence on the effects of welfare benefits on migration includes Edmark (2009), Fiva (2009) and De Giorgi and Pellizzari (2009). Edmark (2009) studies the introduction of stricter rules to receive welfare benefits in Stockholm, finding no effects on out-migration rates. Fiva (2009) tests if Norwegian municipalities offering more generous welfare benefits attract more welfare recipients. He finds a large effect and that accounting for policy endogeneity is important. Finally, De Giorgi and Pellizzari (2009) study the effect of welfare generosity as a factor determining the location decisions of migrants across EU countries. Their results indicate that, conditional on moving, migrants choose countries offering more generous benefits. To our knowledge, Antolín and Bover (1997) is the only paper studying the role of unemployment benefits on migration in Spain. Using micro data from the Labor Force Survey, they find that registered unemployed are less likely to migrate than non-registered ones. Since registered unemployed are more likely to receive benefits, this result suggests that unemployment benefits might deter migration. In 1984, a more generous welfare benefit was introduced by the national government in the lagging and southern regions of Andalucía and Extremadura. This benefit is important both in terms of recipients (8 percent of the labor force in these two regions) and in terms of the welfare generosity change it generates at the border. This paper estimates the effects of this policy on mobility by adopting a ‘border’ identification strategy similar to that developed in Holmes (1998) and Black (1999). The regional implementation of the policy generates a geographic discontinuity in the treatment status of municipalities. Focusing on municipalities that are sufficiently close to (but at different sides of) the border enables us to identify the effect of unemployment benefits on population growth, out- and in-migration rates and on unemployment. Our main findings can be summarized as follows. The program increased population growth by 5 percent between the census years of 1981 and 1991. This population growth effect is explained by both a lower out-migration rate (4 percent) and a higher in-migration rate (1 percent). The effects of the policy on out- and in-migration are stronger for the demographic group with the lowest attained educational level and, thus, the highest probability to be on welfare. In 1991, unemployment was 15 percentage points higher on the welfare ‘generous’ side of the border. Methodologically, this paper is similar to McKinnish (2005, 2007) in that it uses the fact that welfare generosity varies discontinuously at regional borders to (credibly) estimate the effects of interest. However, due to the nature of the ADFC program, McKinnish (2005, 2007) - and the rest of US papers- study the effects of welfare benefits on the mobility of poor single mothers. 3 Here, we can analyze the effects of a more general welfare program that, in contrast to ADFC, can be accumulated over the entire lifecycle. Another attractive feature of this paper is that we estimate the effects of welfare benefits on mobility in a rural environment experiencing intense out-migration and high unemployment. After this introduction, the rest of the paper is organized as follows. In section 2 we describe the institutional details of the analyzed policy. Section 3 introduces the econometric specification and section 4 the data used. The validity of the empirical design is addressed in section 5 and the results are presented and discusses in section 6. Section 7 concludes. 2. Institutional background 2.1. The SIPTEA program In Spain, there is an agrarian unemployment benefit (Subsidio agrario) that is only available to the residents of Andalucía and Extremadura. This benefit is part of a broader national policy that aims to protect temporary agricultural workers of these two southern regions. Its name is SIPTEA (Sistema Integrado de Protección de los Trabajadores Eventuales Agrarios) and was implemented in 1984. Besides the unemployment benefit, it has two other programs. The first one hires local population in (minor) infrastructure projects (Plan de Empleo Rural). The second one offers short training courses (between 60 and 90 days) with the objective of providing agricultural workers with the necessary skills to become employable in other sectors (Formación Ocupacional Rural). Map 1 shows the location of Andalucía and Extremadura within Spain. The bordering regions are Castilla-León, Castilla-la Mancha and Murcia. [Insert Map 1 here] The agrarian unemployment benefit (AUB) is only available to workers that, living in a municipality of Andalucía and Extremadura, are in the Special Agrarian Regime of the Spanish Social Security. Between 1984 and 1990, workers being employed a minimum of 60 days in the agriculture in a given year were entitled to the subsidy the following year, provided their total income was below the minimum wage1. The yearly subsidy was 4.5 months of the national minimum wage. In 1985, a pay of 3.3 months of the minimum wage was introduced for workers employed between 10 and 60 days. The AUB was reformed in 1990. From 1991 onwards, the personal low income requirement (income below minimum wage) was replaced by a household Days worked in the agriculture elsewhere in Spain or abroad are eligible to achieve this minimum of 60 days. 1 4 one (mean household income below minimum wage). At the same time, the pay became agedependant with more generous benefits for older individuals. For the oldest individuals in the system, those aged 60 or more, the subsidy was as high as 7.5 months of the minimum wage. The Plan de Empleo Rural (PER) program consists of ear-marked grants flowing from the Spanish Unemployment Office to municipalities in rural areas of Andalucía and Extremadura with high unemployment rates. In 1987, there were 237,700 jobs in this program. The length of these jobs was typically below one month and there were 30 jobs in the average infrastructure project. The worked days in these projects are eligible to achieve the minimum days of work to access the AUB. In practice, the main objective of the PER projects has been to entitle its workers to the AUB (González, 1990). In this context, the large autonomy of municipal governments in hiring decisions in PER projects might have led to political clientelism behaviours (Cansino, 2000). SIPTEA appeared in 1984 as a reform of the Empleo Comunitario (EC) program in place since 1971. This program hired workers in infrastructure projects as a means to decrease unemployment in the rural areas of Andalucía and Extremadura2. However, these projects were often fictitious, implying that, de facto, the program was partly an unemployment benefit (González, 1990). As a result, the 1984 reform creates an unemployment benefit (i.e. the AUB) separately from a program hiring workers in infrastructure projects (PER). The solid line in Graph 1 shows the evolution over time in (1) the number of ‘employees’ under the EC scheme between 1979 and 1983 and (2) the number of recipients of the UAB between 1984 and 1991. The number of ‘employees’ in the EC scheme increased steadily during the 70s to explode after the Socialist party took office in 1982, going from 53 thousand in 1981 to 159 thousand in 1983. In 1984, the first year of the AUB, there were 192,304 recipients and this number increased to 236,327 in 1991, representing 8 percent of the labor force in Andalucía and Extremadura. [Insert Graph 1 here] Given that (1) a reform extended the PER program to parts of other Spanish regions in 1996 and that (2) decennial census data prior to 1981 are not available at the municipal level, we will use data from 1981 and 1991 to analyze the effects of the described program on mobility and unemployment. Although we are aware that 1981 is not the ideal point in time to start the analysis (EC was in place since 1971), notice that the quantitative importance of the policy increased dramatically between 1981 and 1991 (from 53 thousand EC ‘employees’ to 236 2 A tiny fraction of the EC funds were directed to the region of Castilla-la Mancha. 5 thousand AUB recipients). Notice that the requirement to become an AUB recipient (i.e. to work a given number of days either in the agriculture or in a PER project) are easier to meet in rural than in urban areas. Hence, the SIPTEA policy might have increased the attractiveness of staying in the rural areas of Andalucía and Extremadura vis-à-vis migrating to urban areas. 2.2. Unemployment benefits (elsewhere) in Spain As in most European countries, in Spain there are two types of unemployment benefits: Unemployment insurance and unemployment assistance3. Unemployment insurance is more generous and is more directly linked to each worker past contributions. As for unemployment assistance, there are two programs: The AUB described above (only available to the residents of Andalucía and Extremadura) and the general unemployment benefit (GUB). In 1991 a worker with tenure exceeding 6 months who was fired was entitled unemployment insurance for half of the tenure period (with a maximum of 2 years). Pay was proportional to past wages (with caps) and the proportion decreased over time from 80 to 60 percent. As for the GUB, there were typically two instances in which a fired worker was entitled this less generous form of benefit. The first instance arose with a fired worker with tenure between 3 and 6 months. In that case, the unemployed worker received 75 percent of the minimum wage for the tenure period. This implies that a worker with tenure between 3 and 6 months perceived benefits amounting from 2.25 to 4.5 months of the minimum wage. The second instance where a worker was entitled unemployment assistance arose with the exhaustion of unemployment insurance. This was only the case for workers with 45 years of age or more with family dependents. The length of these benefits ranged from 18 to 36 months and the pay was 75/100/125 percent of the minimum wage for individuals with 1/2/3 dependents. For the period we study, the AUB was more generous than the GUB. Whereas within the AUB scheme a worker needed 60 days of work to be entitled to 4.5 months of the minimum wage, a worker within the GUB needed 6 months of work to receive that same amount. Similarly, whereas a worker within the GUB needed a minimum of 3 months of work to perceive some benefit (2.25 months of the minimum wage), a worker within the AUB only needed 10 days of work to perceive a slightly larger amount (2.5 months of the minimum wage). Besides, the days worked in the numerous PER projects have been eligible to achieve the minimum days of work that provide access to AUB. 3 For details on the reforms on unemployment benefits in Spain between 1987 and 1994, see Bover et al (2002). 6 Table 1 shows the number of unemployment assistance recipients for Andalucía, Extremadura and their neighboring regions in 2001. The first two columns report the number of recipients of the GUB and the AUB schemes in each region. The second and third columns show number of recipients as a fraction of the unemployed. Finally, column five shows the share of unemployed that receive some form of unemployment assistance. [Insert Table 1] The fraction of unemployed that receive the GUB is relatively similar across the regions (between 0.12 and 0.21). These figures are smaller than 0.28, which is the share of unemployed in Andalucía and Extremadura that receive the AUB. As a result, the percentage of unemployed that receive some form of unemployment assistance is 40 and 42 percent in Andalucía and Extremadura versus 12, 21 and 15 percent in the regions of Castilla-León, Castilla-la Mancha and Murcia respectively. Hence, relative to the rest of Spain, unemployment protection seems to have been higher in Andalucía and Extremadura. There are reasons to believe that a significant share of the recipients of the AUB were not genuine agricultural workers. The dashed and dotted lines in Graph 1 show the evolution over time in the number of recipients of the AUB between 1984 and 1991. While the number of male recipients decreased by 32 percent the number of females almost quintupled. In fact, as of 1991, there are more females than males in the system. This change in the gender profile of the recipient does not coincide with a change in the profile of the agricultural worker in the regions of Andalucía and Extremadura. The ratio of male to female workers in the agriculture in these two regions was around 7 both in 1984 and in 1991. Furthermore, the number of recipients of the AUB has systematically exceeded the Labor Force Survey figure for the number of unemployed agricultural workers in Andalucía and Extremadura (García-Pérez, 2004). In 1991, the numbers of unemployed agricultural workers was 159,850 versus 236,327 AUB recipients4. 3. Econometric specification There are two main outcomes of interest in this paper. The first one is the population growth rate (or related measures like out- and in-migration rates) for the 1981-91 period. The second one is the municipal unemployment rate in 1991. The baseline econometric specifications are: y i 81− 91 = β ⋅ SIPTEAi + x ′i 81δ + ε i 81− 91 4 (1) This figure is the number of unemployed in the agriculture sector averaged across the four trimesters of 1991. 7 u i 91 = φ ⋅ SIPTEAi + x ′i 81λ + υi 91 (2) where yi81-91 is the population growth rate between 1981 and 1991 in municipality i. The explanatory variable of interest, SIPTEAi, is a variable that indicates if the policy aimed at protecting temporary agriculture workers was in place in municipality i. Since this national policy was only implemented in the regions of Andalucía and Extremadura, the variable SIPTEAi takes the value of one if the municipality belongs to any of these two southern regions. The error term ε i 81− 91 reflects population growth shocks over the decade spanning 1981-91. In its turn, ui91 is the unemployment rate in municipality i in 1991 whereas υi 91 is an error term capturing shocks in this variable. In some specifications, we will introduce a vector of control variables, x ′i 81 , containing municipality characteristics in 1981 that could affect subsequent population and unemployment rates. The explicit aim of the SIPTEA program was to increase unemployment protection in the regions where seasonal unemployment in the agriculture was more prevalent5. In 1981, the overall and agricultural unemployment rates in Andalucía and Extremadura were the highest in Spain: 19.2 and 15.7 in Andalucía and 16.6 and 9.4 in Extremadura6. The fact that SIPTEA was introduced in the regions with more deteriorated labor market conditions suggests that the estimates of β and φ will generally suffer from a policy endogeneity bias. We address this identification problem by adopting a ‘border’ strategy similar to that developed in Holmes (1998) and Black (1999). In particular, we will use municipality level data coupled with the regional implementation of the policy to try to correctly identify the effects of interest. In other words, the implementation of the policy at pre-existing regional borders creates a sharp geographic discontinuity in the SIPTEA treatment status of municipalities. This identifying strategy is implemented by estimating equations (1) and (2) with municipalities that are sufficiently close to the SIPTEA regional border. That is: y i 81− 91 = β ⋅ SIPTEAi + x ′i 81δ + ε i 81− 91 with db i ≤ db (3) u i 91 = φ ⋅ SIPTEAi + x ′i 81λ + υi 91 (4) with dbi ≤ db 5 Card and Levine (2000) and Lalive and Zweimüller (2004) note that a deteriorated labor market is a factor associated with increases in unemployment protection in US States and Austrian regions, respectively. 6 These two unemployment figures were 13.4 and 5.4 for the Spanish economy, 9.7 and 0.8 for CastillaLeón, 13.5 and 6.9 for Castilla-la Mancha and 12.5 and 7.7 for Murcia. 8 where dbi is the distance from municipality i to the SIPTEA border and db is a threshold distance that we set alternatively at 15, 20 and 25 Km. The distance from municipality i to the SIPTEA border is computed as the air distance between the centre of municipality i and the closest municipality centre at the other side of the border7. The empirical specifications that we use can be interpreted within the regression discontinuity design framework8. The discontinuity is a sharp one since the probability that a municipality is treated changes at the border with a probability of one. Equations (3) and (4) can be seen as parametric specifications with a bandwidth of 15, 20 or 25 Km and a zero-order polynomial in the distance to the SIPTEA border (the forcing variable)9. 4. Data We mostly use data from the 1991 Population and Housing Census. This census contains rich information at the individual level including age, gender, municipality of residence in 1991 and 1981, attained educational level and labor market status (employed, unemployed, inactive…). Unfortunately, the data at the individual level for the municipalities that are relevant to us are not available to researchers10. However, we have data on relevant municipality characteristics such as the number of people by age groups, by gender, by labor market status and by attained educational level. We also know the number of individuals in 1991 that already lived in the municipality in 1981 and the number of in-migrants arriving between 1981 and 1991 by region of origin. Furthermore, we know the number of individuals in each of these migration categories by labor market status and by attained educational level. We also have access (and use) some labor market status data from the 1981 Population and Housing Census. We merge these data on socio-economic characteristics of the population with non-agriculture employment data from the 7 An alternative to air distance is distance by road. This is a better proxy for transport or mobility costs between two municipalities. However, in this application, we are interested in municipalities that are close in terms of confounding unobservables that determine unemployment and out-migration in rural areas. In this respect, plain geographic distance seems a more natural metric. 8 See Imbens and Lemieux (2008) and Lee and Lemieux (2010) for guides to practice in regression discontinuity designs. 9 Alternatively one could set a broader bandwidth (e.g. 100 km) and a higher order polynomial in the distance to the border. However, notice that there is a limit in how close two municipalities can be, meaning that the density of municipalities goes to zero as the distance to the border approaches zero. In this context, it seems more conservative to compare mean differences for observations that are reasonably close to the border. 10 There are public-use microdata files containing 10% of the population in each municipality exceeding 20 thousand inhabitants and 10% of the population in each province that lives in municipalities below 20 thousand inhabitants. Unfortunately, the municipality of residence is not reported for the individuals living in municipalities below 20 thousand inhabitants. 9 1980 Censuses of Establishments and with agriculture employment from the 1982 Agrarian Census. Setting a bandwidth of 15, 20 and 25 km in equations (3) and (4) leaves us with an estimation sample of 184, 236 and 306 ‘border’ municipalities. In Table 2 we provide summary statistics for the 236 municipalities that lie within 20 km from the relevant border. The SIPTEA program was implemented in 40 percent of these 236 municipalities. These are relatively small municipalities with a mean of 1,943 and a median of 754 inhabitants11. On average, municipalities in our sample experience substantial population losses, with an average negative population growth rate of 15.5 percent over the 1981-91 decade. This number is not too different from -12.7 percent, the average municipal population growth in the regions found along the SIPTEA implementation border (i.e. Andalucía, Extremadura, Castilla-León, Castilla-la Mancha and Murcia). This (negative) average population growth at the municipal level contrasts with a positive aggregate growth (3.1 percent) over the same period. These two figures can be rationalized by intense intraregional migration flows from rural to urban municipalities12. The aggregate population growth for the municipalities in the top population quartile in 1981 was 5.1 percent between 1981 and 1991. In contrast, the aggregate population growth for the municipalities in the remaining three quartiles was a negative 9.2 percent. Besides loosing population over the 80s, the average municipality in the sample is one where the 1991 unemployment rate is high (24.3 percent) and the level of education is low. In that same year, only 3.7 percent of the 16 to 64 year-olds had tertiary education whereas 34 percent had less than 5 years of education. [Insert Table 2] 5. Identifying assumptions and validity of the empirical design The (17) current Spanish regions (Comunidades Autónomas) were created between 1979 and 1981, just after the Constitution was passed in 1978. These regions were constructed as aggregations of provinces dating back to 1833. Therefore, regional borders, which in turn determine the geographical implementation of SIPTEA, could not have been defined to reflect current economic conditions. Nevertheless, borders in 1833 might have or might have not been placed at random. The identifying assumption is that, controlling for observables, any confounding For Spain, these figures were 5,013 and 600 in 1991. The difference between the mean and the median is substantially larger for Spain than for the 199 municipalities in our sample. This is probably related to the fact that there is no large city in the used sample of southern municipalities (the largest municipality has 20,595 inhabitants). 12 See Bover and Arellano (2002) for a study on the personal and location determinants of the Spanish intraregional migrations over the 80s and 90s. 11 10 unobservable evolves smoothly (and not discontinuously) at the border. An indirect way to validate this empirical design is to examine if relevant confounding observables evolve continuously or discontinuously at the border. We will examine two types of variables. On the one hand, variables reflecting labor market conditions with the potential of affecting subsequent population growth and unemployment rates. On the other hand, features reflecting the attractiveness of a municipality in terms of consumption opportunities in a rural environment. Ideally, we would like these two groups of variables to be pre-determined and, thus, be measured before 1971 (the year Empleo Comunitario was introduced). Unfortunately, for most of the variables that are relevant to us, the first data available are drawn from different Censuses carried out around 1981. With this caveat in mind and given that the program expanded dramatically between 1981 and 1991, we think it still is an interesting exercise to check if potentially confounding observables evolved continuously or not by 1981. In order to check if physical conditions for the agriculture change discontinuously at the border, the evolution of the ruggedness of terrain is shown in panel A of Graph 2. The horizontal axis is the distance to the border where treated municipalities are assigned positive values. The vertical axis is the (logged) ruggedness index constructed by Goerlich-Gisbert and Cantarino-Martí (2010). Dots are averages for rank percentile bins (representing 39/40 observations) whereas solid lines are local linear regressions fits estimated separately at either side of the border. For presentation purposes, only data for municipalities that belong to provinces that are contiguous to the SIPTEA border are shown (see panel B in Map 1). Municipalities that are close to the border have a more rugged geography than those located further away from it, suggesting that provincial borders in 1833 were not set at random. Nevertheless, there is no evidence that the physical conditions for the agriculture change discontinuously at the border. [Insert Graph 2] In panel B we directly examine the evolution of agriculture employment. To measure it, use data from the 1982 Agrarian Census and divide the number of full-time equivalent jobs in the agriculture (including owners and relatives worked day) over the 1981 population level. The importance of agriculture increases as the border is approached but no clear sign of a discontinuity is observed. In panels C and D we examine the behavior of industry and services employment respectively. Both variables, constructed using the 1980 Census of Establishments, 11 are expressed relative to the 1981 population level. Although services employment seems to increase at the border, no clear evidence of a large discontinuity in these variables is observed. As described above, the unemployment rate in the regions of Andalucía and Extremadura was the highest in the country in 1981. In Panel E we use unemployment data at the municipal level from the 1981 Population and Housing Census. On average, unemployment is higher in the treated municipalities but the difference shrinks as one move closer to the SIPTEA border. In fact, the bin averages suggest a smooth evolution of the unemployment rate at the border. Hence, Empleo Comunitario did not seem to have significantly increased unemployment by 1981. Turning to municipality features capturing consumption opportunities, in Panel F we examine the number of service establishments to reflect the number of bars, restaurants, hotels, shops, hairdressers and bank branches in the municipality. In panels G and H, we show the number of health and education centers, respectively. The data does not show a clear discontinuity in any of these variables, either. 6. Results A graphical preview of the main results is presented in Graph 3. The Population growth over the 1981-91 period is shown in panel A whereas panels B and C illustrate the out- and in-migration results. Out-migration is proxied by the probability to stay, defined as the number of individuals aged 16 to 64 in the municipality in 1991 that lived in the same municipality in 1981, divided by the population in 1981. The in-migration rate is defined as the number of individuals aged 16 to 64 in the municipality in 1991 that lived in another municipality in 1981, divided by the population in 1981. Finally, the unemployment rate in 1991 is plotted in panel D. [Insert Graph 3 here] The municipal population growth rate over the 1981-91 decade seems to evolve discontinuously at the SIPTEA border. Although, on average, municipalities experienced a negative population growth at both sides of the border, population losses were lower in the treated side (around 15 percent) in comparison to the untreated side (around 20 percent). By inspecting panels B and C, both a lower out-migration and a higher in-migration seem to explain the smaller population losses on the treated side of the border. In turn, unemployment ‘jumps’ at the SIPTEA border with the unemployment rate being around 35 and 20 percent on the treated and untreated sides, respectively. Notice that the unemployment rate discontinuity at the SIPTEA 12 border in 1991 contrasts with the smooth evolution of this variable at the border in 1981 (see panel E in Graph 2). The corresponding (baseline) regression results on population growth, probability to stay, in-migration and unemployment rates are reported in Table 3. Columns 1 to 2, 3 to 4 and 5 to 6 show the results with municipalities within 15, 20 and 25 km from the border. Columns 1, 3 and 5 only include region-pairs fixed effects as controls (i.e. Extremadura×Castilla-León, Extremadura×Castilla-la Mancha, Andalucía×Castilla-La Mancha and Andalucía×Murcia dummies). Columns 2, 4 and 6 further include the 1981 municipality characteristics examined in section 5 that might determine subsequent population growth and unemployment rates. Given that these variables were shown to evolve smoothly at the border, their inclusion should leave the estimated effects of interest unaltered. This is indeed the case. [Insert Table 3 here] Starting with the results on population growth (panel A), the estimates reported in column 3 (municipalities within 20 km from the border and no control variables) imply that SIPTEA increased population growth by 5 percent, which amounts to one third of the variable’s mean (-15.5 percent). The graphical analysis suggests that both lower out-migration and higher in-migration contributed to the observed discontinuity in population growth. The regression results reported in panels B and C confirm this finding although assign a more important role to the out-migration mechanism. If the SIPTEA policy brought about 1.3 more in-migrants per 100 inhabitants in 1981, it increased by 4 the number of 16 to 64 year-olds staying in the municipality per 100 inhabitants in 1981. As for unemployment, the regression results reported in panel D confirm that SIPTEA increased the unemployment rate by 15 percentage points. Overall, the graphical and regression results are very much aligned. Besides, the regression estimates do not change to a significant extent if control variables are included (column 4) and are quantitatively similar if alternative bandwidths are specified (15 and 25 km in columns 1-2 and 5-6). We now turn to the effects of SIPTEA on mobility for different levels of education. The Census data contains five categories regarding the attained educational level: illiterates, less than 5 years of education, primary education (5 years of completed education), secondary education (8 years of completed education) and tertiary education (more than 12 years of education). In Spain, a large expansion of education took place in the second half of the 20th century with a major reform in 1970 that made education free and compulsory for children aged 6 to 14. As a result, the proportion of the population with secondary education (8 years of schooling) increased very 13 rapidly. In the regions that are contiguous to the SIPTEA border (see panel A in Map 1), the share of individuals with secondary education was 78, 68 and 43 percent for the cohorts aged 16, 26 and 36 in 1991. In this context, primary versus secondary education is likely to reflect age rather than skill and, thus, we construct the following three educational categories: less than 5 years of schooling (includes illiterates), 5 to 12 years of schooling and tertiary education. For the cohorts aged 26 and 36 in 1991, the proportions of individuals with less than 5 years of schooling were 8 and 20, whereas the corresponding figures for the 5 to 12 years of schooling category were 78 and 67 percent. Finally, 13 and 12 percent of the individuals aged 26 and 36 in 1991 had tertiary education. The results on out-migration (panel A) and in-migration (panel B) behaviors by the described educational groups are reported in Table 4, which shares the structure of Table 3 in terms of columns. The probability to stay for a given educational group is defined as the count of individuals (aged 16 to 64) in 1991 in that group that were in the municipality in 1981, divided by the population in 1981. Likewise, the in-migration rate for a given educational group is the number of individuals (aged 16 to 64) in 1991 in that group that were in a different municipality in 1981, divided by the population in 1981. [Insert Table 4 here] Starting with the probability to stay, the results in panel A indicate that the effect of SIPTEA on out-migration is driven by the behavior of individuals within the lowest educational group (less than 5 years of schooling). In fact, no statistically significant effects are found for the remaining educational categories. If SIPTEA diminished the gains from migrating to an urban area relative to staying in a rural municipality, these results suggest that the returns to migration were lower for the less educated individuals. As for the in-migration rate, quantitatively similar effects are found for the two categories with the lowest educational levels (about 0.5 additional in-migrants per 100 inhabitants in 1981). However, if the effect is measured relative to the variables’ mean, the effect is much larger for the category with less than 5 years of schooling (0.46) than for the group with 5 to 12 years of education (0.15). Overall, the policy seems to affect out- and in- migration rates, with the effects being larger for the less skilled individuals. In Table 5 we analyze the effects of SIPTEA on the in-migration rate by geographic origin of the inflow. In particular, we will consider in-migrants that come from (1) the untreated regions that are contiguous to the SIPTEA border, (2) the core regions of Catalunya and Madrid and (3) other municipalities within the same region. Focusing on the estimates of columns 3 and 4 (those using a bandwidth of 20 km), the largest inflows of in-migrants originated in the core 14 regions of Catalunya and Madrid. Given the massive waves of interregional migrants that received these two leading regions in the 60s and early 70s, the results suggest that SIPTEA caused some returned migration. The results also indicate the existence of modest welfare-induced flows of (cross-border) migrants. [Insert Table 5 here] In Table 6 we explore the program effects on the unemployment rate of different socioeconomic groups. To know if in-migration of welfare-prone individuals can partly explain the unemployment rate discontinuity at the SIPTEA border, panels A and B in Table 6 report the program estimates on the unemployment rate of stayers and in-migrants, respectively. Although the unemployment rate increases discontinuously at the border for both groups, the effect is larger for stayers than it is for in-migrants. Panels C and B report the unemployment results for males and females. The effect is slightly larger for females, a result that can be rationalized by the drastic increase in the proportion of women in the population of recipients of the agrarian unemployment benefit shown in Graph 1. [Insert Table 6 here] Given that SIPTEA especially affected the migration behavior of the less-educated, the discontinuous evolution of the unemployment rate at the SIPTEA border could potentially be the result of sorting by skills. In the second column of Table 7, we estimate the effect on the unemployment rate in 1991 holding fixed (1) the pre-treatment controls described in section 5 and (2) the shares of the municipal labor force in each of the educational groups. The results of a similar exercise are reported in column 3 where we hold fixed the shares of the municipal labor force in 18 different occupations. To ease comparability, the results of the baseline unemployment specification with pre-treatment controls (panel D in Table 3) are reproduced in the first column. [Insert Table 7 here] Controlling for the educational level of the labor force in the municipality leaves the estimated effect of SIPTEA on unemployment virtually unaffected. In contrast, including the share of the labor force in narrowly defined occupations reduces the estimated effect by half. This result suggests that one channel through which unemployment benefits increase 15 unemployment is to reduce the mobility of workers from high to low unemployment occupations. 7. Concluding remarks In this paper we have analyzed the effects on mobility of a national policy providing higher unemployment benefits to the residents of Andalucía and Extremadura, two lagging regions in the south of Spain. By comparing municipalities that are sufficiently close to (but at different sides of) the policy border, we identify the effect of unemployment benefits on population growth, out- and in-migration rates and on unemployment. In a rural environment with high unemployment and significant migration outflows, we find that unemployment benefits significantly reduce out-migration rates and, thus, increase population growth. We also find evidence that unemployment benefits increase unemployment and in-migration (from the ‘less generous’ side of the border and from core regions that attracted migrants in the 60s and 70s for labor market reasons). This paper shows that unemployment benefits can significantly affect the location of people and economic activities within a country by reducing the incentives to move from low to high productivity areas. Other welfare state programs may have similar effects. By providing uniform levels of basic services such as education or health care financed through ability-to-pay national taxes, the public sector redistributes income from rich to poor regions. In Spain, for the period 1991-1996, the annual difference between what the national government spent and collected in the regions of Andalucía and Extremadura represented 10.5 and 19.5 of the region’s value added13. This income redistribution at the regional level coexists with large and permanent unemployment rate differentials. In 1991, the unemployment rate in Andalucía and Extremadura was 25 and 23, well beyond 12 and 11 percent which were the unemployment rates in the leading regions of Catalunya and Madrid, respectively. In fact, Jimeno and Bentolila (1998) find that, in Spain, a reduction in a region’s unemployment rate does not cause any substantial inflow of migrants. Hence, the overall effects of the welfare state on migration may well exceed the effects identified in this paper. These figures correspond to the average of the period 1991-1996. See Table 7.4 (b) in Castells et al (2000) for details. 13 16 References Antolín, P. and Bover, O. (1997), “Regional migration in Spain: The effects of personal characteristics and of unemployment, wage and house price differentials using pooled crosssections”, Oxford Bulletin of Economics and Statistics 59, 215-235. Black, S. (1999), “Do Better Schools Matter? Parental Valuation of Elementary Education”, Quarterly Journal of Economics 114, 577-599. Bover, O. and Arellano, M. (2002), “Learning About Migration Decisions From the Migrants”, Journal of Population Economics 15, 357-380. Bover, O., Arellano, M. and Bentolila, S. (2002), “Unemployment Duration, Benefit Duration and the Business Cycle”, The Economic Journal 112, 223-265. Cansino, J. M. (2000), “El subsidio agrario. Principales magnitudes (1984-1999)”, Estudios Agrosociales y Pesqueros 189, 11-28. Card, D. and Levine, P. (2000), “Extended benefits and the duration of UI spells: Evidence from the New Jersey extended benefit program”, Journal of Public Economics 78, 107138. Castells, A., Barberán, R., Bosch, N., Espasa, M., Rodrigo, F. and Ruiz-Huerta, J. (2000), Las balanzas fiscales de las Comunidades Autónomas (1991-1996). Análisis de los flujos fiscales de las Comunidades Autónomas con la Administración Central, Ariel, Barcelona De Giorgi, G. and Pellizzari, M. (2009), “Welfare migration in Europe”, Labor Economics 16, 353-363. Edmark, K. (2009), “Migration effects of welfare benefit reform”, Scandinavian Journal of Economics 113, 511-526 Fiva, J.H. (2009), “Does Welfare Policy Affect Residential Choices? An Empirical Investigation Accounting for Policy Endogeneity”, Journal of Public Economics 93, 529-540. García-Pérez, J. I. (2004), “Problemas de incentivos en el diseño de políticas asistenciales para la protección por desempleo: una aplicación al subsidio agrario”, Universidad Pablo de Olavide, mimeo. Gelbach, J. (2004), “Migration, the lifecycle and state benefits: how low is the bottom?”, Journal of Political Economy 112 (2004), 1,091–1,130 Goerlich-Gisbert, F. and Cantarino-Martí, I. (2010), “Rugosidad del terreno: una característica del paisaje poco estudiada”, Fundación BBVA Working Paper 2010-10. González, J. J. (1990), “El desempleo rural en Andalucía y Extremadura”, Agricultura y Sociedad 54, 229-266. 17 Holmes, T. (1998), “The Effects of State Policies on the Location of Industry: Evidence from State Borders”, Journal of Political Economy 106, 667-705. Imbens, G. and Lemieux, T. (2008), “Regression discontinuity designs: A guide to practice”, Journal of Econometrics 142, 615-635. Jimeno, J. F. and Bentolila, S. (1998), “Regional unemployment persistence (Spain, 1976– 1994)”, Labor Economics 5, 25-51. Katz, L. and Meyer, B. (1990), “The impact of the potential duration of unemployment benefits on the duration of unemployment”, Journal of Public Economics 41, 45-72. Lalive, R. and Zweimüller, J. (2004), “Benefit Entitlement and Unemployment Duration: The Role of Policy Endogeneity”, Journal of Public Economics 88, 2,587-2,616. Lee, D. and Lemieux, T. (2010), “Regression Discontinuity Designs in Economics”, Journal of Economic Literature 48, 281-355. Levine, P. and Zimmerman, D. (1999), “An Empirical Analysis of the Welfare Magnet Debate Using the NLSY”, Journal of Population Economics 12, 391-409 McKinnish, T. (2005), “Importing the Poor: Welfare Magnetism and Cross-Border Migration”, Journal of Human Resources 40, 57-76. - (2007), “Welfare-induced migration at state borders: New evidence from micro-data”, Journal of Public Economics 91, 437-450. 18 Table 1. Unemployment assistance in selected Spanish regions in 2001 Number of Recipients Recipients per unemployed aged 16-64 AUB AUB GUB + GUB 0.12 0.28 0.40 0.14 0.28 0.42 0.12 -.0.12 0.21 -.0.21 0.15 -.0.15 Region Andalucía Extremadura Castilla-León Castilla-la Mancha Murcia AUB GUB 85,021 14,806 17,301 17,670 9,752 206,248 30,079 -.-.-.- Source: Boletín de Estadísticas Laborales. AUB and GUB stand for agrarian and general unemployment benefits, respectively. 19 Table 2. Descriptive statistics (municipalities within 20km from border; N=236). Variable SIPTEA Population in 1981 Population growth, 1981-91 (%) Probability to stay, 1981-91 In-migration rate, 1981-91 Unemployment rate, 1991 (%) Individuals with < 5 years of schooling, 1991 (%) Individuals with 6 to 12 years of schooling, 1991 (%) Individuals with tertiary education, 1991 (%) Ruggedness (logged) Agriculture employment per 100 inhabitants, 1982 Industry employment per 100 inhabitants, 1980 Services employment per 100 inhabitants, 1980 Unemployment rate, 1981 (%) Number of service establishments, 1980 Number of health care centres, 1980 Number of education centres, 1980 Prob. to stay., <5 years of school., 1981-91 Prob. to stay., 5-12 years of school., 1981-91 Prob. To stay., tertiary education, 1981-91 In-migration rate, <5 years of school., 1981-91 In-migration rate, 5-12 years of school., 1981-91 In-migration rate, tertiary education, 1981-91 Mean 0.398 1,943 -15.496 48.082 5.781 24.256 33.824 62.504 3.672 3.482 13.591 1.851 6.104 13.591 58.114 1.826 3.216 17.120 29.524 1.438 1.214 4.019 0.547 S.D. -.5,188 12.692 8.091 3.185 14.486 18.154 18.085 2.239 0.570 13.911 2.604 3.017 13.911 150.104 4.979 7.387 9.426 9.705 0.969 1.124 2.494 0.497 Min. 0 45 -57.778 21.875 0 0 0 18.033 0 1.579 0 0 0.978 0 0 0 0 0 7.657 0 0 0 0 Max. 1 60,627 20.368 64.622 15.868 81.429 80.562 100 13.333 4.591 91.270 22.500 22.813 91.270 1,666 55 81 42.057 55.709 5.941 8.721 12.079 3.425 Notes: Probability to stay defined as the number of 16 to 64 year-olds in 1991 that were in the municipality in 1981, divided by the population in 1981. In-migration rate defined as the number of 16 to 64 year-olds in 1991 that were not in the municipality in 1981, divided by the population in 1981. Data on attained educational levels in 1991 are referred to individuals aged 16 to 64. Ruggedness of terrain from GoerlichGisbert and Cantarino-Martí (2010). 20 Table 3. The effects of SIPTEA on population growth, probability to stay, in-migration rate and on unemployment. <15 km <20 km <25 km Panel A: Population growth 1981-91 SIPTEA 4.274** (1.733) 4.233** (1.679) 5.151*** (1.548) 4.925*** (1.449) 5.646*** (1.356) 5.418*** (1.289) 2.792** (1.154) 4.048*** (1.031) 3.963*** (1.054) 4.273*** (0.950) 4.003*** (1.006) 1.303*** (0.433) 1.316*** (0.415) 1.195*** (0.406) 1.971*** (0.497) 2.108*** (0.527) Panel B: Probability to stay 1981-91 SIPTEA 2.622** (1.147) Panel C: Move-in rate 1981-91 SIPTEA 1.493*** (0.454) Panel D: Unemployment rate in 1991 SIPTEA Control variables N 14.051*** (2.032) 13.718*** (1.983) 14.790*** (1.878) 13.487*** (1.880) 16.378*** (1.562) 15.002*** (1.586) N Y N Y N Y 184 236 306 Notes: Robust standard errors in parenthesis. ***,** and * statisitically significant at 1, 5 and 10%. All specifications include region-pairs fixed effects. Control variables are terrain ruggedness, agriculture employment per 100 inhabitants in 1982, industry and services employment levels per 100 inhabitants in 1980, the unemployment rate in 1981 and the number of establishments in the services sector, healthcare centres and education centres in 1980. 21 Table 4. The effect of SIPTEA on the probability to stay and on the in-migration rate, by educational level. <15 km <20 km <25 km Panel A: Probability to stay 1981-91 <5 years of schooling SIPTEA 4.049*** (1.169) 3.829*** (1.294) 5.005*** (1.063) 4.389*** (1.189) 5.681*** (0.968) 4.841*** (1.055) -0.624 (1.185) -1.619 (1.039) -1.073 (1.042) 0.198 (0.123) 0.211* (0.111) 0.236** (0.111) 0.789*** (0.176) 0.844*** (0.197) 0.541* (0.300) 1.008*** (0.341) 1.106*** (0.348) 5 to 12 years of schooling SIPTEA -1.534 (1.311) -1.185 (1.291) -1.160 (1.171) Tertiary education SIPTEA 0.107 (0.139) 0.149 (0.139) 0.203* (0.122) Panel B: Probability to move-in 1981-91 < 5 years of schooling SIPTEA 0.535*** (0.168) 0.524*** (0.180) 0.557*** (0.145) 0.558*** (0.160) 5 to 12 years of schooling SIPTEA 0.826** (0.340) 0.669** (0.311) 0.621* (0.318) Tertiary education SIPTEA 0.132* (0.067) Control variables N N 0.110 (0.067) 0.138** (0.064) 0.095 (0.065) 0.174*** (0.060) 0.158** (0.064) Y N Y N Y 184 236 306 Notes: Robust standard errors in parenthesis. ***,** and * statisitically significant at 1, 5 and 10%. All specifications include region-pairs fixed effects. Control variables are terrain ruggedness, agriculture employment per 100 inhabitants in 1982, industry and services employment levels per 100 inhabitants in 1980, the unemployment rate in 1981 and the number of establishments in the services sector, health-care centres and education centres in 1980. 22 Table 5. The effect of SIPTEA on the in-migration rate, 1981-91, by geographic origin of the inflow <15 km <20 km <25 km Panel A: Cross-border SIPTEA 0.220* (0.117) 0.147 (0.105) 0.251** (0.103) 0.175* (0.098) 0.267*** (0.082) 0.206** (0.081) 0.405* (0.220) 0.499*** (0.178) 0.401** (0.197) 0.522*** (0.148) 0.482*** (0.165) 0.248 (0.207) 0.157 (0.199) 0.144 (0.197) 0.715** (0.339) 0.868** (0.379) Y N Y N Y Panel B: From the Core SIPTEA 0.475** (0.203) Panel C: Within the region SIPTEA 0.261 (0.208) Control variables N N 184 236 306 Notes: Robust standard errors in parenthesis. ***,** and * statisitically significant at 1, 5 and 10%. All specifications include region-pairs fixed effects. Control variables are terrain ruggedness, agriculture employment per 100 inhabitants in 1982, industry and services employment levels per 100 inhabitants in 1980, the unemployment rate in 1981 and the number of establishments in the services sector, health-care centres and education centres in 1980. 23 Table 6. The effects of SIPTEA on different unemployment rates in 1991 <15 km <20 km <25 km Panel A: Stayers SIPTEA 14.335*** (2.102) 14.040*** (2.034) 15.089*** (1.936) 13.748*** (1.934) 16.834*** (1.605) 15.370*** (1.631) 9.991*** (2.538) 9.384*** (2.903) 12.023*** (2.235) 10.749*** (2.468) 12.318*** (2.086) 11.433*** (2.273) 12.872*** (2.103) 12.589*** (2.053) 13.848*** (1.930) 12.582*** (1.912) 15.260*** (1.593) 13.931*** (1.613) 14.371*** (2.477) 14.697*** (2.546) 14.481*** (2.418) 14.109*** (2.624) 16.130*** (2.063) 15.547*** (2.226) N Y N Y N Y Panel B: In-migrants SIPTEA Panel C: Males SIPTEA Panel D: Females SIPTEA Control variables N 184 236 306 Notes: Robust standard errors in parenthesis. ***,** and * statisitically significant at 1, 5 and 10%. All specifications include region-pairs fixed effects. Control variables are terrain ruggedness, agriculture employment per 100 inhabitants in 1982, industry and services employment levels per 100 inhabitants in 1980, the unemployment rate in 1981 and the number of establishments in the services sector, health-care centres and education centres in 1980. 24 Table 7. The effect of SIPTEA on the unemployment rate in 1991: Further results. Pre-determined controls Pre-determined controls + educational shares Pre-determined controls + occupational shares <15 km (N=184) SIPTEA 13.718*** (1.983) 13.632*** (2.016) 7.281*** (2.247) <20 km (N=236) SIPTEA 13.487*** (1.880) 13.307*** (1.928) 6.823*** (1.953) <25 km (N=306) SIPTEA 15.002*** 14.746*** 7.601*** (1.586) (1.609) (1.692) Notes: Robust standard errors in parenthesis. ***,** and * statistically significant at 1, 5 and 10%. All specifications include region-pairs fixed effects as well as terrain ruggedness, agriculture employment per 100 inhabitants in 1982, industry and services employment levels per 100 inhabitants in 1980, the unemployment rate in 1981 and the number of establishments in the services sector, health-care centres and education centres in 1980. The results in column 2 correspond to a specification that further includes the shares of the population aged 16 to 64 in different educational levels. The results in column 3 correspond to a specification that further includes the shares of the labor force in 18 narrowly defined occupations. 25 300 100 200 Socialist first election win 0 Number of employees/recipients (in thousands) Graph 1. ‘Employees’ in Empleo Comunitario (EC) and recipients of the Agrarian Unemployment Benefit (AUB). 1979 1981 EC 1983 1985 AUB 1987 AUB, female Source: INEM and González (1990). 26 1989 1991 AUB, male Graph 2. ‘Pre-treatment’ municipality characteristics B: Agriculture employment in 1982 2 Ruggedness index (logged) 2.5 3 3.5 4 Full-time equivalent jobs per 100 inhab. 5 10 15 20 A: Ruggedness of Terrain -200 -100 0 100 200 -200 -100 0 100 200 D: Services employment in 1980 0 4 Employees per 100 inhab. 2 4 6 8 Employees per 100 inhab. 6 8 10 12 10 14 C: Industry employment in 1980 -200 -100 0 100 -200 200 0 100 200 F: Services establishments in 1980 0 5 Unemployment rate 10 15 20 25 Number of establishments 100 200 300 30 400 E: Unemployment in 1981 -100 -200 -100 0 100 200 -200 -100 100 200 H: Education centers in 1980 0 0 Number of centers 5 10 Number of centers 5 10 15 15 20 G: Health centers in 1980 0 -200 -100 0 100 200 -200 -100 0 100 200 Notes: The horizontal axis is the distance to the border where treated municipalities are assigned positive values. Ruggedness index from Goerlich and Cantarino’s (2010) study. 27 Graph 3. Population growth, probability to stay and in-migration 1981-91, and unemployment in 1991. B: Probability to stay 1981-91 45 Probability to stay 50 55 Population growth rate -20 -10 0 10 60 A: Population growth 1981-91 -200 -100 0 100 200 -200 -100 0 100 200 D: Unemployment in 1991 -200 -100 0 100 200 10 5 In-migration rate 10 Unemployment rate 20 30 15 40 C: In-migration 1981-91 -200 -100 0 100 200 Notes: The horizontal axis is the distance to the border where treated municipalities are assigned positive values. Probability to stay defined as the count of individuals aged 16-64 in the municipality in 1991 that lived in the same municipality in 1981, divided by population in 1981. In-migration rate defined as individuals aged 16-64 in the municipality in 1991 that lived in another municipality in 1981, divided by population in 1981. The unemployment rate is for individuals aged 16-64. 28 Map 1. SIPTEA regional implementation and border municipalities. A: Regions in Spain Castilla-León B: Provinces at the border Catalunya Madrid Extremadura Castilla-la Mancha Murcia Andalucía C: Municipalities within 20 km from the border Notes: From left to right in panel B, the provinces that are contiguous to the SIPTEA implementation border are Cáceres, Badajoz, Córdoba, Jaén, Granada and Almería (treated) and Salamanca, Ávila, Toledo, Ciudad Real, Albacete and Murcia (untreated). There are 1,801 municipalities in these 12 provinces. Within these provinces, there are 236 municipalities within 20 km from the border (Panel C). 29