The Economic Impact of Schools for Local Labor Markets

The Economic Impact of Schools for Local Labor Markets

Ronny Freier ^∗ Martin Simmler ^† Christian Wittrock ^‡

Abstract

This paper tests the presumption that urbanization due to agglomera- tion economies is fortified by the centralization of public good provision in rural areas due to economies of scale. Our empirical analysis focuses on the impact of the availability of grammar schools on local employment in mid-sized towns in East Germany. We study the short-run effect using a difference-in-differences design that exploits grammar school closures between 1997 and 2008. To understand the long-run changes, we em- ploy an IV estimator for school openings after the German re-unification using historic information about the local availability of institutions for higher education as excluded instrument. We find that school closures reduce local employment in the short run by about 10% if the share of out-commuters is small otherwise local employment is unaffected. The reasons is that out-commuters are first movers and thus may mitigate the full shock by reducing house prices. The long run effect of grammar school openings on local employment is much larger with 120%. Almost 60% of the effect are due to less out-commuting (due to higher wages in the jurisdiction) and the reaming part is due to an increase in the population. Further, we study the incidence of grammar schools into house prices and find consistent results.

JEL Classification: J1, J2, R1, R5

Keywords: local schools, employment, commuting, house price

∗

DIW Berlin, Mohrenstrasse 58, 10117 Berlin, Germany, and Freie Universitaet Berlin, Garys- trasse 21, 14195 Berlin, Germany. E-mail: [email protected].

†

University of Oxford Centre for Business Taxation, Park End Street, OX1 1HP, Oxford, UK, and DIW Berlin. Tel: +44 1865 614845. E-mail: [email protected].

‡

Corresponding author: Ruhr Graduate School in Economics and U Bochum, Universittsstrae

150, 44801 Bochum, Germany. Tel: +49 234 32 22874. E-mail: [email protected].

I. Introduction

The UN estimates that in 2050 almost 70% of the world population will live in urban areas, up from 30% in 1950 (UN 2014, World Urbanization Prospects - The 2014 Revision). The driving force behind this trend is seen in agglomeration and urban- ization economies which increase productivity (Rice et al. (2006); Graham (2007);

Graham and Kim (2008)) and thus wages (e.g. Melo et al. (2009)). While adjusting house prices mitigate the trend in urbanization (e.g. Combes et al. (2012)), the centralized provision of public goods for which economies of scale are important due to the decline in population density in rural areas may re-enforce the urbaniza- tion trend. Our paper tests this presumption by investigating the role of grammar schools for local employment in mid-sized towns. An important dimension of town heterogeneity we explore is the fraction of out-commuters. The importance of com- muting has been increasing in recent decades (e.g. Hymel (2009) for the US) due to higher wages and house prices in cities, and a decline in transportation costs. Our work highlights that out-commuters are pivotal in understanding the short run as well as the long run effect of local public goods on employment.

Our paper starts by setting out an a stylized, theoretical model of local employ- ment in a (small) village and a (large) town. All potential villagers have different preferences for living in the village and different commuting costs. We show that the impact of an increase in the locally provided public good on employment depends largely on commuting costs. Further, we show that the long term effect is likely to be much larger due to the additional agglomeration economies effect which changes local employment along two margins. First, the attractiveness of the city increases which leads to an inflow of people. Second, less inhabitants out-commute as the wage differential to high-wage jurisdiction decreases.

We test the theoretical predictions using grammar schools in mid-sized towns in East Germany between 1990 and 2008. We study the short term impact of gram- mar schools by investigating school closures, which are arguable more exogenous especially with regard to the exact timing, using a difference-in-differences design.

Our treatment group experienced the closure of its only grammar school, while the

only grammar school of the control group was not closed. Our results reveal that

on average school closures did not reduce average local employment. We only find

a significant and relatively large effect of 9% in jurisdictions with a small share

of out-commuters. This is consistent with the theoretical prediction. To address

whether differences in the preference for grammar schools between treatment and

control group can explain our results, we investigate the incidence of school closures into house prices. Using mirco-level offer data from 2004 to 2008, we find that house price dropped in all towns with a school closure by around 11%. This suggests sim- ilar preference for grammar schools in jurisdictions with a high and a low share of commuters.

To test the long run predictions of our model, we focus on grammar school openings after the German re-unification in 1990. Although higher education schools existed in the German Democratic Republic, much less pupils went there and also the number of grammar schools was lower. We implement an instrumental variable approach which accounts for omitted variable bias due to, for example, political governance quality. Our excluded instruments are the distance to cloisters before 1800 as well as the presence of higher education schools in 1914. Conditional on our set of control variables which capture past agglomeration benefits we estimate the long run (18 years) impact of grammar schools on local employment to be in the order of 120%. More than half of the increase is due to a reduction in the number of out-commuters as overall population increases “only” by 50%. Consistent with the increase in the population, we find that house prices increased on average by 46%.

Our analysis contributes to several streams of literature: (Agglomeration and Public Good Provision) First, researchers have been evaluating the economic effects of other public inputs. Universities, e.g., have been shown to effect regional de- velopment positively (see Audretsch and Lehmann (2005); Abel and Deitz (2011);

Audretsch et al. (2012); Frenette (2009); Goldstein et al. (1995)). Less clear is the evidence of airports (Kanafani and Abbas (1987); Redding et al. (2011)), public transport (Koh et al. (2013)) and highways (Garcia-L´ opez et al. (2015)). (Forces that Drive Agglomeration Economies) Further, our study can be linked to the lit- erature that is concerned with agglomeration economies. is concerned with local economic development and the distribution of fiscal means. Those studies asks how the local economy can benefit from additional funds, without specifying the tar- geted public goods or services for which those funds are used (see Becker et al.

(2010, 2013); von Ehrlich and Seidel (2015)).

(Incidence and Public Good Provision) Second, we contribute to the literature that evaluate the impact public good provision inputs on land and property prices (Fack and Grenet (2010); Cellini et al. (2010)) or wages (Beeson and Eberts (1989);

Greenwood et al. (1991)).

(Local Labor Market) Third, a broad literature focuses on the determinants

of local employment and the functioning of local labor markets (see Topel (1986);

Enrico (2011)). Here, our effects can be compared to, e.g., the closures of large military bases (aus dem Moore and Spitz-Oener (2012)), plant closures/openings (Greenstone and Moretti (2003)) or industry specific shocks (Black et al. (2002, 2005b,a); Marchand (2012)). Proximity of educational institutions has been shown to matter greatly on the individual level (Do (2004); Frenette (2006); Spiess and Wrohlich (2010); Denzler and Wolter (2011); Freier and Storck (2012)). Our results here answer the question how the individual effects add up to the aggregate labor market improvement of the jurisdictions.

The remainder of this article is structured as follows. In Section 2, we present the institutional settings necessary to understand our research approach. Section 3 sets out a stylized theoretical model to motivate our empirical analysis. Our two empirical strategies are described in section 4, while section 5 includes information about the data and reports descriptive statistics. Our findings are presented and discussed in section 5. Section 6 concludes.

II. Institutional Background

With the reunification on the 3rd of October 1990, East Germany adopted – among other institutions – the West German school system. Before reunification, East Ger- many used a system in which children attended a common school (Polytechnische Oberschule ) until the age of 15/16. Part of the pupils then continued education at a higher secondary school which ended with a university entrance diploma (Erweiterte Oberschule ). After the re-unification, pupils only attended a common elementary school until the age of 10/12 (depending on the state). Then, pupils were tracked into different school types, distinguished by the level of academic ambitions, curricu- lum and years to receive a degree. To put it simply, we can differentiate three main types of schools by the overall quality of the education: grammar schools (Gym- nasium ) as the highest quality school, other secondary modern schools (Realschule, Gesamtschule, Oberschule ) or basic secondary schools (Hauptschule ). Importantly, the Gymnasium is the only school that provides a regular path for all pupils to obtain 12/13 years of schooling and an university-entrance diploma.

1

Note that good students at the other schools may have the chance to switch schools after a

number of years and move up to a Gymnasium. In some states, different school tracks are offered

within one schools (Gesamtschule.

Variation in school availability in small and medium sized cities during our ob- servation period stems from two sources: (i) the availability of higher secondary schools in the GDR and (ii) the demographic trend in East Germany after the re-unification. The number of pupils going to higher secondary schools was much higher in West Germany before 1990 than in the GDR. In West Germany 33.5% of all pupils leaving school in 1990 had a university entrance degree and 24.4% went to a grammar school. In East Germany, only 13.3% of the pupils leaving schools had a university entrance degress and only 8.8% have went to a grammar school.

After the re-unification, the share of pupils going to grammar school and leaving schools with university entrance degree increased. In 2014, on average 41% of the pupils went to grammar schools and almost 52% of the pupils leaving school had a university entrance degree. No differences between East and West German states are observed in these shares.

. The number of pupils at higher education schools in the GDR is also reflected in the number of jurisdictions that had a grammar school in the GDR. Half of the roughly 250 jurisdictions with one grammar school today had no grammar school before 1990, and 25 jurisdictions that had a grammar school before 1990 have none today. Thus, the re-unification increased the important of secondary schools for parents and pupils and reduced transport costs dramatically due to a higher number of grammar schools.

The re-unification changed, however, not only the number of grammar schools but also the spatial distribution of inhabitants. Most important, the re-unification triggered a massive migration wave from East to West Germany. Overall, about four million (out of 16 million) moved to the western part for at least some time.

As it was mostly the younger people moving, this had direct effects on the number of pupils attending east German schools. Second, the reunification also triggered a wave of migration from rural to urban areas. Saxony-Anhalt, a mostly rural area saw its population decrease by more than 20% from 1993-2013 (after the initial east-west migration) and had to close about 60% of all schools in rural areas during the process. Finally, East Germany experienced a massive drop in child births when families postponed the fertility decision due to economic uncertainty after the reunification (see Chevalier and Marie (2013)). Estimates of the fertility gap in the early 90s show that birth rates dropped to half and stayed significantly lower ever

2

http://www.bildung-weltweit.de/pdf/kurzdarstellung_deutschland.pdf

3

https://www.destatis.de/DE/Publikationen/Thematisch/BildungForschungKultur/

Schulen/BroschuereSchulenBlick0110018169004.pdf?__blob=publicationFile

since. This implies that 6 years later, elementary schools saw a large drop in the number of pupils and ten years later the number of students for higher secondary school was declining.

The responsibilities for schools are shared between all tiers in the German fed- eral tier system (states, counties and municipalities). The largest burden falls to the state level which is responsible for hiring and overseeing teachers and providing general funds per student. Counties are often responsible for student transporta- tion, youth welfare services and schools for children with special needs. The local municipalities are usually in charge of the school infrastructure and maintain school buildings, facilities and equipment. A school opening or closure is not to be decided individually by one of the tiers and is usually a long political process. In the end, the state level which provides the decisive funding has the ultimate decision power, however, municipalities do have a say. It is important to note that neither openings nor closures are happening over night. In the data, we assign openings and closure to the year of the actual school opening and closing. Anecdotal evidence and inter- views with municipality administrations suggest that the announcement for closure is, however, usually done two years in advance.

III. Theoretical Motivation

In the following section, we present a stylized theoretical model to motivate our empirical analysis. There are two jurisdictions, a small village (v) and a large town (t). We normalize the potential village population to 1 and assume the village population to be small compared to the town population; changes in the village population have thus no impact on towns.

Every inhabitant in the village either works in the village and receives w

or works in the town and receive w

. There are many firms in the town and the village and all firms produce an export good, for which the price is normalized to one. Production uses labor as input and has constant returns to scale. All firms have the same production function except for a different jurisdiction specific total factor productivity (A). Wages in the town and in the village are thus given by

w

= A

(1)

with j = t, v. We assume that total factor productivity in the town is higher than in the village which is motivated by agglomeration economies (A

> A

).

4

Qualitatively results are unchanged when relaxing this assumption.

We refer to a village inhabitant who works in the town as a commuter as we are focusing on local employment in the village. Commuters incur costs (c

), which we assume to be heterogeneous across inhabitants and uniformly distributed between c and c. Inhabitants in the village and town pay housing costs of p

and p

respectively.

Housing demand is given by the number of inhabitants in the village, thus workers and commuters. We assume that housing supply adjustments ensures that house prices equal the return for an alternative use of land (e.g. agricultural value of land).

Every inhabitant in the village and the town values (v(.)) a locally provided non-rival public good pg

and pg

respectively, e.g. grammar schools. Further, every inhabitant has a preference (η

) for living in the village. Preferences are uniformly distributed between η and η. To simplify the analysis, we normalize the range between η and η to 1.

The utility of worker i in one of the two jurisdictions is given by

U

= w

− p

+ v(pg

) + D(j 6= t)η

(2)

with j = t, v and D as an indicator variable. Utility of a commuter i in one of the two jurisdictions is given by

U

= w

− p

+ v (pg

) + D(j 6= t)η

− c

(3)

.

We are interested in how the provision of the public good in the village affects the number of workers in the village. To assess this, we first derive an expression for the number of workers and commuters in the village and characterize the equilibrium.

Then, we study how a change in the public good affects equilibrium values in the short and long run.

For each potential worker in the village two conditions need to be met. First,

utility of living and working in the village (U

) needs to be larger or equal to

living and working in the town (U

). Second, utility of living and working in

the village needs to be larger or equal to living in the village and working in the

town (U

) . Re-formulating the first condition provides an expression for the

minimum preference of all workers (η

) for living in the village (equation (4)). It

states that the minimum preference is equal to the sum of wage differential, housing

price differential and value of public good differential. Re-formulating the second

condition provides an expression for the minimum commuting costs of all workers (c

) in the village (equation (5). For the marginal worker the wage in the village equals the net wage in the town (wage minus commuting costs).

η

= (w

− w

) − (p

− p

) + (v(pg

) − v(pg

)) (4)

w

= w

− c

(5)

η

− c

= −(p

− p

) + (v(pg

) − v(pg

)) = η

− c

(6) For out-commuters two similar conditions have to be fulfilled. For them utility of being a commuter (U

) needs to be larger than being a worker in the village (U

). This is equivalent to the second condition for the workers (see equation (5)).

Thus, there will be only commuters in the village if wages are higher in the town than in the village (which we assumed in the beginning) and minimum (maximum) commuting costs of workers (commuters) c

exceed minimum commuting costs, c, which we will assume in the following.

The second condition states that utility of commuters in the village must bet larger than utility of being a worker in the town (U

). This provides an expression for the minimum net preference (preference minus commuting costs) for living in the village of all out-commuters (see equation (6)). By using equation (4) and (5) we can show that the minimum preference of commuters is smaller than the minimum preference of workers since by assumption minimum costs of workers c

exceed minimum commuting costs c, thus for some commuters commuting costs (c

) is lower than c

.

Using the minimum preference of workers and commuters as well as the minimum

commuting costs for workers, the number of workers is given in equation (7) and the

number of commuters in equation (8). There are two types of commuters. First,

commuters with similar location preferences as workers but lower commuting costs

(first term). Second, commuters with an even lower preference of living in the village

but also even lower commuting costs on average than the first group of commuters

(second term). Intuitively, the first group is more likely to be a worker in the village,

the second group more likely to be a worker in the town. Figure 1 illustrates the

number of workers and commuters graphically. The horizontal axis captures the

preference parameter, the vertical axis the probability density function as well as

commuting costs. The number of workers is given by the surface W, the number of

commuters by the sum of the surfaces C1 and C2.

Figure 1: Illustration Share of Workers and Commuters - Equilibrium.

η^w c^w

W

C1

c^w c̲/η̲ η̅

C2 c̅/1

η^c

N

= (c − c

)(η − η

) (7) .

N

= c

(1 − η

) + (c

)

/2 (8) .

If we assume that commuters and workers respond depending on their net pref- erence and thus similar to an decrease in public good provision and wages are un- affected, the minimum preference of commuters and workers shifts to the right by the same amount (see Figure 2). If we assume that commuters and workers re- spond according to the (gross) preference, commuters move away first as they have the lowest preference (see Figure 3). The implications for local employment are different depending on who moves first although population changes to the same extend. Since there is not proof for one or the other presumption, we leave this theoretical un-determined and assess the empirical evidence on this question taking the different prediction into account.

To analyze the long run effect, we introduce agglomeration economies and link

total factor productivity to the number of workers in a jurisdiction. To be closer to

the following empirical analysis, we consider an increase in the public good, which

Figure 2: Illustration Share of Workers and Commuters - Short Term Effect (Net Preference).

η^w c^w

W

C1

c^w

η̅

Δpop

C2 c̅/1

c̲/η̲ η^c

Figure 3: Illustration Share of Workers and Commuters - Short Term Effect (Gross Preference).

η^w c^w

W

C1

c^w

η̅

Δpop

C2 c̅/1

c̲/η̲ η^c

Figure 4: Illustration Share of Workers and Commuters - Short Term Effect (Gross Preference).

η^w c^w

W

C1

η̅

C2 c̅/1

c̲/η̲ η^c

c^w1

η^w1

increases the number of workers and thus increases wages in the jurisdiction. This affects the number of workers along two margins. First, the share of inhabitants with the same preference that are workers increases by (c

− c

)(η − η

) (see Figure 4). Second, due to an additional inflow of workers and commuters the minimum preference of inhabitants shifts from η

to η

and house prices increases increase such that utility is the same for the marginal commuter and marginal worker. The long run effect on local employment is thus much larger. Allowing for a change in housing supply increases the effect further as house prices decreases lead to a further inflow of inhabitants. If housing supply adjust completely house price return to the before change level. Without adjustment in housing supply, house prices change in a similar magnitude as the change in population.

To sum up, we expect that the effect of school closures on local employment is decreasing with the share of commuters in a jurisdiction if commuters are first movers (Hypothesis 1). In both jurisdictions, house price should, however, change to the same extend as population has to change to a similar extend (Hypothesis 2).

The long run impact of school openings is likely to be much larger (Hypothesis (3))

due to agglomeration economies. Part of the increase in local employment stems,

however, from the fact that less inhabitants commute and thus, the population

increase is smaller than the increase in the number of workers (Hypothesis (4).

IV. Empirical Strategy

In this section, we explain our empirical strategy to test the predictions outlined in the previous section. To assess the short run impact of schools we rely on a DiD design that exploits school closures in all East German states between 1997 and 2008. To test the long run impact, we apply an IV strategy for school openings after 1990 in East Germany using information about the historical availability of higher eduction institutions as excluded instruments. In the following we describe each approach in detail.

A. Differences-in-Difference Strategy

Table 1: School closings and openings in municipalities with one school and with at least two schools in 1997.

Schools = 1 Schools ≥ 2

Schools Opening Closing Schools Opening Closing

1997 262 0 0 345 1 2

1998 263 1 0 341 0 3

1999 263 0 0 343 0 1

2000 264 2 1 339 2 3

2001 262 0 2 334 2 7

2002 261 1 2 326 1 9

2003 254 0 7 305 0 14

2004 251 1 4 288 1 16

2005 248 4 7 270 2 14

2006 243 0 5 263 1 7

2007 240 2 5 255 0 7

2008 238 3 5 253 0 2

Source: Own data collection and calculations.

The identification strategy for the short term effect is to compare jurisdictions with and without the closure of their only grammar school before and after the school closure. Our baseline fixed effects estimation equation reads as follow:

Y

= α

+ β

T R

∗ Closure

+ γX

+ ν

+ λ

+

(9) Our dependent variable is the natural logarithm of the number of employees (liv- ing and working in the same municipality) and stems from the Federal Employment Agency.

Our treatment group (T R

) includes all jurisdictions with one grammar

5

In a robustness analysis, we also estimate a Poisson model. Results are basically unchanged.

school (Gymnasium) in 1997 that was closed until 2008. The data for grammar schools is hand-collected and cross-checked with the number of grammar schools available on a county level for Germany. We focus on jurisdictions with one gram- mar school as we believe that this provides the best variation for our analysis. In principle, school closures in cities with more than one grammar school could also be used if these closures leads to excess demand for grammar schools. However, we do not have information on the number of students on the school level and are thus not able to account for school size (changes).

Since not all school closures happened in the same year, the reform variable (Closure

) is jurisdiction-specific and is one for all years for which the jurisdiction had one higher secondary education school less. The number of school closures for each year is shown in Table 1. Most of the school closures happened after 2002. The geographical location of schools and school closings in East Germany is depicted in Figure 5. Most of the school closings took place in Saxony and Saxony Anhalt.

Our control group consists of jurisdictions with one grammar school which was open at least until 2008. We focus on jurisdictions with one school to ensure that treatment and control group jurisdictions are similar in size without further sample restrictions. On average jurisdictions with one grammar school had between 500 and 30,000 inhabitants in 1997 (see Figure 7 in Appendix A).

Our set of control variables, captured in the matrix X

includes the logarithm of population in 1997 interacted with year dummies, the local business tax (which is set by the municipality), and the number of jurisdictions within 10km distance that have a higher education school. The variables stem from Statistik Lokal provided by the Federal Statistical Office. The variables ν

and λ

in equation (10) represent time and municipality fixed effects. We report heteroscedasticity-robust standard errors clustered at the municipality level.

To account for a potential different impact of school closures on local employment in jurisdictions with a large and a small share of out-commuters, we construct an indicator variables (C

) that is one if the share of out-commuters in 1998 is above the mean and interact it with the treatment and reform interaction (see equation (10)).

6

We use 1998 data as out-commuters are not observed for all jurisdiction in 1997.

Figure 5: Schools and School Closings in East-Germany between 1997 and 2008.

Y

= α

+ β

T R

∗ Closure

+ β

T R

∗ Closure

∗ C

+ +γX

+ ν

+ λ

+

(10) Descriptive statistics for the whole sample and for treatment and control group for 1997 are reported in Table 2. The average jurisdiction in our sample has 12,280 inhabitants and around 1,865 workers. Control group jurisdictions are on average slightly larger (12,780 to 9,590) and have a higher number of workers, in all other dimension the two groups do not differ. This mitigates concerns about the validity of the common trend assumption. To provide also empirical support for this as- sumption, we include in our empirical specification two-forward leads for the school closures.

To address whether different preferences of commuters and thus different pref-

erences for grammar schools in jurisdictions with a large and a small share of out-

commuters co-founds our analysis, we also estimate the short-run impact of gram-

mar schools on house prices using the above outlined difference-in-differences design.

Table 2: Descriptive Statistics Local Employment Sample for 1997.

Full Sample Control Group Treatment Group Difference

mean/sd mean/sd mean/sd t-value

Employees 1795.28 1880.56 1257.35 623.2*

(1231.21) (1285.10) (586.39) (2.43)

Commuter Share 0.61 0.61 0.65 -0.0368

(0.13) (0.13) (0.11) (-1.66)

Population in 1000 (1997) 11.93 12.40 9.43 2.974**

(6.38) (6.62) (4.18) (2.67)

Area in km

70.51 69.73 74.60 -4.867

(63.03) (64.59) (54.75) (-0.44)

# Schools (0 − 10 km) 1.07 1.08 0.97 0.110

(2.53) (2.70) (1.28) (0.24)

Local Business Tax in % 3.28 3.26 3.36 -0.0953

(0.39) (0.40) (0.36) (-1.38)

Observations 237 199 38

* p < 0.10, ** p < 0.05, *** p < 0.01. Source: Own data collection and calculations.

The data stems from Empirica AG and includes offer price data for a wide range of objects from newspapers as well as online ads covering the years 2004 to 2008. We merged municipality information to this data set using information on the location of the properties. We include the following properties in the analysis: single family homes, semi-detached houses, and terraced houses. We focus on purchase offers and not rental offers as housing markets are mainly purchase markets in our jurisdic- tions. Further, in rural areas property markets consists mainly of houses. Due to a very low number of treated observations (below 50) for the states of Thuringia and Brandenburg, we exclude these two states completely. Although the offer price data has the potential disadvantage of selection driving the results, we believe that this is captured by municipality and time fixed effects.

Since houses are usually sold with land, we use the natural logarithm of the overall price as dependent variable and control for the amount of land as well as living space. Adverts differ strongly regarding the non-essential information included in the ad (e.g. location in the city, close to public transport or not, etc), which could bias our estimates. Therefore, we include in our analysis only baseline property characteristics as amount of land and living space, information on the condition of the property (high quality, newly built, renovated, in need of rehabilitation) and availability on balcony, garage and basement. Further, we include the same municipality characteristics as for the analysis of local employment.

Descriptive statistics for the whole sample and treatment and control group are

shown in Table 3. On average a house in the sample costs about 140,000 Euro and comes with 860 square meter of land and 140 square meter living space. The houses in the treatment group are substantially less expensive and have less living space but more land. The property tax in the treatment group is higher but local business tax rates are similar. Further, in the treatment group the share of out-commuters is larger.

Table 3: Descriptive Statistics House Price Sample for 2004 to 2008.

Full Sample Control Group Treatment Group Difference

mean/sd mean/sd mean/sd t-value

Price 135273.04 137383.21 124087.26 13296.0***

(72185.85) (72957.61) (66852.25) (11.83)

Amount of Land 834.32 827.24 871.86 -44.62**

(888.34) (880.04) (930.31) (-3.22)

Living Space 137.31 137.65 135.51 2.138*

(55.82) (56.37) (52.82) (2.45)

Single Family House 0.78 0.78 0.80 -0.0234***

(0.41) (0.42) (0.40) (-3.62)

Detached House 0.13 0.13 0.12 0.0113*

(0.34) (0.34) (0.32) (2.15)

Terraced House 0.09 0.09 0.08 0.0121**

(0.29) (0.29) (0.27) (2.69)

Population in 1000 (1997) 14.51 15.18 10.99 4.193***

(6.41) (6.59) (3.78) (43.16)

Area in km

68.93 66.93 79.53 -12.60***

(48.62) (46.90) (55.67) (-16.68)

Local Business Tax in % 3.56 3.56 3.55 0.00324

(0.41) (0.43) (0.34) (0.50)

Property Tax in % 3.68 3.67 3.73 -0.0560***

(0.41) (0.41) (0.37) (-8.83)

Share Commuter 0.66 0.65 0.70 -0.0435***

(0.12) (0.12) (0.10) (-23.59)

Observations 30761 25879 4882

* p < 0.10, ** p < 0.05, *** p < 0.01. Source: Own data collection and calculations.

B. Instrumental Variable Strategy

To asses the long run impact of grammar schools we focus on grammar school openings in East Germany after the German re-unification in 1990. This allows us to account for adjustments over a period of 18 years and we are less constrained by the availability of data and the small number of observations used for school closures.

As outlined in Section 2, we believe that the German re-unification provides a

suitable setting as it provides exogenous variation in the availability of grammar schools due to the regime change. Despite this unique setting, there are two chal- lenges when estimating the impact of grammar schools on local employment in the cross-section. First, grammar school openings are likely to correlate with political governance quality which is not observed. We address this concern by implementing an instrumental variable approach using historical information on the local avail- ability of higher eduction institutions as excluded instrument. In particular, we use an indicator variable for whether the centroid of the jurisdiction was close (less than

≈ 11km - the median in the sample) to a cloisters that existed before 1800 as well as the presence of higher education schools around 1914 in Prussia. Since higher education schools in Prussia had a school track allowing graduates to apply for uni- versity and thus are similar to a Gymnasium today, we expect them to be positively correlated with the openings of grammar schools after 1990. Out from the different school types in Prussia, we use grammar schools as well as oberlyceums.

For a graphical illustration of the higher education schools in Prussia see Figure (6).

The use of distance to cloisters builds upon the idea that although cloister were relevant for the emergence of public education in the medieval, their importance declined with the emergence of publicly funded schools and most of the cloisters were dissolved with the secularization around 1800. We hypothesize that if a jurisdiction was close to a cloister before 1800 the probability that the jurisdiction would fund its’ own school was much lower and expect thus a negative correlation between distance to the next cloister and the probability of having a grammar school today.

The location of cloisters in our data set is illustrated in Figure (6). The second problem which is created by the first is the correlation of historical institutions for higher education with the emergence of agglomeration economies, Thus, we include a wide range of agglomeration economies measures in 1990 as control variables to capture this correlation.

To exploit the binary variation whether a grammar school is available or not, we limit the sample to jurisdictions with not more than 1 grammar school and at least 500 and not more than 30,000 inhabitants (in 1990) since this is the size of jurisdictions for which the probability of having a grammar school is between 0

7

The data stems from a publication in the “Die hoeheren Lehranstalten fuer die maennliche

Jugend”’ from 1914. We believe these two school types to be most closely related to grammar

schools today.

Figure 6: Higher Education Schools in Prussia and Cloister.

and 1. Further, to assess the long-run impact we only use data for the year 2008.

We implement the IV strategy by a two-stage least squares estimation. Our IV estimation equation reads:

D(School)

= α + β

D(Prussia)

+ β

D(Cloister)

+ δX

+ φ

(11) Y

= α + β D(School) ˆ

+ γX

+ ω

(12) In a first step, we regress the dummy for the presence of a grammar school on our excluded instruments and the control variables. We then us the predicted value for the presence of a grammar school in a second step, to estimate the causal relationship of the presence of a higher education school on the local economy, measured by the number of employees. Our set of control variables includes the deciles of the population in 1990, the natural logarithm of the population in 1990 within a 50km radius, the ratio of jurisdiction population to the largest city population within

8

Results are similar when using 2007.

50km distance, the natural logarithm of jurisdiction size, area used for buildings and distance to the next highway in 1937. Furthermore, in a robustness check we also include state fixed effects and an indicator variable for the presence of a higher education school before 1990.

Descriptive statistics for the local employment sample are reported in Table (4).

Compared to the DiD sample municipalities are much smaller with 3,630 inhabitants (in 1990) and 276 employees. The reason is that we include jurisdictions with no grammar school in the estimation sample. These are on much smaller than jurisdictions with grammar schools. The average distance to the next highway in 1937 is 14.83km and the average population of the biggest municipality within 50km is about 363,460.

Table 4: Descriptive Statistics (Instrumental Variable Strategy) Local Employment Sample in 2008.

Mean Median Std. Dev.

Employees 274.66 83.00 500.86

Area in km

40.67 27.50 45.11

Area Buildings in km

(1996) 148.07 90.00 161.64

Highway Distance (1937) 14.83 11.80 12.75

Population in 1000 (1990) 3.63 1.81 4.62

Population with 50km distance (1990) 361.71 147.28 754.16 Populatin to Largest Population within 50km distance (1995) 0.03 0.01 0.05

Observations 2,084

Source: Own data collection and calculations.

Descriptive statistics for the house price sample are reported in Table (5). The average house is very similar to the DiD analysis, it is slightly more expensive but comes with more land. The size of the jurisdiction is somewhat smaller, measured by population and area.

V. Results

In the following we present the results of our DiD and IV strategy starting with the first.

A. Difference in Difference Results

Table (6) shows the DiD results. Specifications (1) to (5) use the natural logarithm

of employees working and living in the jurisdiction as the dependent variable and

Table 5: Descriptive Statistics (Instrumental Variable Strategy) House Price Sample for 2004.

Mean Median Std. Dev.

Price 144,191 139,400 72,290

Amount of Land 935.16 663.50 896.53

Living Space 133.62 123.00 48.42

Single Family House 0.84 1.00 0.37

Detached House 0.11 0.00 0.31

Terraced House 0.06 0.00 0.23

Population in 1000 (1990) 7.83 5.72 6.64

Area in km

65.66 46.17 65.05

Distance to Highway in 1937 11.91 7.55 12.47

Observations 62018

Source: Own data collection and calculations.

specification (6) the number of employees working in the jurisdiction. In the first column we do not distinguish between the share of out-commuters in jurisdictions and do not find evidence that school closures affect local employment. From column (2) onwards we account for a potential different impact and our results confirm the presumption. The results suggest that closing a grammar school reduces employ- ment by about 9% in a jurisdiction with a small share of out-commuters but has no impact on employment in a jurisdiction with a large share of out-commuters.

The point estimate changes little when including additional control variable as the local business tax rate or the number of neighboring jurisdictions with a grammar school (column (3)) or state-year fixed effect (column (4)). Further, treatment and control group seem to follow a common trend before the reform as the lead school closures included in column (2) are basically zero (although not statistically sig- nificant). In an additional robustness test, we include up to 4 leads and lags to assess the common trend and timing of the response in more detail (see Figure 8 in Appendix B). Both municipality groups with a high and low share of outgoing commuters show a common trend before the treatment (as point estimates of the leads are insignificant). Further, the results suggest that roughly after 3 to 4 years the negative effect of a school closure has a significant impact on the number of local employees in muncipalities with a low share of outgoing commuters of around 15%.

In column (5) in Table (6) we use the continuous commuting share, which gives

very similar results. Finally, in column (6) we use instead of the narrow defined

local employment, which includes only local residents that work in the jurisdictions,

all employees in the jurisdiction as dependent variable. Results are less precisely

estimated but point estimates are very similar.

Table 6: Baseline Estimation Results for Local Employment.

(ln) Employees (Residents) (ln) Employees (All)

(1) (2) (3) (4) (5) (6)

b/se b/se b/se b/se b/se b/se

Treatment X Treatment Year -0.025 -0.090** -0.087** -0.078* -0.589*** -0.074

(0.027) (0.042) (0.038) (0.040) (0.120) (0.049)

Treatment X Treatment Year X High Commuter 0.092* 0.094** 0.086* 0.109*

(0.053) (0.048) (0.048) (0.056)

Treatment X Treatment Year X Share Commuter 0.882***

(0.179)

School Close (t+1) -0.003

(0.026)

School Close (t+1) X High Commuter -0.022

(0.041)

School Close (t+2) -0.004

(0.024)

School Close (t+2) X High Commuter 0.001

(0.035)

# Schools (0−10 km) -0.012 -0.015 -0.011 0.008

(0.010) (0.010) (0.008) (0.010)

(ln) Local Business Tax in % 0.099 0.029 0.095 0.206*

(0.140) (0.136) (0.139) (0.120)

N 2591 2591 2591 2591 2591 2570

F 37.00*** 34.32*** 34.84*** 32.97*** 35.27*** 18.86***

R sq 0.54 0.54 0.54 0.56 0.54 0.33

Year Fixed Effects X X X X X X

Municipality Fixed Effects X X X X X X

(ln) Population () X Year X X X X X X

(ln) Area X Year X X X X

State Fixed Effects X

Robust standard errors clustered at the municipality level in parenthesis. Significance levels: *p <0.10, **p <0.05, ***p <0.01.

Source:Own data collection and calculations.

One particular strong assumption for the validity of our results is that preferences

for grammar schools are similar in jurisdictions with a large and small commuting

share. Prior literature has found that some individuals are more likely to commute,

e.g. males and richer households. If households sort across space, this would mean

that jurisdictions with a large share of out-commuters may value grammar schools

differently. To assess the plausibility of this presumption we investigate whether

house prices as an indicator of households valuation of grammar schools are similar

effected. The results are reported in Table (7). In the first column, we do not

distinguish between jurisdictions with a small and a large share of out-commuters

and include two leads. The results suggests that there is a price drop already in

t+1 which reflects the role of expectations in house prices. This is consistent with

anecdotal evidence that school closures are announced up to 2 years earlier. We

thus change the treatment effect such that it is one from t+1 onwards. The point

estimate is around 7% which suggest that house price dropped by 7% in response to

an average school closure. In column (3) we test whether the response is different

in jurisdictions with a large and a small share of commuters. The interaction effect

is literally zero and statistical insignificant. Including state-year fixed effects in column (4) increases the coefficient but the interaction effect is further insignificant and close to zero. We conclude that preferences for grammar schools seem to be similar in jurisdiction with a low and high share of commuters as house prices drop to a similar extend. Further, the results confirm that people value the availability of grammar schools within a jurisdiction. The average willingness to pay is about 13,000 euro or roughly 11% of the property value. This seems plausible as pupils go 8 years to grammar school. Thus, using a discount rate of 3% and 1 hour additional time consumed to due the commuting, we derive at a hourly rate between 8 and 9 Euro.

Table 7: Baseline Estimation Results for House Prices.

(ln) House Price

(1) (2) (3) (4)

b/se b/se b/se b/se

School Close (t+2) 0.007 0.007 -0.006 -0.007

(0.020) (0.019) (0.019) (0.028)

School Close (t+2) X High Commuter 0.022 0.013

(0.029) (0.035)

School Close (t+1) -0.070

(0.046)

Treatment X Treatment Year -0.071*

(0.040)

School Close (t+1) or Treatment X Treatment Year -0.070* -0.082* -0.127***

(0.039) (0.041) (0.044) School Close (t+1) or Treatment X Treatment Year X High Commuter 0.019 0.045

(0.062) (0.058)

(ln) Property Tax in % 0.131 0.132 0.130 -0.054

(0.292) (0.291) (0.292) (0.266)

(ln) Local Business Tax in % -0.061 -0.060 -0.060 -0.006

(0.257) (0.257) (0.257) (0.234)

# Schools (0−10 km) -0.005 -0.005 -0.005 -0.009

(0.009) (0.009) (0.009) (0.008)

N 31243 31243 31243 31243

F 87.30*** 90.22*** . .

R sq 0.29 0.29 0.29 0.30

Property Controls X X X X

Year Fixed Effects X X X X

Municipality Fixed Effects X X X X

(ln) Population (1997) X Year X X X X

(ln) Area X Year X X X X

State-Year Fixed Effects X

*p <0.10, **p <0.05, ***p <0.01.Source: Own data collection and calculations.

B. Instrumental Variable Results

In the following we report the results for the long run impact of school closures on

local employment. The results are presented in Table (8). Column (1) presents the

OLS result which gives a positive and significant point estimate. From column (2)

onwards we use the instrumental variable approach. The IV statistics suggest that distance to Cloister is not strong enough although the point estimate is significant on the first stage (column (2)). The instrument quality for the schools in 1914 in Prussia is, however, sufficient (column (3)). Interesting to note is that the point estimates using cloisters and schools respectivley as excluded instrument are almost identical.

In column (4), we account for state-fixed effects but the point estimate is unchanged.

In column (5) we include an indicator which is one if there was higher education school before 1990 in the jurisdiction. The instrument quality decreases strongly and thus standard errors increase, but the point estimate is again almost unchanged. In column (6) we explore to which extend the increase in local employment is offset by a reduction in out-commuters and use overall population as dependent variable.

The point estimate suggests that population increases by only 51%. Thus, roughly 60% of the change in employment is due to less out-commuting.

To validate the analysis further, we assess the long run change in house prices using the same strategy. Results are reported in Table (9) in Appendix C. The OLS estimate is 0.05 and statistically significant. Using distance to cloisters as the excluded instrument, we find a positive and at the 5% level significant point estimate of 0.46, using schools in Prussia the point estimate is insignificant and 0.61. Instrument relevance is, however, only sufficient for cloisters. Including state- fixed effects increases standard errors and decreases the coefficient slightly. When including an indicator for the existence of grammar schools before 1990 results are very similar to the baselines specification. Thus, we conclude that house prices increase in the long run due to grammar school openings by about 46%. This is comparable to the results report by Combes et al. (2012). They find that land prices increased by roughly 0.9% for a 1% percent change in the population.

VI. Conclusion

This article evaluates the importance of local schools for a jurisdictions’ economy.

Motivated by a stylized theoretical model, we investigate the short run impact of school closures on local employment and to which extent the effect is influenced by the share of out-commuters in a jurisdiction. Further, we study the long run impact on local employment using school openings.

As a testing ground, we focus on East Germany, which showed a unique devel-

opment in the schooling landscape due to and after the reunification. We analyze

the short run impact by using a difference-in-differences design and find that em-

Table 8: Long-run effects of Higher Education Schools on Local Employment.

(ln) Employees (living and working) (ln) Population

(1) (2) (3) (4) (5) (6)

Cloister (C) Schools (S) 1914 Schools

b/se b/se b/se b/se b/se b/se

Second Stage:

Gymnasium 0.35*** 1.28 1.18** 1.26* 1.35 0.52*

(0.04) (1.67) (0.56) (0.65) (1.28) (0.27)

Polytechnical (High-) School -0.11

(0.44)

(ln) Area in km² -0.07*** -0.04 -0.04 -0.01 -0.04 -0.06***

(0.02) (0.07) (0.03) (0.04) (0.05) (0.01)

(ln) Area Buildings in km² 0.29*** 0.27*** 0.27*** 0.42*** 0.27*** 0.22***

(0.04) (0.05) (0.04) (0.04) (0.05) (0.02)

(ln) Distance Highway (1937) in km -0.00 0.00 0.00 0.00 0.00 -0.04***

(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)

(ln) Population with 50km Radius 0.16*** 0.12 0.12** 0.09 0.12** 0.21***

(0.05) (0.09) (0.06) (0.06) (0.06) (0.04)

(ln) Population to Largest Population within 50km 0.24*** 0.18 0.18*** 0.13** 0.18*** 0.18***

(0.06) (0.12) (0.06) (0.06) (0.07) (0.04)

First Stage:

Cloister -0.01

(0.01)

Prussia Gymnasium 0.21*** 0.19*** 0.10 0.21***

(0.07) (0.07) (0.08) (0.07)

Prussia Oberlyzeum 0.05 0.05 0.02 0.05

(0.06) (0.06) (0.07) (0.06)

N 2037 2037 2037 2037 2037 2037

F 1087.58*** 814.26*** 857.10*** 698.01*** 739.65*** 2124.21***

Underidentification LM 2.02 6.82 6.23 2.36 6.82

Weak IV 2.01 27.61 24.20 5.87 27.61

Hansen J p-Value 0.29 0.28 0.30 0.30

Year Fixed Effects X X X X X X

Population Percentiles X X X X X X

State Fixed Effects X

*p <0.10, **p <0.05, ***p <0.01.Source:Own data collection and calculations.

ployment decreases by roughly 20% after five years in jurisdictions with a small share of out-commuters. Further, we show that house price decreased by 10% in all jurisdictions with school closures. This rules out that different preferences drive the results and quantifies the valuation of the availability of grammar schools. The reason for the different impact is that commuters seem to be first movers. Thus, if they account for a large share of the inhabitants, their outflow might cause a sufficient drop in house prices.

To assess the long run effects of schools we analyze school openings after the Ger- man re-unification in East Germany. We employ an instrumental variable approach which accounts for omitted variable bias and control for agglomeration economies in 1990. Our excluded instrument is the distance to cloisters before 1800 and higher education schools in Prussia. We find a long run effect in the order of 120%. Further, in line with the theoretical prediction we find that roughly 60% of the additional employment is due to less out-commuting as population increases only by 50%. The additional house price analysis suggests that house prices increased by 50% as well.

Our findings have several implications. First, they show that the centralization

of public good provision does affect employment, in particular in jurisdiction with a

low share of out-commuter. Thus, urbanization may be intensified due to changes in

locally provided public goods. This should be taken into account in the cost benefit

analysis of grammar schools in rural areas. Second, public goods, and in particular,

grammar school may generate agglomeration economies and can thus be used to

foster economic development in rural areas. Third, a large part of the additional

local employment generated by grammar schools is due to less out-commuters and

thus a reduction in congestion costs, which again should be included in any cost-

benefit analysis. Fourth, our results suggest that housing supply changes little in the

long run and thus house prices reflect strongly the availability of grammar schools.

References

Abel, J. R. and Deitz, R. (2011). Do colleges and universities increase their region’s human capital? Journal of Economic Geography, pp. 667–691.

Audretsch, D. B. , H¨ ulsbeck, M. and Lehmann, E. E. (2012). Regional competitiveness, university spillovers, and entrepreneurial activity. Small Business Economics, 39 (3), 587–601.

— and Lehmann, E. E. (2005). Does the knowledge spillover theory of entrepreneurship hold for regions? Research Policy, 34 (8), 1191–1202.

aus dem Moore, J. P. and Spitz-Oener, A. (2012). Bye bye, GI – The impact of the US military drawdown on local german labor markets. Tech. rep.,

Sonderforschungsbereich 649, Humboldt University, Berlin, Germany.

Becker, S. O. , Egger, P. H. and Von Ehrlich, M. (2010). Going nuts: The effect of eu structural funds on regional performance. Journal of Public

Economics, 94 (9), 578–590.

— , — and — (2013). Absorptive capacity and the growth and investment effects of regional transfers: A regression discontinuity design with heterogeneous treatment effects. American Economic Journal: Economic Policy, 5 (4), 29–77.

Beeson, P. E. and Eberts, R. W. (1989). Identifying productivity and amenity effects in interurban wage differentials. The Review of Economics and Statistics, pp. 443–452.

Black, D. , Daniel, K. and Sanders, S. (2002). The impact of economic

conditions on participation in disability programs: Evidence from the coal boom and bust. The American Economic Review, 92 (1), 27–50.

— , McKinnish, T. and Sanders, S. (2005a). The economic impact of the coal boom and bust. The Economic Journal, 115 (503), 449–476.

Black, D. A. , McKinnish, T. G. and Sanders, S. G. (2005b). Tight labor markets and the demand for education: Evidence from the coal boom and bust.

Industrial and Labor Relations Review, 59 (1), 3–16.

Cellini, S. R. , Ferreira, F. and Rothstein, J. (2010). The value of school facility investments: Evidence from a dynamic regression discontinuity design.

The Quarterly Journal of Economics, 125 (1), 215–261.

Chevalier, A. and Marie, O. E. (2013). Economic uncertainty, parental selection, and the criminal activity of the’children of the wall’.

Combes, P.-P. , Duranton, G. and Gobillon, L. (2012). The Costs of

Agglomeration: Land Prices in French Cities. Tech. rep., IZA Discussion Paper

No. 7027.

Denzler, S. and Wolter, S. C. (2011). Too far to go? does distance determine study choices?

Do, C. (2004). The effects of local colleges on the quality of college attended.

Economics of Education Review, 23 (3), 249–257.

Enrico, M. (2011). Local labor markets. Handbook of labor economics, 4, 1237–1313.

Fack, G. and Grenet, J. (2010). When do better schools raise housing prices?

evidence from paris public and private schools. Journal of Public Economics, 94 (1), 59–77.

Freier, R. and Storck, J. (2012). The treatment effect of attending a high-quality school and the influence of unobservables.

Frenette, M. (2006). Too far to go on? distance to school and university participation. Education Economics, 14 (1), 31–58.

— (2009). Do universities benefit local youth? evidence from the creation of new universities. Economics of education review, 28 (3), 318–328.

Garcia-L´ opez, M.- ` A. , Holl, A. and Viladecans-Marsal, E. (2015).

Suburbanization and highways in Spain when the romans and the bourbons still shape its cities. Journal of Urban Economics, 85, 52–67.

Goldstein, H. A. , Maier, G. and Luger, M. (1995). The university as an instrument for economic and business development: Us and european

comparisons. Emerging patterns of social demand and university reform:

Through a glass darkly, pp. 105–133.

Graham, D. (2007). Variable returns to agglomeration and the effect of road traffic congestion. Journal of Urban Economics, 62 (1), 103–120.

— and Kim, H. (2008). An empirical analytical framework for agglomeration economies. ANNALS OF REGIONAL SCIENCE, 42, 267–289.

Greenstone, M. and Moretti, E. (2003). Bidding for Industrial Plants: Does Winning a’Million Dollar Plant’Increase Welfare? Tech. rep., National Bureau of Economic Research.

Greenwood, M. J. , Hunt, G. L. , Rickman, D. S. and Treyz, G. I. (1991).

Migration, regional equilibrium, and the estimation of compensating differentials. The American Economic Review, 81 (5), 1382–1390.

Hymel, K. (2009). Does traffic congestion reduce employment growth? Journal of

Urban Economics, 65 (2), 127 – 135.

Kanafani, A. and Abbas, M. S. (1987). Local air service and economic impact of small airports. Journal of Transportation Engineering, 113 (1), 42–55.

Koh, H.-J. , Riedel, N. and B¨ ohm, T. (2013). Do governments tax agglomeration rents? Journal of Urban Economics, 75, 92–106.

Marchand, J. (2012). Local labor market impacts of energy boom-bust-boom in western canada. Journal of Urban Economics, 71 (1), 165–174.

Melo, P. C. , Graham, D. J. and Noland, R. B. (2009). A meta-analysis of estimates of urban agglomeration economies. Regional Science and Urban Economics, 39 (3), 332 – 342.

Redding, S. J. , Sturm, D. M. and Wolf, N. (2011). History and industry location: Evidence from german airports. Review of Economics and Statistics, 93 (3), 814–831.

Rice, P. , Venables, A. J. and Patacchini, E. (2006). Spatial determinants of productivity: analysis for the regions of great britain. Regional Science and Urban Economics, 36 (6), 727–752.

Spiess, C. K. and Wrohlich, K. (2010). Does distance determine who attends a university in germany? Economics of Education Review, 29 (3), 470–479.

Topel, R. H. (1986). Local labor markets. Journal of Political economy, 94 (3, Part 2), S111–S143.

von Ehrlich, M. and Seidel, T. (2015). The persistent effects of place-based

policy: Evidence from the West-German Zonenrandgebiet. Tech. rep.

Appendix A: Schools

Figure (7) depicts the number of higher education schools in population bins from 0 to 60 thousand.

Note that the number of higher education schools increases strictly with population.

Figure 7: Share of Schools in Municipalities in East Germany.

0.00 0.25 0.50 0.75 1.00

0−5 5−10 10−15 15−20 20−25 25−30 30−35 35−40 40−45 45−50 50−55 55−60 >=60

Population in 1000 (1997)

Share of Municipalities

0 Schools 1 School

2 Schools 3 Schools

4 Schools

>= 5 Schools

Appendix B: Timing of Events

Figure (8) depicts the point estimates and 5% confidence intervals of a regression similar to spec- ification (2) of table (6). In detail, the estimation equation reads as follows

Y

=

m

X

τ=0

δ

D

+

m

X

τ=0

γ

D

×C

+

q

X

τ=1

δ

D

+

q

X

τ=1

γ

D

×C

+X

β +λ

+ν

+

(A.1) with D

a treatment dummy which takes the value 1 if a school closure happened in municipality i in year t. C

is the indicator for municipalities with a high share of outgoing commuter. We control for municipality fixed effects, year fixed effects, year times population (1997) fixed effects and year times area fixed effects. The estimated point estimates for δ (left side) and δ + γ (right side) are depicted in Figure (8). Note that we include 4 leads and lags such that m = q = 4.

Figure 8: Estimation using Leads and Lags.

Appendix C: Instrumental Variable Results for House Prices

Table 9: Long-run effects of Higher Education Schools on House Prices.

(ln) House Price

(1) (2) (3) (4) (5)

Cloister (C) Schools (S) 1914 Cloister (C) Schools

b/se b/se b/se b/se b/se

Second Stage:

Gymnasium 0.06*** 0.49*** 0.40 0.45 0.50***

(0.02) (0.18) (0.40) (0.34) (0.18)

Polytechnical (High-) School -0.11**

(0.05)

(ln) Area in km² -0.08*** -0.06*** -0.06*** -0.06** -0.06***

(0.01) (0.02) (0.02) (0.02) (0.02)

(ln) Area Buildings in km² 0.04* 0.07** 0.06* 0.07* 0.07**

(0.02) (0.03) (0.04) (0.04) (0.03)

(ln) Distance Highway (1937) in km -0.00 -0.01 -0.01 -0.01 -0.00

(0.01) (0.01) (0.01) (0.01) (0.01)

First Stage:

Cloister -0.10*** -0.06** -0.10***

(0.03) (0.03) (0.03)

Prussia Gymnasium 0.11***

(0.04)

Prussia Oberlyzeum 0.06

(0.07)

N 62018 62018 62018 62018 62018

F 282.22*** 264.24*** 272.94*** 257.51*** 267.68***

Underidentification LM 13.32 4.49 3.49 12.39

Weak IV 14.16 5.60 3.62 13.16

Hansen J p-Value 0.31

Year Fixed Effects X X X X X

Population Percentiles X X X X X

Property Characteristics X X X X X

State Fixed Effects X

*p <0.10, **p <0.05, ***p <0.01.Source: Own data collection and calculations.