Estimating the Impact of Enterprise Zones on Commercial Rents Clément Malgouyres July 21, 2011

Research Master in Economics, PhD Track, Sciences Po Paris [email protected] Thesis defended on June 14th 2011.

Under the supervision of Etienne Wasmer, Professor of Economics and Head of Graduate Studies in Economics at Sciences Po

Second reader: Thierry Mayer, Professor of Economics at Sciences Po

1

Abstract Capitalization of local tax cuts into property values can potentially hinder the ability of enterprise zones to stimulate local employment by subsidizing labor. This research aims at i) formalizing the interactions between the real estate market and enterprise zones (EZ) policies and ii) attempts to empirically measure the eect of EZs on the level of commercial rental value. We use geocoded transactions gathered by leading real estate consulting rms on the Parisian regional market to construct several cross-sections from 2000 to 2010. Resorting to a dierence-in-dierence (DD) approach, we nd mixed evidence of the process at work. While baseline specications nd positive and often signicant estimates, the results are very sensitive to the inclusion of a variable accounting for property type (oce versus industrial property). It seems that the apparent increase in rental values caused by enterprise zones is driven by an inated share of transactions pertaining to oces (whose rental value is higher than industrial locals). That change in the composition of the cross-sections makes DD estimates of EZs eects on rental values hard to interpret due to the violation of the constant quality assumption. Keywords: urban policy, enterprise zones, commercial real estate, land markets

1

Introduction

Enterprise zones (EZ) programs have been implemented in many European countries and US states. The common goal of such programs is to stimulate economic activity in places where businesses tend to be reluctant to locate by attracting them with mostly scal incentives. Most of these programs also share the explicit objective to enhance local employment . The mechanism at work is straightforward: by decreasing the cost of local factors of production such as labor, EZ programs aim at increasing local factor demand. In France for instance, Urban Enterprise Zones (UEZ) program requires businesses to conform to a  local hiring clause in order for them to benet from tax reliefs. Exemptions of contributions to social security are granted only if more than a third of a rm's employees live in deprived areas within the agglomeration where the UEZ is located. If demand for labor increases in a given location so will demand for complementary local factors of production notably demand for oor space which is arguably one of the most location-specic production factors and probably a complement with respect to labor.

2

Hence among the set of parameters

likely to inuence EZs' impact on local employment, the elasticity of oor space supply is probably playing an important role.

In case of a very low

elasticity of supply, the EZ-induced shift in local oor space demand is expected to drive rents up, partially o-setting the cost-reducing eect of EZs' tax-cuts.

In the end, increase in the cost of oor space is likely to curb

the boost in local labor demand.

Hence, the low elasticity of commercial

property supply implies that demand will cause prices to rise and therefore, capitalization of exemptions into property values (market prices or rents) could be an explanation of why the literature on EZs has often found these policies to be rather inecient and ineective (Peters and Fisher (2004)). It is therefore relevant to the understand EZ programs' impact on most outcomes of interest, such as local employment, to estimate their impact on the commercial real estate market. This study has two main goals. First it intends to formalize the argument made above in order to deepen our understanding of the interactions between real estate markets and EZ policies. Then it aims at measuring the extent to which EZs exert an upward pressure on the cost of oor space, as measured by rents of commercial property.

Our empirical analysis is based on an

exhaustive dataset of commercial rental transactions in the Parisian region between 2000 and 2010 for oce and industrial properties. Transactions were geocoded in order to determine whether they took place within or outside of an UEZ. Due to the low number of observations in the actual zones, empirical analysis was however carried out at the municipality level. Using a dierence-in-dierence model, we nd some evidence of increases in rents in cities where urban EZs were set up in 2004 and late 2006. In the most complete specications of our model however, we nd no signicant effect of the UEZ policy on the level of rents. We emphasize that these results might stem from the fact that the very low level of observations in the important zones forced us to consider treatment at the municipality level, thus reducing the interest of geocoding transactions and exposing the estimation to downward bias due to poor delineation. This issue could be addressed by working with a dataset covering other French large cities where the policies were led. Another important improvement would be to gather data on small retail businesses. It contributes to the literature on the evaluation of enterprise zones by evaluating a generation of French enterprise zones (labeled in late 2006) that has not been subjected to an econometric evaluation yet. Moreover it is to our knowledge the rst econometric analysis of geocoded commercial real estate data in relation with French urban policy. Therefore descriptive ndings have an interest on their own, even though the dataset was not exactly appropriate to carry out a causal analysis of policies taking place in deprived

3

neighborhoods where there are very few observations. Also it formalizes the interactions between EZ programs and local real estate markets. This study is structured as follows.

In next section, we will describe

the French UEZ policy and proceed to a parsimonious literature review. In Section 3 we provide a simple formal exposition of the eects of EZs on the market for oor space and on how EZ-induced increase in rents could curb the extent to which EZs boost local employment. We also determine what are the main drivers of EZs ecacy are. In section 4, we present turn to the empirical part. We introduce the dataset, provide descriptive statistics and present the econometric model we will run before commenting on the results. Finally, we conclude in section 5.

2

Literature Review and Poliy Description

2.1 Enterprise Zones in France Three dierent types of zones During the 70s and 80s the social and economic decline of some neighborhood emerged as a public policy issue in France.

Concerns about the

rising level of spatial inequalities and the appearance of so-called  ghettos led to the creation of a  City Ministry ( Ministère de la Ville ) in 1991. In 1996, national urban policy was redened and strengthened through a the so-called  Pacte de Relance de la Ville .

Among the new tools mobilized

in order to revitalize blighted neighborhoods, enterprise zones, i.e. localized scal incentives, were implemented for the rst time in France. Three types of zones were designated, ranked according to an increasing order of economic and social diculty: Urban Sensitive Zones (USZ), Urban Revitalization Zones (URZ) and Urban Enterprise Zones (UEZ). Each of these labels is associated with specic policies which we will describe below. 751 neighborhoods were designated as USZ. Among them 416 were classied as URZ. Finally among the 416 URZs, 44 zones were labeled as UEZ in 1997. While the geography of USZ and URZ has remained stable since the 1996 law, there have been two additional waves of UEZ designation: 41 URZ were turned into UEZ in 2004 and 15 more in late 2006. There are currently 93 such zones in continental France (7 in overseas regions). This study focuses on the Parisian region (Île-de-France) where there are 36 such zones.

Designation Process

4

The decision to classify an area as a URZ was based on a so-called synthetic index that sums up the degree of deprivation of a neighborhood. The index's formula is the following:

(T otal population∗Share of less than 25∗U nemployment rate∗Share of school dropouts) F iscal potential Neighborhoods with high index value were granted the status of URZ. Among URZs, neighborhoods of more than ten thousands inhabitants with the highest index values were picked to receive the status of UEZ which is associated with the highest level of local tax exemptions.

While the URZ

selection process seems to have been solely based on the index value (hence we have selection on observables), the UEZs designation process was not as systematic and it is very possible that authorities exhibited a tendency to choose areas with the highest growth potential as has been observed in other contexts (Boarnet & Bogart, 1996). An interesting point for identication however is that in 2006 the population criterion was relaxed and areas comprising only 8500 inhabitants became eligible. This rather arbitrary change in the minimum population criterion allowed new zones to be eligible. Such zones are comparable under other aspects and the discontinuity in the population criterion leads us to believe that 2006 municipalities form a suitable control group when looking at the treatment eect on the cities that received an UEZ in 2004. This has been conrmed by ndings in Rathelot et Sillard (2009) as will be seen in the literature review.

Descriptions of tax incentives granted by the UEZ program The evaluation of EZs in the United States has been complicated by the heterogeneity of programs across cities and states, an issue raised and addressed in Bondonio et Greenbaum (2007). On the contrary, the UEZ program in France is part of national urban policy plan and has been uniformly implemented across the territory. Five types of tax cuts are implemented:

Corporate tax:

10

Companies with less than 50 employees and less than

¿ millions of sales can benet from a full exemption on corporate tax for

5 years. The exemption is then progressively cut back to zero during a 3 or 9 years period depending on the size of the rm's workforce (more or less than 5 employees).

Employers' social contribution to social security:

Companies re-

specting the same criteria as above are eligible for  wage tax cuts within a certain limit (1.4 times the minimum wage) for ve years, under the condi-

5

tion that they respect the  local hiring clause which mandates that at least a third of employees must live in a USZ that is part of the same agglomeration as the UEZ. The exemption is progressively decreased after 5 years along the same lines than for corporate tax.

Tax on prot:

Companies tting the same criteria as above are eligible

for 5 year of prot tax cuts. Exemption is total for the rst two year and progressively decreases for the last 3 years.

Property tax:

All building owned by rm located in a UEZ are ex-

empted from property tax.

URZs comprise some of these tax exemptions

but their intensity is lower. More details on the exact nature of UEZ exemption can be found in Bachelet (2007) and a more accurate description of French urban policies and the state of so-called  sensitive neighborhoods is provided by ONZUS (2010).

2.2 Literature Review There is a large body of literature concerned with the evaluation of enterprise zones.

Many dierent outcome variables have been studied, mainly

employment and stock of establishments. Most studies are concerned with EZs evaluation in the United States or the United Kingdom where EZs were rst implemented in the late 70s. In the context of France, there are two important papers already available respectively evaluating the rst and second generations of EZs, but none are concerned with local real estate markets. Rathelot and Sillard (2009) use propensity score matching to compare economic outcomes (employment and plants inows) in areas that received a UEZ in 2004 with areas that did not but nevertheless benet from a another type of geographically targeted policies named Urban Revitalization Zones (URZ). They nd that employment and plants inows growth rates are both signicantly aected by the UEZ policy but one year after the zoning. They attempt to estimate the same treatment eect through a dierencein-dierence approach, using neighborhoods that were designated a UEZ in late 2006 as a control group. The results from the DD estimation are very similar to those obtained via the propensity score matching which suggests that 2006 UEZs form a valid control group when evaluating 2004 UEZs. The short duration of EZs' eects is a rather general nding in the literature and is also documented in the second econometric evaluation of French UEZ by Gobillon et al. (2010). In this paper, the authors measure the ef-

6

fect of the rst generation of UEZ (labeled in 1996) on job-nding rate of unemployed workers residing in cities that receive an UEZ in 1996.

They

nd a small, signicant but rather short-lived positive eect: the rate out of unemployment increases by 3 percent but that eect vanishes 3 three years after the program. An important contribution of that paper is to look at the eect of EZ on the labor market outcomes of residents of the targeted areas (or their vicinity) as opposed to outside commuters. O'Keefe (2004) also nds that eects of EZs on employment are transitory and vanishes after 6 years. She points to the low availability of commercial real estate as an explanation for the temporary nature of the eects. Hence the dynamics of EZ eects could be related to the state of real estate markets in targeted areas. If this is the case, one would expect the price of oor space to go up, which is an additional motivation to the present study.

O'Kee

carries an evaluation of EZs' impact on wages but nds no eect.

This is

hardly surprising as the mobility of labor probably impedes any capitalization of EZ benets into wages. On the other hand, commercial property being a rather inelastic and immobile factor of production, one would expect stronger eect on commercial rents than on wages. In many econometric EZ evaluations, the unit of observation is larger than the EZ. For example Boarnet and Bogart (1996) and Enberg and Greenbaum (2000) work at the municipality level while Bondonio and Greenbaum's (2007) unit of observation is the ZIP code. That can be problematic for the evaluation of EZs eects as the share of a municipality that is treated can be rather low which most likely implies a downward bias in these estimates (Lynch and Zax (2010)).

For instance there can be negative externalities

whereby EZs areas grow at the expense of their vicinity. In that case using a unit of observation larger than the EZ itself will fail to measure the eect of the EZ but rather will merely capture the net eect of EZ on a given community. A major innovation, even though purely technical, has been to resort to precisely geolocalized data through the use of Geographical Information System (GIS). It has allowed rening the denition of treatment and ensuring that units associated with an EZ are actually included within an EZ. This study by resorting to GIS techniques and geolocalization of transactions was an attempt to overcome the issues mentioned in last paragraph.

Due to

data limitations however we have to operate at the city level, thus losing the advantages associated with geolocalization. Lynch and Zax (2010)resorts to GIS techniques and test whether EZ subsidies aect the proportion of factors used. Depending on the degree of substitutability between capital and labor, EZs' incentives directed towards capital have an ambiguous eect on overall employment.

7

The sign of the

total eect depends indeed on the relative magnitude of output eect versus substitution eect. If the former dominates the latter, then employment will grow and inversely, if substitution more than oset scale eects, employment will diminish.

Most EZs in the US combine labor and capital subsidies ,

even though subsidies directed towards labor should have an unambiguous eect, the overall eect of EZs is still theoretically ambiguous which might contribute to explain the absence of consensus in the empirical literature on EZ evaluation. The simple model presented in section III goes a bit further and introduces another mechanism through which labor subsidies could have an ambiguous eect on employment. A subsidy on labor induces a scale and a substitution eect that both push up the demand for labor. Equilibrium eects on the market for oor space however can entail a third eect: increase in demand for oor space can drive up rents equilibrating supply and demand, thus contributing to higher average cost and, in a increasing costs industry, cut-backs in the level of production and employment. Bondonio and Greenbaum (2007)emphasize the need to disentangle gross from net eects of EZs.

Their results shows that even moderate net in-

creases in establishments are often associated with large inows of new establishments that are partially oset by inated outows of preexisting establishments. The eect of EZs on shut-down probability of existing rms is positive and signicant, so EZs mean net impact might appear negligible but still result in a signicant reorganization of the productive system of targeted areas. An inated level of failure of preexisting plants could be explained by a raise in competition for local customers (competition on the output market) but could also stem from an increase in the cost of local inputs following the arrivals of new, more productive rms (competition on the input markets). Floor space being a very localized input, the hypothesis that EZs lead to increase in rents appear consistent with Bondonio and Greenbaum's ndings. Even when positive eects are found such as in Rathelot and Sillard (2009), cost-benet analysis raises doubts about the eciency of EZs as a tool to temper spatial inequalities and reduce local unemployment. The cost per job created of EZ program is likely to be aected by capitalization of EZ tax cuts into local price factors, as rise in, for instance, price of oor space erodes the cost dierential created by tax cuts. This is an additional motivation to study the eect of EZs on the price of commercial property. The capitalisation of EZ incentives into the value of immobile factors has been focused on the price of real estate and land. One can distinguish two strands in that literature. The rst strand looks at the price of housing and considers an increase as a positive outcome. The hedonic approach to pricing thinks of the market price of a housing unit as the sums of consumers' will-

8

ingness to pay for its dierent attributes. Hence increase in housing prices, keeping private characteristics constant, can be interpreted as an increase in the attractiveness of an area. If they are truly successful, then EZ program should raise local real estate value.

Enberg and Greenbaum (2000)

follow the housing prices in sample of small US cities located in dierent states where some have implanted EZs and some have not. They nd that EZs have in general no impact but that in cities where vacancy rate were low prior to the implementation of the policy, EZs do actually raise prices. Hence vacancy rate, and other indicators of market tightness, seems to be an important covariate to control for and to interact with the treatment. Boarnet and Bogart (1996) use a panel of cities in New Jersey and nd no evidence of an eect of EZs on municipal housing prices. Both studies suer from the fact that they do not resort to geocoded data and look at outcomes aggregated at the city level while EZ are usually just a subset of a city. Hence the estimates of EZs' eects on property values are likely to be biased downwards. A second strand in the literature on EZ and local real estate markers considers that a lift in real estate prices as a result of EZ designation is not necessarily desirable for reasons exposed in the introduction: increase in the price of either real estate or land can go against the EZ's goal of creating a cost dierential in order to attract businesses. That strand is mainly interested in the price of commercial property as opposed to housing. To our knowledge the rst work interested in EZ eect on the value of commercial properties is Erickson and Syms (1985).

Looking at trends

in rental values in the commercial market within the boundaries and in the vicinity of EZs in the United Kingdom, they nd descriptive evidence that EZ programs create a dual property market: while rents inside EZs pick up, they are depressed outside. This suggests some negative externalities of EZs. They estimate that about two thirds of the incentives granted are capitalized into rents of EZ property.

Landers (2006) also investigates impact of

EZs on the commercial property market. He resorts to precisely geocoded microdata which is a clear improvement upon Boarnet and Bogart (1996) and Enberg and Greenbaum (2000).

He does not consider at all however

the issue of endogeneity of EZ treatment, hence it is hard to interpret the estimates as reecting a causal impact. Using several cross-sections of transactions, he simply includes a binary variable for the treatment. Landers nds no signicant eect of EZ on property price, even if it estimates are positive across all specications, with elasticities ranging from 18 to 21 percents. An interesting result is the signicantly negative interaction eect between the treatment and the number of EZs in the county where the property was sold. Hence, a larger supply of EZ land seems to go against the capitalization of tax incentives into EZ property prices (the elasticity is about -13 percents).

9

The present study improves on issues of treatment endogeneity by following a dierence-in-dierence approach and including pre-treatment variables explaining selection into treatment, hence the assumptions under which our estimates can be given a causal interpretation are much less stringent than in Landers' work. Finally in a very recent paper, yet unpublished, Burnes (2011)provides very convincing evidence that EZs programs in California led to an increase in commercial property price. The author's approach follows Neumark and Kolko (2010). EZs were progressively extended in California. They use that fact in order to select as control group areas that were part of EZ before the start of the period of interest and areas that became part of an EZ after the period. The treatment group is the part of EZ that switched from non-EZ to EZ during the sample period. The denition of such groups is made possible by resorting to GIS techniques.

The author runs a dierence-in-dierence

estimation on successive cross-sections of sales transactions and nds a signicantly positive a point estimate of about 6 percent. This result is robust to a large number of robustness checks. The author notably implements a border-matching strategy as in Billings (2009)allowing to control for local unobservable factors determining the attribution of an EZ. The relatively large sample (almost 15000 observations) contrasts with what we have in the case of the Parisian market. It can partly be explained by the large size of the region consider (California as opposed to the sole Parisian region) and also by the relative large size of EZs in California. More than 1.3 million jobs are located in a EZ in California (Neumark et Kolko 2010, p.9) while French EZs account for a much smaller number of jobs .

3

Theoritical Framework

Here we present a simple model motivating the empirical work that follows. This model aims at formalizing the arguments mentioned in the introduction and the literature review regarding the eect of local tax cuts on real estate and land values and the potential impact of tax cuts capitalization on EZs ecacy. We consider an economy where a binding minimum wage results into unemployment even in the presence of inelastic labor supply. We formalize the EZs programs as a decrease in the level of the minimum wage in a given neighborhood.

We analyze how from an equilibrium where the minimum

wage is binding (i.e.

above the market wage) the eect of a drop in the

minimum wage on employment will vary according to parameters of supply and demand on the market for oor space. We choose here to assimilate EZs as a local decrease in the minimum wage

10

because one of French UEZ policy's most important features is to cut in employer social contributions on compensations lower than 1.4 times the French minimum wage. Considering the rather large fraction of workers working for a minimum wage in the targeted neighborhoods, one can consider that the main eect of such cuts is to relax the constraint imposed on businesses by minimum wage.

Moreover as pointed out by Lynch and Zax (2010), sub-

sidies of labor should have a non-ambiguous eect on labor demand while incentives directed towards capital are ambiguous as the sign of their eect on labor demand depends on whether scale eect outweighs substitution effect. Restricting our attention to subsidies on labor, modeled as a decrease in a binding minimum wage, allows us to capture more clearly how the interplay of property and labor markets contribute to shape UEZs' ecacy independently from substitution eects titled towards capital.

3.1 Basic setting The economy is composed of q dierent neighborhood. Each neighborhood is characterized by a given population (Nq ) and a given level of amenities (aq ). There are 3 types of agents: workers, representative rms in a traded good sector, and representative rms in the construction sector. We rst present the settings and then go on to establish a relevant comparative static.

Workers Each individual is identical. She supplies inelastically one unit of labor at the given local wage (wq). Individual utility come from consumption of the composite good (z). Workers in neighborhood q follow a constrained utility maximization program:

Uq = U (zq ) s.t. zq ≤ wq

(1)

The composite good is taken as the numéraire. Optimization yields the indirect utility function

Vq = V (wq ).

Hence the level of welfare of workers is

only a function of their wage (as we consider that housing is provided freely). Workers do not commute (they cannot or the cost of doing so, C, is considered large enough to oset any gains:

C > maxi6=q (wi − wq ).

Moreover

we assume that workers cannot move and are housed in free public projects (which explains why housing is not included in their budget constraint). These are naturally very restrictive hypothesis that it would be interesting to relax in further research.

11

Firms There is a single representative rm in each neighborhood. It produces the composite traded good whose price is determined on the international markets and that we consider as the numéraire. The production technology uses oor space and labor as inputs and exhibits diminishing returns to scale.

Y = F (Sq , Nqd ; aq ) where

Sq

is the quantity of oor space used,

demanded by the rm and

aq

(2)

Nqd

is the quantity of labor

is the level of productive amenity in neighbor-

hood q. The representative rm is assumed to behave competitively on both the output and the input market, that is when it determines the optimal quantities of locally supplied oor space and labor it wishes to purchase, it does not take into account the impact of this decision on input prices (i.e. it behaves as a price taker). We note AC and MC to refer, respectively, to the average and marginal costs of production of the representative rm in neighborhood q. In the long run, both prot maximization and zero prot conditions hold in all neighborhoods. We can write these conditions:

M C(w, r; a; Y ∗ ) = 1 (prof it maximization)

(3)

AC(w, r; a; Y ∗ ) = 1 (f ree entry, zero prof it)

(4)

We add to these conditions that of full employment: in all neighborhoods, the level of production (Yq ) that satises the equilibrium conditions above is high enough so that all of the population is employed in each neighborhood. ∗ We note Nd the contingent input demand function for labor and writing Y the optimal level of production, we can state the full employment condition as:

N d (w, r; a; Y ∗ ) = N

(5)

where N refers to the population of the total neighborhood population and labor supply. Prices of inputs and optimal level of production vary across neighborhoods in order to adjust for the dierent level of amenities.

Construction sector The production function of oce building companies has constant returns to scale and transform land (L) and capital(K) into oor space.

12

In the

neighborhood q, the oce supply is:

Sq = G(Lq , Kq )

(6)

The rm maximizes prot taking r, the price of oor space as given. We write

v

for the price of capital, it is xed across neighborhoods and

determined exogenously on world markets.

We denote

h

the price of land

which varies across neighborhoods. Equating marginal cost to price allows us to derive the supply function of oor space.

∂C (h, v; S) = r ∂S

(7)

where C(h,v;S) is the total cost function of the construction sector. The condition denes the supply function:

S s = S s (r, h, v)

(8)

Zero prot condition is assumed to hold in the long-run. Land being the only truly immobile factor, its price will adjust to bring construction rms prot to zero. We have then the following condition:

rG(L, K) − hL − vK = 0

(9)

The equilibrium on the oce market is reached when demand and supply d are equal. Denoting S the demand for oor space the condition simply writes:

S d (r, w, a, Y ∗ ) = S s (r, h, v) .

Given that Y is chosen to maximize prot we have ∗ where Y is dened by:

(10)

S d = S d (r, w, a, Y ∗ )

∂C (r, w; Y ∗ ) = 1 ∂Y

(11)

3.2 Minimum wage, local unemployment and enterprise zones Now we consider a situation equilibrium where a minimum wage w is imposed which prevents the full employment condition to be fullled for some values of local productive amenities a, that is:

∃a ∈ R++ : ∀a ≤ a, AC(w, r; a; Y ∗ ) > 1, ∀r≥0 13

(12)

The conditions states that the minimum wage is such that given the level of amenities, there is no level of rents that could allow the representative rm to maintain its previous level of employment without making losses. In order not break even the rm has to reduce production (and thus employment) in order not to make losses:

AC(w, r∗∗ ; a; Y ∗∗ ) = 1, with Y ∗∗ < Y ∗

(13)

This fall in production implies local unemployment:

N d = N d (w, r∗∗ ; a; Y ∗∗ ) < N

(14)

We now model the concept of urban enterprise zones as a policy reducing the level of the binding minimum wage. We will investigate how much a marginal change (decrease) in minimum wage will increase employment starting from the equilibrium described in the last equation above. Taking the total derivative of the labor demand with respect to minimum wage we obtain:

−

∂N d ∂N d ∂Y ∗∗ dN d dr∗∗ dN d (w, r∗∗ ; a; Y ∗∗ ) = − − − ∗∗ dw ∂w ∂Y ∗∗ ∂w dr dw

(15)

The right hand side if the equation above has three components. The rst is the partial derivative of labor demand with respect to minimum wage, it is expected to be negative (hence a marginal diminution in the minimum wage is expected to increase labor demand) under very general conditions. It describes a substitution eect. The second term expresses the fact that ∗∗ a change in the minimum wage aects the overall production level Y and ∗∗ that these changes in Y aects the overall level of labor demand in the neighborhood. Hence this term reects a scale of production eect. It is also expected to be negative under very general conditions.

The third term is

the most involved one and its sign is not as obvious as that of the rst. It expresses the extent to which change in the binding minimum wage eects on the level of local labor demand depends on equilibrium eects on the real estate market. That is the product of two derivatives. Labor demand depends on directly on the price of oor space (rst derivative). The price of oor space is itself determined by the market clearing condition between oce demand and supply and therefore moves when the price of other inputs shifts its own demand curve.

Therefore a marginal decrease in the level

of minimum wage, or equivalently a marginal decrease in tax wage (with inelastic labor suply), should increase employment due to substitution and output eects. These two eects might be oset by the reaction of the real

14

estate market as expressed in the third term. The empirical part of this paper is concerned with estimation of the term dr∗∗ that is how much equilibrium rents will increase in reaction to a EZ dw induced shift in inputs price (subsidy to labor). We can decompose this term further and show that under general conditions it will be positive. If we con∗∗ ∗∗ d sider the conditional demand for oor space which we note S (w, r ; a; Y ). In equilibrium we have:

S d (w, r∗∗ ; a; Y ∗∗ ) = S s (r∗∗ , h, v) The partial derivative of scale and output eect.

(16)

S d (w, r∗∗ ; a; Y ∗∗ ) with respect to w includes both

Taking the total dierential of that demand and

supply we get:

dS d (w, r∗∗ ; a) =

dS d dS d (w, r∗∗ ; a; Y ∗∗ )dw + ∗∗ (w, r∗∗ ; a; Y ∗∗ )dr∗∗ dw dr dS s (r∗∗ , h, v) =

dS s ∗∗ (r , h, v))dr∗∗ ∗∗ dr

(17)

(18)

Market clearing condition requires that these two dierentials be equal. Setting them equal and rearranging we get:

dr∗ = dw

dS d dw dS s dr∗∗

−

dS d dr∗∗

(19)

dS s dS d > 0 and dr ∗∗ < 0, that is oor space supply is upward dr∗∗ slopping in price and oor space demand is downward slopping. Hence the We consider

denominator is positive.

The numerator and the total derivative of condi-

tional oor space demand with respect to rent can both be rewritten in terms of scale and output eect:

dS d = Swd + SYd ∗∗ Yw∗∗ Q 0 dw

(20)

dS d = Srd∗∗ + SYd ∗∗ Yr∗∗ ∗∗ < 0 ∗∗ dr

(21)

∂S i . We see from equation 21 above that a marginal change ∂j in minimum wage has an ambiguous eect on the demand for oor space as where

Sji =

subtitution and output eect go in opposite directions. The same hold for dN d the derivative of labor demand with respect to the price of oor space ( ∗∗ ) dr .

15

If the cost function

C(w, r; a; Y ) is of class 2, its cross derivative are equal:

∂ 2C ∂ 2C (w, r; a; Y ) = (w, r; a; Y ) ∂r∂w ∂w∂r

(22)

Moreover following Shephard lemma we know that:

∂C (w, r; a; Y ∂w

) = Nd

and

∂C (w, r; a; Y ∂r

) = Sd

Hence we have

dN d dS d = dw dr

(23)

dN d dr∗∗ ≥0 dr∗∗ dw

(24)

which implies

dr∗∗ is of outmost interest in order to evaluate how feedbacks dw on the commercial property market will aect the ecacy of enterprise zones. h It is more intuitive to express in terms of elasticities. Denoting ek the toh tal elasticity of function h with respect to parameter k and ek the partial The value of

elasticity, we can write: d

∗ erw

=

d

Y εSw + εN Y ∗∗ εw s

d

∗∗

d

(25)

∗∗

εSr∗∗ − (εSr∗∗ + εSr∗∗ εYr∗∗ )

Hence the elasticity of equilibrium rents with respect to the minimum wage will be high (in absolute value) when the supply is inelastic and the demand is (low denominator) or when the demand for oor space is very sensitive to changes in the minimum wage (high numerator) which would typically be the case when there is strong complementarity between inputs. We can also rewrite the eect of a marginal decrease of minimum wage in terms of elasticities: d

d

d

∗∗

d

∗∗

N N Y N r −eN w = −(εw + εY ∗∗ εw ) − er∗∗ ew

(26)

To be illustrate how does EZs eect on unemployment vary with according to elasticity of supply (that we note

a)

16

and the complementarity between

inputs we consider a CES production function for the representative rm of the form:

F (N, S) = (L

σ−1 σ

+K

σ−1 σ

σ

) σ−1 where σ ≥ 0 and 0 < θ < 1

(27)

The associated cost function is :

C(w, r; Y ) = Y j (r1−sv + w1−sv ) 1−sv 1

1

After some manipulations, we can reexpress

r∗

ew as

(28) a function of the un-

derlying parameters: 1−σ

∗

(σ(θ−1)+1)w −erw = − (θ−1)(α+σ)w 1−σ +(α(θ−1)−1)r 1−σ That term turns out to be positive for most plausible values of the parameters. We now describe graphically comparative statics. More results are exposed in the appendix. w r∗ We remark that −ew is locally increasing in σ for low values of . This r can be seen in Figure 1 where r is set equal to 1 (so that w is equal to ∗ w > 1, −erw is the inputs price ratio) and α is equal to 1. However, when r decreasing in σ . That relationship makes intuitive sense: as the ease of substitution increases, the subsidy to labor entails less and less additional purchase of oor space, hence limiting the upward pressure on rental values. Interestingly we remark that it is equal to the inputs price ratiowhen both

α

and

σ

are equal to zero. It implies that all the cost reduction generated

by the decrease in minimum wage is oset by increase in equilibrium rent, without any increase in the level of employment. That case is show in Figure 2with

w = r.

When inputs are stricly complementary, the ecacy of EZs will depend the elasticity of supply and the initial input price ratio. It is equal to:

d

−eN w =

w wr + (σ = 0) (θ − 1)(r + w) ((θ − 1)αw + r(−1 + (θ − 1)α))(r + w)(θ − 1) (29)

which we plot in the plan(w, α) in Figure 3. The price of r is xed at 1. We can see that as long as supply is inelastic, no matter how much labor is overpriced with respect to oor space, EZs will not increase local labor demand. We also not that even for small value of α w when the price ratio is high (high level of the minimum wage), EZs have r a strong potential to increase local labor demand.

17

That section allowed us to see that the term we wish to estimate depends on both rms' technology (elasticity of substitution) and the reactivity of the real estate market (elasticity). In the model as exposed here, we do not take into account general equilibrium feedbacks of EZs policies, notably because we do not allow for the possibility for workers to commute and/or move and we consider how subsidy to labor through a decrease in social contributions on the minimum wage is nanced. Hence there seems to be some scope for interesting extensions. One could introduce workers' commuting and housing consumption, thus modeling the competition of rms and workers for land and how it is aected by scal incentives. In the spirit of Roback (1982), we could rene the analysis of competition for land by allowing workers to move between neighborhoods. By making moving between neighborhood a costly action, one could further investigate the welfare consequences of EZs on local workers depending on how the housing market is regulated (private or public housing), the intuition being that regulations on land and housing markets might introduce some ineciencies but can also allow the workers not to be forced out of their neighborhood due to an increase in competition for land. So if the social welfare function gives a lot of weight to workers living in deprived (low production amenity) area, EZs and public housing could be a welfare maximizing policy mix.

Again, those are only some speculative

statements that we think would be worth formalizing. The model gives us rough guidelines to interpret the empirical estimates we will retrieve. The absence of signicant positive increase could be signal either the absence of increase in labor demand or an elastic supply of oor space. Finding a signicant increase in rents in EZs would imply that EZs work to a certain extent but that feedback on the local real estate market are strongly undermining their eectiveness. As the eect is determined by a complex combination of the primitive parameters of the economy, one could argue that following a more structural approach would lead to more easily interpretable results.

4

Empirical Analysis

4.1 Data We signed a convention with BNP Real Estate Research in order to access a database on oce and industrial property rental transactions. This database contains data on transactions realized by BNP Paribas Real Estate as well as transactions realized by all members of the Immostat Group. Established in 2001 Immostat gathers all transactions on oce properties

18

realized by its members, i.e.

BNP Paribas Real Estate, CB Richard Ellis,

DTZ Jean-Thouard and Jones Lang LaSalle. Moreover it also contains data on transactions published in the specialized written or electronic press as well as leading websites for oce rental operations on the Parisian market. In order to create a homogenous database, the four founding consulting rms have adopted: property type, type of contract, rents, denition of areas taken into account.

All oce transactions and industrial properties

whose areas exceed 5000 square meters are registered. Immostat considers this database to be exhaustive since 2000 for oces and 2004 for industrial properties above 5000 square meters. The outcome of interest we will analyze here are rental values expressed in annual rent per square meters. Using the GIS Mapinfo and the geocoding software of Pitney Bowes, we geocoded about 34000 transactions from 2000 to 2010 at the building block level (Îlot ) that took place in the Parisian region.

We can then construct

11 cross-sections of transactions that can be used to analyze the growth of rental values within UEZs using a dierence-in-dierence approach.

4.2 Descriptive Statistics Table 2 clearly shows that an analysis at the level of the zone is not possible because the number of transactions is very limited. Less than 150 transactions occurred between 2000 and 2010 within the boundaries of UEZs, regardless of the generation to which they belong. This severe data limitation forced us to consider the treatment as being a located in a city that received an UEZ regardless of whether the transaction took place within or outside of the UEZ boundaries. This cancels the advantages of geocoding and forces us to expose our analysis to type of bias mentioned in the literature review. There are reasons to believe that transactions taking place in UEZ are under-sampled, at least in the market for industrial properties. As explained in the presentation of data, the Immostat dataset only contain data on transactions on warehouses and other industrial properties whose size exceeds 5000 square meters. UEZs and cities that contain UEZs tend to have a higher industrial to oce transactions ratio (this is very clear in Table 1) and are accordingly more aected by this censoring than other types of zones. We have no information regarding the constitution of the database that could explain the very low number of transactions on the oce market in UEZs by some sort of similar under-sampling.

We tend to think that it reects

the well-documented specialization of these neighborhoods in sectors intensive in industrial locals, such as construction (Bachelet (2007))and also a low number of jobs per capita in these areas (Gobillon et al. (2010)).

Descriptive Hedonic Model

19

We estimate a descriptive hedonic model in order to document dierences of rents across types of neighborhoods and across types of cities. The model writes as:

log(renti tz) = α + δXi + ηt(i) yeart(i) + βz(i) zonez(i) + θi of f icei +µz(i) zonez(i) of f icei + εitz The subscripts i,

(30)

z(i) and t(i) refer to respectively a transaction, the zone

and the period where the transaction took place.

Xi

is a vector of charac-

teristics specic to the transactions, it includes variables such as surface and type of contract.

zonez(i)

action took place in zone

is a binary variable that is equal to one if the trans-

z.

of f icei

is a binary variable equal to one if the

transaction is an oce as opposed to an industrial property. We consider an interaction term between the type of zone and the type of property because the determinants of industrial and oce rents are likely to be dierent. We estimate two dierent descriptive models. In the rst model, we consider only transactions in cities that contain a zoned-area (a USZ, a URZ or a UEZ) and the variable

zonez(i) is

actually took place in a zone of type

equal to one for transactions that

z.

Hence we document the infra-

municipality variations in rents between the three dierent types of zones using the unzoned part of cities as a reference. We run a second regression where

zonez(i) = 1(transaction i

took place in city comprising a zone of type

z ). Here we document inter-cities variations in rents depending on the type of zone they contain. In case a city has both a USZ and a URZ, it is coded as a  URZ city and when a city has both a URZ and a UEZ, then it is coded as a  UEZ city .

Results are shown in Table 3.

UEZ transactions do not

seem to be much lower than the rest of  zoned cities . We can see in column 2 that oce property in cities with a UEZ have rents signicantly lower than properties of the same type in cities with no zone, the average dierence for

¿ or 0.7 standard deviation. As expected rents for of-

oces is of about 180

ces are higher, but it interesting to notice that the coecient is much larger when looking at all cities than when looking at cities with a zone. It suggests that industrial and oce prices tend to be closer in zoned cities. Cities with USZ and URZ have on average lower rents than the reference group but to a lesser extent than UEZ cities. This is consistent with the idea that USZ, URZ and UEZ targeted areas were ranked by an increasing order of economic deprivation.

Time Series for Control and Treatment Groups In the next section we will estimate the eect on the level of rents when a

20

city is granted an enterprise zone. We rst evaluate the 2004 UEZ generation using cities that received a UEZ in 2006 as a control group. We then attempt to estimate the eect of 2006 UEZ using URZ cities as a control group. Figure 3 shows the evolution of the mean log annual rents for these three groups of cities. Cities that received an UEZ in 2004 have lower rents per square meters than cities that labeled in late 2006 or than cities with a URZ. 2006 UEZ cities were actually granted an UEZ late in fall 2006.

However rents

seem to pick as soon as 2006. This suggests that it will be important to test whether allowing for group specic time trend substantially alter our DD estimates. Figure 4 displays the evolution share of oce in all transactions for the three types of cities. There seems to have been a slight increase of that share in UEZ 2004 cities. UEZ 2006 cities see the share of oce increase before and after treatment until 2008. The generalized decrease from 2008 to 2010 can be explained by the well-documented cyclical behavior of oce markets (Malle 2009), while transactions on the oor space for industrial activities is much less sensitive to economic uctuations.

4.3 Dierence-in-Dierence As was made clear in the descriptive section, carrying out a causal statistical analysis requires to operate at the municipality level because of the very low number of observations.

The baseline model is described in the equation

below.

log(rentitz ) = α + δTz(i) + ηpostt(i),t0 + γpostt(i) Tz(i) + εitz

(31)

Tz is a binary variable that is equal to 1 when a transaction takes place in a city that received the treatment. Postt is a binary variable equal to 1 when t is larger or equal to the treatment date t'. The term

postt Tz

is therefore

equal to 1 only when the transaction takes place after treatment in a city that received the treatment (enactment of an enterprise-zone). parameter of interest.

E(log(renti ) | postt = 0, Tz = 0) = a E(log(renti ) | postt = 1, Tz = 0) = a + η E(log(renti ) | postt = 0, Tz = 1) = a + δ E(log(renti ) | postt = 1, Tz = 1) = a + δ + η + γ

21

γ

is here the

γ = [E(log(renti ) | postt = 1, Tz = 1) − E(log(renti ) | postt = 0, Tz = 1)] (32) − [E(log(renti ) | postt = 1, Tz = 0) − E(log(renti ) | postt = 0, Tz = 0)] Running an ordinary least square on Equation 32 yields the standard dierence-in-dierence (DD) estimate, i.e. an estimate of the average treatment on the treated.

We estimate three main equations in DD. We rst

follow Rathelot and Sillard (2009) and apply DD using the 2006 generation as the control group and 2004 wave as the treatment group. As mentioned in the policy description section, there are reasons to believe that the discontinuity in the population criterion for UEZ eligibility allowed areas very comparable to areas that were labeled UEZ in 2004 to be designated in 2006. Moreover as explained in the literature review, the fact that Rathelot and Sillard (2009) found very similar results when constituting the control group through propensity score matching techniques and when simply using 2006 UEZs as control provides some evidence that 2004 and 2006 areas are comparable. In order to evaluate 2006 UEZs, we use cities with an URZ as a control group. We control for covariates that explain the selection into treatment: unemployment rate in 1999, average scal income, overall population and share of public housing. Two approaches have been used. Including covariates in the regression, and in the spirit of Bondonio and Greenbaum (2007) estimating rst the probability of for a municipality of being treated based on observables through a logit model and including that score as an explanatory variable. It turns out that controlling for pre-treatment dierences at the city level in terms of unemployment, scal revenue, share of public housing and overall population does not inuence signicantly the estimates of treatment eect.

4.4 Results and Comments 2004 UEZs Results for the 2004 generation are presented in Table

4.

Standard errors

are clustered at the municipality level and are computed using classic robust

1

standard errors . The treatment eect in the simplest specication is not signicant but has the expected sign. The point estimates suggests an increase

1 Clustered

standard errors have also been computed using wild bootstrap as exposed in Cameron et al. (2008) and applied in Neumark and Kolko (2010) within the context of EZ evaluation but results are very similar and the change in method of estimation never 22

of about 14 percent following the treatment. Unsurprisingly the coecient associated with the variable UEZ 2004 is negative, reecting dierences in rents prior to the treatment in UEZ 2004 cities than in UEZ 2006 (as was apparent in Figure 4). Allowing for group specic linear trends does not alter signicantly our results (column 2), which provides us with evidence that the common trend assumption is satised. Including surface as a control variable (column 3) does not aect the coecient of interest either. The coecient associated with surface is signicantly negative (but very close to zero) which reects the decreasing cost of unit of oor space with the size of the property being rented. In column 4 we include a variable accounting for property type that is equal to unity when it is an oce and zero when it is an industrial property and we interact that variable with surface.

Treatment eect estimate is

not signicant any longer. The R-square increases strongly, suggesting that property type is an important determinant of rent per square meter. In column 5, we look at the potential heterogeneity of UEZ treatment effect by property type by interacting the property type dummy with the three terms of the baseline specication. We see that the the treatment eect on industrial properties is actually signicantly negative.

The eect on oce

properties is almost null and UEZs seem therefore to increase the relative rental value of oces. Controlling for property type seems natural but one can wonder whether the type of property is not itself partly determined by the UEZ treatment.

In that case, UEZ would have an eect on the share

of oces as a share of all transactions. We test this by estimating a linear probability model in DD, where the outcome variable is equal to unity if a property is an oce and zero if it is not. Results are presented in Table 5. It appears that UEZs in 2004 have a signicant eect on the share of oce property among all transactions: the DD estimate is .25 and signicant at the 5% level. This conrms graphical evidences mentioned in Figure 5.

2006 UEZs We estimate the same equation, this time, using 2006 UEZ cities as a control group and URZ cities as a control group.

There are more reasons

to doubt of the validity of common trend assumption in this case, as URZ cities were never designated to receive a UEZ which suggests that they are fundamentally dierence from UEZ cities. Results are displayed in Table 6.

The basic specication in column 1

seems indicates a positive impact of UEZs on rental values.

Allowing for

cause a coecient of interest to move in or out of the 95 percent condence interval. 23

group specic linear time trends signicantly alter the DD estimate. It becomes much stronger and remains signicant. This suggests that the common trend assumption is not satised, 2006 UEZ cities exhibit a pre-treatment trend with a lower slope, hence causing baseline DD estimates to understate the treatment eect. We account for group specic quadratic trend in column 3 but this does not modify substantially the DD estimate. So we only use the linear trend in the additional specications we try. Because URZ cities were never chosen to receive a UEZ it is crucial to account for dierences in characteristics on which selection into treatment may have been decided.

We could directly control for city characteristics

that are relevant to selection into treatment (these variables are listed in Table 1) or use these variables to estimate the probability of a city to receive a UEZ in 2006 conditionally on being a URZ city and use that propensity score as a covariate. It turns out both options have little impact on the estimated treatment eect. We choose to display here the former one. Column 4 shows that the inclusion of city level covariates does not matter a great deal as far as the estimation of the treatment eect is concerned. Column 6 we introduce property type and, once again, we see that introducing an oce binary variable removes any signicance from the treatment eect coecient. Contrary to the 2004 UEZ generation, including interaction term between treatment and oce indicator does not lead to dierentiated treatment eect along property type as neither overall treatment eect estimate nor the interaction coecient is signicantly dierent from zero, as can be seen in Column 7. Running a basic dierence-in-dierence estimation of a linear probability model where the dependant variable is the oce binary variable we nd a strong eect of the policy on that share.

As displayed in Table 7 the

coecient is signicant and positive, with a point estimate varying between 0.27 and 0.30. We can therefore see that UEZ zoning entails an increase in the share of transactions pertaining to oces as opposed to industrial properties. As oces are rented for a much higher rents on average a change in the share of oce is sucient to drive the overall increase in the average rental value of transactions. That increase could be explained by the fact that it is easier and less costly for rms to move their oces rather than their warehouses. Bondonio and Greenbaum (2007) show that EZs overall resulted in an increase in the turn-over of rms: quasi-null growth in the number of establishments, but signicant amplications of the numbers of entries and exits. In case this  turn-over eect is stronger for establishments occupying ofces than for those based in industrial property, EZs have the potential to increase average rental values by simply boosting the oce share of transac-

24

tions, while the overall distribution across types of the property stock in the EZ remain stable. That makes the dierence-in-dierence based successive cross-sections of transactions not appropriate to capture the true extent to which UEZs cause rents to increase. A change in the composition of property type among transactions could be considered to be positive outcome if it truly reects a change of a municipality's sectoral specialization.

However if it operates only through a

stronger turnover eect among oce-intensive companies, this would be a poor proxy for economic recomposition. In case there is indeed a turn-over eect, we should see an increase in the share of second-hand buildings go up among transactions taking place in UEZ cities in comparison with URZ cities.

Figure 6 seems to provide some evidence supporting the idea of an

increase in the second-hand share in 2006 UEZ cities in comparison to URZ cities. An implementable improvement apt to curtail the issue of the moving of oces would be to resort to the matching method based on a matching algorithm in the spirit of that used in the last section of Burnes (2011). It would consist of constituting a control group by rst matching perfectly on property type and submarket (one could think of department in the case of the Parisian market or other submarkets as dened by BNP Paribas Re-

2

search ).

Then the second stage would be to match imperfectly based on

covariate such as physical distance, surface, age.

5

Conclusions

In this study, we present the relevance of estimating the eect of EZs on the commercial property rental values in the overall evaluation of EZ policies. We saw that the reactions of local real estate markets to EZs can potential hinder EZs ability to reach their goals in terms of promoting local employment and isolated the key parameters likely to drive such reactions, that is the degree of complementarity between commercial property and the inputs being subsidized (labor in our model) and the price elasticity of oce space supply. Then we estimate the change in market rents following the implmentation of EZs using a dataset on transactions in the Parisian market and resorting to GIS techniques to precisely documenting transactions' locations. The low

2 BNP

Paribas Real Estate divided the commercial real estate market in Île-de-France into several sub-markets: La Défense, Paris Central Business District, Paris Outside CDB, Paris South ...

25

number of observations obliged us however to proceed at the city level, thus curtailing the advantages associated with GIS techniques. The timing of successive EZ extensions allows us to use a dierence-indierence approach.

Baselines specications of our model when estimated

nd signicant and positive eects.

Such ndings however are not robust

to the inclusion of property type as a covariate.

UEZs seem to induce a

positive eect on the relative number of oce transactions and oce transactions exhibit higher rental values.

This can explain why positive eects

are not robust to the inclusion of property type.

We face here a typical

violation of the constant property assumption which in the end makes the standard dierence-in-dierence not amenable to estimate the causal eect of treatment. Two main improvements could be brought to solve the issues faced by our empirical study. We could collect data on other large French cities where identical UEZs where implemented in order to alleviate the issues associated with the low number of observations within UEZ boundaries.

Second, we

could adopt a cross-sectional matching approach. We could then circumvent the issues associated with varying proportions of property types by perfectly matching transaction on property type. This dataset is not well suited to the study of EZ due to the low number of observations in the area of investigation. However it seems to have a great potential to investigate many economic urban issues such as the eect of infrastructure or local taxation on rental values (e.g. Wheaton and Sivitanidou (1992)).

References Bachelet, M., 2007. Les zones franches urbaines en 2005: : des embauches encore fortement concentrées dans les anciennes zfu. Premières synthèses 26.1. Billings, S., 2009. Do enterprise zones work? Public Finance Review 37 (1), 6893. URL

http://pfr.sagepub.com/content/37/1/68.abstract

Boarnet, M. G., Bogart, W. T., 1996. Enterprise zones and employment: Evidence from new jersey. Journal of Urban Economics 40 (2), 198215. URL

http://econpapers.repec.org/RePEc:eee:juecon:v:40:y:1996:i:2:p:198-215

Bondonio, D., Greenbaum, R. T., January 2007. Do local tax incentives aect economic growth? what mean impacts miss in the analysis of enter-

26

prise zone policies. Regional Science and Urban Economics 37 (1), 121136. URL

http://ideas.repec.org/a/eee/regeco/v37y2007i1p121-136.html

Burnes, P. D., March 2011. An empirical analysis of the capitalization of enterprise zone tax incentives into commercial property values, job Market Paper. Cameron, A. C., Gelbach, J. B., Miller, D. L., 2008. Bootstrap-based improvements for inference with clustered errors. The Review of Economics and Statistics 90 (3), 414427. URL

http://econpapers.repec.org/RePEc:tpr:restat:v:90:y:2008:i:3:p:414-427

Enberg, J., Greenbaum, R., 2000. State enterprise zones and local housing markets. Journal of Housing Research 10, 163187. Erickson, R., Syms, P., 1985. The eects of enterprise zones on local property markets. Regional Studies 20, 114. Gobillon, L., Magnac, T., Selod, H., Oct. 2010. Do unemployed workers benet from enterprise zones?

the french experience. CEPR Discussion

Papers 8084, C.E.P.R. Discussion Papers. URL

http://ideas.repec.org/p/cpr/ceprdp/8084.html

Landers, J., 2006. Why don't enterprise zones work? estimates of the extent that ez benets are capitalized into property values. Journal of Regional Analysis and Policy 36(1), 1530. Lynch, D., Zax, J. S., 2010. Incidence and substitution in enterprise zone programs: The case of colorado. Public Finance Review. URL

http://pfr.sagepub.com/content/early/2010/11/04/1091142110386210.abstract

Malles, R., 2009. Modélisation des loyers de bureaux : le cas de la région parisienne. Ph.D. thesis, Université Paris Dauphine. Neumark, D., Kolko, J., July 2010. Do enterprise zones create jobs? evidence from california's enterprise zone program. Journal of Urban Economics 68 (1), 119. URL

http://ideas.repec.org/a/eee/juecon/v68y2010i1p1-19.html

O'Keefe, S., January 2004. Job creation in california's enterprise zones: a comparison using a propensity score matching model. Journal of Urban Economics 55 (1), 131150. URL

http://ideas.repec.org/a/eee/juecon/v55y2004i1p131-150.html

27

ONZUS, 2010. Rapport 2010 de l'onzus. Peters, A., Fisher, P., 2004. The failures of economic development incentives. Journal of the American Planning Association 70(1), 2737. Rathelot, R., Sillard, P., 2009. Zones franches urbaines :

quels eets sur

l'emploi salarié et les créations d'établissements ? Economie et Statistique 415-416, 8196. Sivitanidou, R., Wheaton, W. C., March 1992. Wage and rent capitalization in the commercial real estate market. Journal of Urban Economics 31 (2), 206229. URL

http://ideas.repec.org/a/eee/juecon/v31y1992i2p206-229.html

Thélot, H., 2006. Les zones franches urbaines en 2004 :

lancement de 41

nouvelles zones. Premières Synthèses 6.2.

6

Appendix

6.1 Comparative Statics of Rent Elasticity with Respect to Subsidy to Labor In this subsection we present a few relevant comparative statics regarding the elasticity of equilibrium rents with respect to a marginal decrease in minimum wage in order to complete the graphical arguments given in the body of the text. As above, they are established in the context of a CES production function.

The elasticity of interest is equal to:

1−σ

∗

(σ(θ−1)+1)w −erw = − (θ−1)(α+σ)w 1−σ +(α(θ−1)−1) We now look at how it varies with respect to the other parameters: a) Elasticity of supply ∗∗

−

derw dα

=

(σ(−1+θ)+1)w1−σ ((−1+θ)w1−σ −1+θ) . ((−1+θ)(α+σ)w1−σ −1+(−1+θ)α)2

For a given value of

w/r

et

θ,

this term is negative for low values for

28

σ.

σ   ∗∗ derw sign − dα Comments:

0

σ ∗∗ =

+

0

1 1−θ

+∞ -

The sign does not depend on the value of

α

b) Initial input price ratio ∗∗

derw − dw For

1−σ

(−1+(−1+θ)α)(σ(−1+θ)+1)(−1+σ) = ww((−1+θ)(α+σ)∗w 1−σ −1+(−1+θ)∗α)2 σ = α = 0, this term is equal to 1.

For

σ = 1,

this term is equal to

0, that is elasticity of equilibrium rents does not depend on the initial input price ratio but only on

α.

c) Elasticity of substitution ∗∗

derw − dσ

((−θ+1)p1−σ +(1−σ(1−θ)ln(p)−θ+1)p1−σ (−1+(−1+θ)α) ((−1+θ)(α+σ)p1−σ −1+(−1+θ)α)2 ∗∗ There is a value ofσ , denoted σ such that this term is equal to 0.

=

θ

pends on

and

w/r

but is indedendent of

σ ∗∗

de-

α.

((−θ+1)w1−σ +(1−σ(1−θ)ln(w)−θ+1)w1−σ (−1+(−1+θ)α) ((−1+θ)(α+σ)w1−σ −1+(−1+θ)α)2

=0

⇒ ((−θ + 1)w1−σ + (1 − σ(1 − θ)ln(w) − θ + 1)w1−σ (−1 + (−1 + θ)α) = 0 | {z } <0 (as α≥0 and 0<θ<1)

(as the denominator is always positive)

⇔((−θ + 1)w1−σ + (1 − σ(1 − θ))ln(w) − θ + 1)w1−σ = 0 α ≥ 0) ⇔(−θ + 1)w1−σ + (1 − σ(1 − θ))ln(w) − θ + 1 = 0

(as we assume

(as we assume

w > 0)

∗∗

−

derw dσ

T (w)

Q 0 ⇔ T (w) = (1 − θ)w1−σ + (1 − σ(1 − θ))ln(w) − θ + 1 R 0 is monotonic in ∗∗

w

so that

T (w) = 0

denes the couples

derw that − = 0. T (w) is increasing or decreasing in dσ relative value of σ and w .

w

(w, σ)

such

depending on the

In the production function, labor and oor space have the same coe-

29

cient (namely 1), hence for the minimum wage to be binding we need at least

w r

> 1.

Now, it is clear that when

∗

w > 1, −erw is

decreasing in

σ.

∗∗

−

derw dσ

|w=1 =

((−θ+1)−θ(−1+(−1+θ)α) ((−1+θ)(α+σ)−1+(−1+θ)α)2

By assumption ∗∗

derw − dσ

<0

for all

θ ∈]0, 1[ (i.e.

Q0⇔θQ1

decreasing returns to scale) , hence we have:

w > 1.

6.2 Figures and Tables 6.2.1 Figures

30

Figure 1: Elasticity of in the

(w, σ)

equilibrium rents wrt decrease in minimum wage

plane

Comments: Values of other parameters:α = r = 1. The elasticity generally decreases in σ except for very low values of the the input ratio wr . The elasticity increases with almost linearily in wr . In the limiting case α = σ = 0, the elasticity is simply equal to wr (Cf. Figure2).

Figure 2: Elasticity of in the

(α, σ)

equilibrium rents wrt decrease in minimum wage

plane

Comments: Values of other parameters: wr = 1. With strict complementarity and inelastic supply, elasticity of equilibirum rents wrt decrease in minimun wage is equal to the input price ratio. Feedback on the real estate market in that case osets all EZ-induced costs savings thus preventing local labor demand to grow (Cf. Figure3).

31

Figure 3: Elasticity of in the

(w, α)

local labor demand wrt decrease in minimum wage

plane (strict complementarity)

Comments: Value of the other parameters: σ = 0, r = 1. We see that no matter how much minimum wage are binding, in case of inelastic supply (α = 0), the eect of EZs on local labor demand is null. However when supply is even moderately elastic (e.g. α = 0), EZ's eects are strongly increasing in w/r.

32

Figure 4: Average rents in 2004 and 2006 UEZs waves and in URZs

33

Figure 5: Share of oce transactions among all transactions, UEZ and URZ cities

Comments : Each transaction is weighted by property surface.

34

Figure 6: Share of second hand transactions

Comments: Transactions are not weighted by surface.

35

6.2.2 Tables

36

Vacancy Estimated Vacancy Share of rate in probability Unemployment Average fiscal rate public commercial of Type of income (1999) housing housing property* designation city Statistics Population rate Mean 55579,8 0,167 13160,6 0,111 0,409 8,181 0,246 Sd 22209,1 0,028 1348,1 0,029 0,085 9,778 0,127 UEZ 2004 p10 35004 0,130 11165,7 0,074 0,350 1,200 0,156 (N=16) p50 46492 0,167 13394,1 0,116 0,393 4,100 0,188 p90 91536 0,189 13948,3 0,153 0,505 31,600 0,438 Mean 44735,500 0,190 11842,4 0,085 0,455 4,170 0,374 Sd 13940,122 0,034 1526,9 0,028 0,112 2,581 0,181 UEZ 2006 p10 25795,5 0,134 10248,3 0,060 0,323 1,250 0,150 (N=10) p50 42975 0,200 11480,1 0,076 0,413 4,500 0,393 p90 61332 0,226 14058,8 0,127 0,589 7,900 0,633 Mean 41697 0,137 14773,0 0,082 0,423 3,218 0,149 Sd 22023 0,032 2368,01 0,026 0,120 3,059 0,154 URZ p10 19048 0,104 12308,1 0,060 0,279 0,300 0,017 (N=22) p50 35436 0,134 14382,01 0,072 0,414 2,350 0,094 p90 79324 0,183 17958,6 0,124 0,605 6,000 0,298 mean 66741,3 0,143 15052,3 0,089 0,371 6,647 0,160 Sd 53248,9 0,035 3449,8 0,032 0,158 12,364 0,139 USZ p10 11284 0,099 12170,7 0,051 0,154 0,600 0,018 (N=52) p50 48796,5 0,147 14599,2 0,085 0,361 2,800 0,143 p90 170076 0,190 17790,0 0,134 0,605 12,500 0,312 Mean 26310 0,089 24335,3 0,078 0,184 8,536 0,022 Sd 30121,2 0,028 9817,1 0,038 0,124 38,374 0,054 No zone 3991 0,062 15759,370 0,041 0,030 0,600 0,000 (N=154) p10 p50 19256 0,085 21480,9 0,070 0,173 2,300 0,002 p90 51497 0,127 36558,370 0,125 0,362 12,400 0,063 *Figures are from the year 1999 except for vacancy rate in the commercial property market which are for the years 2000 and 2001 The estimated probability of treatment refers to the probability for a municipality to receive a UEZ during the year 2004 or 2006 modeled by a logit including all the other variables described in this table. Sources : French census 1999, BNP Paribas Immostat Table 1: City level descriptive statistics

37

Table 2: Distribution of transactions across zones from 2000 to 2010

UEZ URZ USZ Not in a zone Total

Office market Frequence Percentage 71 0.29 77 0.31 588 2.39

Other types (industrial) Frequence Percentage 76 1.58 1,124 23.42 1,311 27.31

23,883

97.01

2,289

47.69

24,619

100.00

4,800

100.00

Comments: There is not enough transactions to carry analysis at the inframunicipal level.

38

Table 3: Descriptive hedonic model

Variables UEZ URZ USZ Office (industrial is the reference group) Office and UEZ Office and URZ Office and USZ New building (2nd hand reference) Surface Surface and Office Surface Squared Constant

(1) Rents -6.4 (10.81) -1.517 (10.40) 31.03 (22.44) 124.5*** (25.17) -43.76 (42.64) -86.90*** (27.67) -15.20 (33.71) 19.71*** (6.233) -0.0033* (0.0019) 0.0104*** (0.00215) -4.75e-08

(2) Rents -6.72* (3.97) 4.9 (5.58) 53.95 (39.17) 231.6*** (28.97) -172.6*** (30.85) -157.7*** (30.90) -122.1** (53.46) 50.73*** (13.13) 0.0003 (0.0011) 0.0082*** (0.00111) -1.04e-07***

94.25*** (4.950)

82.21*** (8.174)

Observations 9,263 27,391 Adjusted R-squared 0.062 0.210 Clustered robust standard errors in parentheses (60 clusters for column 1 and 352 clusters for column 2) *** p<0.01, ** p<0.05, * p<0.1

39

Table 4:

DD estimation, UEZs 2004 vs UEZ 2006 cities:

Y = ln(annual rents) UEZ 2004 and Post 2004 Post 2004 UEZ 2004

log (annuel

(1) 0.132 (0.084) -0.138 (0.081) -0.348*** (0.074)

(2) 0.141 (0.112) -0.138 (0.082) -0.343*** (0.082)

(3) 0.144 (0.112) -0.149* (0.083) -0.349*** (0.085) -0.000** (0.000)

(4) -0.019 (0.087) 0.086 (0.055) -0.226*** (0.050) -0.000** (0.000) 0.486*** (0.034) 0.000** (0.000)

(5) -0.210** (0.090) 0.266*** (0.056) 0.015 (0.094) -0.000** (0.000) 0.748*** (0.085) 0.000*** (0.000) -0.326*** (0.093) -0.215*** (0.071) 0.227** (0.080)

No

Yes

Yes

Yes

Yes

Surface Office Office and surface Office and UEZ 2004 Office and Post 2004 Office, Post 2004 and UEZ 2004 Group specific trends

Observations 782 782 782 782 782 Adjusted R-squared 0.096 0.096 0.118 0.463 0.475 Robust standard errors clustered at the municipality level in parentheses *** p-value<0.01, ** p-value<0.05, * p-value<0.1 Difference in difference OLS estimates of UEZ treatment effect on commercial rental values. Treated group contains the cities that received a UEZ in 2004. The control group is composed of the cities that received one in late 2006.Time period goes from 2000 to 2006 included.

40

Table

5:

for

property

a

DD

estimation

of

to

oce

be

an

the

eect (as

UEZ 2004 and Post 2004 Post 2004 UEZ 2004 Linear trend group specific

of

UEZ

opposed

to

(1) Property indicator 0.258** (0.109) -0.357*** (0.066) -0.127 (0.093) Yes

on an

the

probability

industrial

(1) Property indicator 0.241** (0.125) -0.361*** (0.063) -0.129 (0.089) Yes

Sales and rental operations Yes Only rental Observations 929 782 Adjusted R-squared 0.105 0.105 Clustered robust standard errors at the municipality level (21 clusters), *** p<0.01, ** p<0.05, * p<0.1 Sample period is 2000-2006 which can be problematic as there is a change in the coverage rate of industrial properties between 2003 and 2004. If that change however is uniform across groups, it should pose no problem to the estimation.

41

prop-

Office and Post 2006

Office and UEZ 2006

Office

Office and surface

Surface

UEZ 2006

Post 20061

Y = ln(annual rents) UEZ 2006 and Post 2006

42

(2) 0.268*** (0.0946) 0.0629* (0.0365) -0.0981 (0.115)

(3) 0.236** (0.0926) 0.0629* (0.0365) -0.0746 (0.123)

(4) 0.222** (0.0880) 0.0816*** (0.0290) 0.155* (0.0842)

(5) 0.215** (0.0884) 0.0824*** (0.0279) 0.158* (0.0828) 1.84e-05** (8.96e-06)

(6) -0.0689 (0.0595) 0.118*** (0.0252) -0.0154 (0.0732) -2.74e-05*** (8.59e-06) 6.74e-05*** (1.24e-05) 0.467*** (0.0460)

Group specific linear trend No Yes Yes Yes Yes Yes Group specific quadratic No No Yes No No No trend No No No Yes Yes Yes Control for pre-treatment city characteristics2 N 3,338 3,338 3,338 3,338 3,338 3,338 Adjusted R-squared 0.0207 0.022 0.023 0.264 0.274 0.343 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 1 Post 2006 takes the value 1 from 2007 included onwards, because UEZ were enforced in December 2006. 2 Pretreatment city characteristics include: unemployment rate, share of public housing, housing vacancy rate, population, average household fiscal revenue all values for 1999.

Office, Post 2006 and UEZ 2006

(1) 0.1009** (0.0426) 0.0629* (0.0365) -0.2062* (0.1121)

3,338 0.5028

Yes

Yes No

(7) -0.125 (0.0867) 0.140*** (0.0305) -0.149 (0.115) -2.78e-05*** (8.54e-06) 6.80e-05*** (1.25e-05) 0.453*** (0.0605) -0.0326 (0.0393) 0.166 (0.0991) -0.0269 (0.0726) Table 6: DD estimation, UEZs 2006 vs UEZ 2004 cities: log (annuel rents)

Table 7: Linear probability model for oce indicator: UEZ 2006 vs URZ cities, DD estimation

UEZ 2006 and Post 2007 Post 2007 UEZ 2006 Estimated treatment probability Group specific linear trend

(1) Property indicator

(2) Property indicator

(3) Property indicator

0.3081*** (0.0799) 0.0313 (0.0316) -0.0858 (0.1596)

0.2906*** (0.0781) 0.0313 (0.0316) -0.2760 (0.3401)

No

Yes

0.2706*** (0.0853) 0.0312 (0.0304) -0.1878 (0.3396) -0.3127 (0.2344) Yes

Group specific quadratic No No Yes trend N 2,362 2,362 2,362 R-squared 0.0171 0.0172 0.0374 Clustered robust standard errors at the municipality level (36 clusters) *** p<0.01, ** p<0.05, * p<0.1 The sample period in this table is 2004 to 2010. The reason we do not include pre-2004 is there is a change in the coverage of transactions pertaining to industrial properties.

43