Urban Public Transportation and Firm Location Choice
Evidence from the Regional Express Rail of Paris Metropolitan area
Thierry Mayer∗ and Corentin Trevien†
October 4, 2013
Preliminary version, please do not cite
Abstract This paper identifies the causal impact of urban rail transport on firm location and employ- ment in the Paris metropolitan region. The evaluation of transport infrastructures always faces an important issue of endogeneity since rail lines are not located randomly. We use the natural experiment offered by the Regional Express Rail (RER) progressively opened between 1970s and 2000s. We show that ignoring the endogeneity issue lead to overestimate the treatment effect. We find that the increase in the number of firms due to a station opening lies between 8 and 15 % in the 1975-1990 period. Places located between 15 and 30 km from Paris are the most affected by the RER. The estimated impact is stronger for foreign firms. Despite we find no effect on the total population, our results suggest the commissioning of the RER may have increased the competition for land and the share of wealthier households in the population.
JEL Codes: D04, H43, R42
∗SciencesPo., 28 rue des Saints-Pères, 75007 Paris, France, [email protected] †Insee-SciencesPo.-Crest, 15 bd Gabriel Péri, 92240 Malakoff, France, [email protected]
1 Introduction
The evaluation of transportation infrastructures appears to be a major topic as the construc- tion of roads, railway lines or airports represents a substantial public spending. At the same time, many transportation investments are planned, for example, the "Crossrail" project in London, the "Grand Paris" in France or the Obama plan for high speed rail in the United-States. Policymakers generally argue that such infrastructures are priority steps for the economic development of their region or country. This assertion has to be evaluated in the light of rigorous empirical works. This article aims to estimate the causal impact of Paris urban transportation systems on firm location, employment and population. It proposes an original identification strategy based on a natural experiment provided by the implementation of a fast commuter rail system, the Regional Express Rail (RER)1. The RER has been progressively opened from the 1970s to the 1990s (see figure 1). It has currently reached 587 km all over the Paris metropolitan region. Finally, it provides empirical evidence on the way transports influence firm behavior.
The evaluation of transport systems faces an unavoidable problem of endogeneity. In fact, new infrastructures are obviously not randomly assigned. For example, urban planners are likely to connect in priority economically promising areas. They can also use transport infrastructure as a tool for urban renewal and thus link deprived areas. The consequence is that a naive estimation of the effect of a new infrastructure which would compare directly connected to unconnected areas would be biased. The sign of this bias remains unknown as the intentions and anticipations of the urban planners are not observed.
The literature offers some examples of identification strategies that can address this major en- dogeneity issue and provide the unbiased effect of new infrastructures. Econometricians have used natural experiments to identify the causal effect of transport infrastructures. Duranton and Turner (2012) evaluate the impact of the highway network in the United-States on the local evolution of employment. They use an instrumental variable strategy, based upon a 1947 plan of the interstate highway system and a 1898 map of railroads, to address the endogeneity of the highways location in 1980. Michaels (2008) also uses the 1947 plan as an exogenous variation of road for trade on interstate trade. Donaldson (2010) shows that the Indian railway extension led to a decrease in interregional trade costs and increases both incomes and trade. To do so, he uses a natural experiment provided by 40,000 km of planned lines which were never built for exogenous reasons. Banerjee et al. (2012) find a moderate positive effect of transportation access on income growth in China. They use the fact that railroad lines were built in China to connect European concessions on the coast and inland historical cities in the 19th century. They bring out that crossed areas, which were located in between these two sorts of cities were "quasi-randomly" linked to the railway network and can be compared to similar unconnected areas.
On the other hand, other papers study the determinants of firm location, including trans- portation, but they do not use a natural experiment to address the endogenity issue of transport infrastructure location. All these works highlight a positive impact of roads (Coughlin and Segev, 2000; Gibbons et al., 2012; Holl, 2004a,b), rail (Cheng and Kwan, 2000; Kang and Lee, 2007) or airports (Strauss-Kahn and Vives, 2009) on firm location. Data availability and precision is a key issue to accurately estimate the impact of transportation on firms. Gibbons et al. (2012) insist on the fact that such an evaluation is more complicated in developed countries as transportation networks are already widely extended. In this conditions, it is necessary to measure not only if a
1Réseau express régional in French
2 given area is linked to a network but also how well it is connected. To do so, it is necessary to have journey time data.
This article combines a rigorous approach of transport infrastructure evaluation with a stan- dard frame of firm location choice. More concretely, we consider both domestic and foreign firm location choice across municipalities of the Paris metropolitan region. To compute the treatment effect of RER, we use the proximity to a commuter train station and an accessibility index based on travel time by public transport. We provide two strategies to address the endogeneity issue we previously reported. The first one is similar to Banerjee et al. (2012). Indeed, the RER net- work has been implemented with the aim of connecting "new towns" located 30 km away from the historic center of the city. Commuter train happens to cross municipalities which are located between the historical core and the five new towns of the metropolitan area. These municipalities are "quasi-randomly" treated and can be compared to similar untreated municipalities in order to estimate the causal impact of railway infrastructures. The second strategy is closer to Duranton and Turner (2012) and relies on significant differences between the initial plan presented in 1965 and the current network.
We find that ignoring the endogeneity issue lead to overestimate the treatment effect. We find that the increase in the number of firms due to a station opening lies between 8 and 15 % in the 1975-1990 period. Places located between 15 and 30 km from Paris are the most affected by the RER. The estimated impact is stronger for foreign firms. Despite we find no effect on the total population, our results suggest the commissioning of the RER may have increased the competition for land and the share of wealthy households in the population.
The first section of this article presents the datasets used for estimations. Then, the second section details the model and the estimation strategy. The third section sets out the results and finally, the fourth section concludes and discusses our findings.
1 Data description
Our dataset provides information on firm, employment and socio-demographic features of Paris metropolitan region at the municipality level. We also build new data to precisely describe the evo- lution of urban transportation system between the 1970s and the 2000s. Municipality is the more accurate geographical scale available since more precise data are computed only since the 1990s. However, French municipalities are particularly small in comparison with other European countries. The Paris metropolitan region contains 1300 municipalities which proves to be an adapted geo- graphical scale for this type of estimations.
Our data give information on firms between 1974 and 2004, including the industry sector, the location at a municipality level, the opening and closing years. Unfortunately, employment at firm level is not accurate enough before 1993 to be used in this study. Note also that until the 1990s, foreigner investors had to register every acquisition of French firms with the treasury. A dataset of foreign direct investment (FDI) has been computed using this administrative requirement. So, we have data on shareholding including the nationality of the stakeholder and the proportion of capital held by foreigners. As this compulsory registration has been phased out in the 1990s, reliable data on FDI are available only until 1994. We use census data to know the population, the employment, the unemployment rate and the social composition of municipalities at each census year in 1968,
3 1975, 1982, 1990, 1999 and 2009.
We constructed new data on the transportation system of Paris metropolitan area. They include the precise surface of each municipality within 500 meters of a train or metro station, each year between 1969 and 2009. We also computed the travel time by rail between every municipality for the same period of time. It is an essential information, as the RER conversion of existing lines led to mixed improvement of the transportation service quality. This travel time data are used to compute an accessibility index.
2 Identification strategy and econometric model
Our identification strategy need to address the endogeneity issue, in order to provide the un- biased effect of the RER on firm location, employment and population. The main problem of transportation system evaluation is that new infrastructures are obviously not randomly located. The consequence is that a naive estimation which would directly compare connected to uncon- nected areas will be biased. As explained in the introduction, the sign of this bias remains unknown.
To address the endogeneity issue, we obviously do not select all municipalities but only a sub- sample for which treatment can be regarded as exogenous. Our method relies on two strategies which exhibit two different subsamples. For each of these strategies, we exhibit a control group and a treatment group. We argue that location of transport infrastructures can be seen as "quasi- random" across groups. Finally, we compute a difference-in-difference estimate to provide the effect of RER station opening on various local variables. The first strategy is similar to Banerjee et al. (2012) and compares existing lines which were upgraded thanks to the RER policy with other unimproved lines. The second strategy uses the differences between the 1965 initial urban master plan and the commuter train network actually implemented as Duranton and Turner (2012). We compare municipalities equipped with a RER station but which should not have been equipped according to the initial plan to the opposite case (municipalities not equipped with a RER but which should have been). In that case, we consider that treatment was driven by exogenous tech- nical reasons. We do not include municipalities for which the initial plan and the current network coincide.
2.1 Why Paris needed a new commuter rail system? In the early 1960s, Paris metropolitan region was growing rapidly. Public infrastructures prove insufficient to meet the demographic increase demands. Population of the metropolitan region2 rose from 6.6 million in 1945 to 8.5 million in 1962 and 9.9 million in 1975. To avoid an uncon- trolled spread of the city, the central Government finalized a plan to reorganize the region and guide its development. It included the redistribution of administrative boundaries, the building of housing, especially in nine "new towns", 30 km away from Paris, the opening of highway and rail lines. New towns designate planned cities located 30km away from Paris in relatively undeveloped areas. The goal of this policy was to disperse population away from Paris urban core and address urban congestion issues. Most of these new policies were presented in the "SDAURP" in 19653.
2"Île-de-France" administrative region population 3Schéma directeur d’aménagement et d’urbanisme de la Région Parisienne in French
4 The communter train network was clearly inadequate for the development of the city, since it had hardly evolved for 60 years. The French railways have been mostly built during the 19th century by private companies which obtained concessions of the French Government to link French cities to Paris. This organization has had a long-lasting impact on the geographical shape of the railway network. French railway system is highly centralized and the majority of lines is directed towards Paris. When the network was built, each private company was in charge to develop rail infrastructure in a specific part of France. Competing companies poorly connected their networks together which often end in a dead-end station in Paris. As a consequence, circular tracks were scarcer than radial ones, especially in the Paris metropolitan region (see figure 1). As main lines, all commuter lines ended in Paris and did not allow direct travels from East to West or North to South in the Paris Metropolitan region. Suburban commuters may have to change train several times. The French railway network had been unified under state control in 1937 but a lack of investment in infrastructure allowed this situation to persist. Concretely, considering the case of a commuter who lives in the north and works in the south of the Paris metropolitan area, he or she had to take a train to a first train station, then use the metro inside Paris to reach the second train station and finally take another commuter train to his or her final destination. The organization of the commuter rail network was clearly suboptimal. In addition, some lines were still served by steam trains in the 1960s.
Even if car played a central role in projects of 1960s urban planners, the "SDAURP" plan men- tioned a very ambitious commuter rail system, the so-called Regional Expresse Rail. This project corresponded to a rapid and high capacity network. It planned the construction of tunnels under Paris to connect isolated parts of the city to the rest of the metropolitan area and the building of hundreds of kilometers of news railways across the agglomeration. This new rail system was supposed to link the city of Paris, which is limited to the historical core of the agglomeration, to the major subcenters of the Paris metropolitan area (see figure 2). This subcenters include mainly the nine "new towns", the two airports and the business district of "La défense".
Because of the 1970s crisis, the implementation of the initial urban plan has been more modest and principally consisted in upgrading existing railroad lines, connected together with tunnels under the historical city core of Paris (see figure 3). Only five news towns were actually built, new line construction has been far scarcer than initially planned and major parts of the 1965 plan has been dropped (see figure 2.3). RER project finally included the construction of new branch lines to- wards airports and "new towns" not connected by an existing line, the commissioning of new trains and higher frequencies. Despite limited new track segments, the RER finally led to a significant improvement of the commuter rail network and reached the goals assigned by the 1965 plan to connect the five new towns and both airports. Coming back to the previous example, the journey from North to South is easier thanks to the Regional Express Rail. Only one train change is needed, in one of the new underground RER stations in the city center, instead of two before the opening of the RER service. According to our simulations, the median travel time to Paris4 was 68 min- utes in 1969, for all the municipalities of Paris metropolitan region with at least one train station. Between 1969 and 2005 it decreased by 5.3 minutes for municipalities connected to the Regional express rail while it decreased only by 2 minutes in municipalities left apart from the new network. Thus, RER station opening offers a 5 percent push in commuting time for connected municipalities.
4Travel time to Paris is the mean of the 20 travel times to the 20 boroughs of Paris.
5 1960 1990
PARIS PARIS
1975 2005
PARIS PARIS
Commuter rail line RER line Department border
Figure 1: Evolution of commuter train network and regional express rail between 1965 and 2005
Roissy
La Défense PARIS Marne−la−Vallée
St.−Quentin−en−Yvelines
Orly
New line Evry Line to be upgraded
Melun−Sénart
Figure 2: The SDAURP plan for Regional Express Rail in 1965
Cergy−Pontoise
Roissy
La Défense PARIS Marne−la−Vallée
St.−Quentin−en−Yvelines
Orly
Existing line Evry Reopened line New line
Melun−Sénart
Figure 3: New and reopened lines in Paris metropolitan region since 1969
7 Cergy−Pontoise
● Roissy
La Défense PARIS Marne−la−Vallée
St.−Quentin−en−Yvelines
●Orly
Initially unplaned line Evry Canceled project Conform to initial plan ● New town ● Airport Melun−Sénart
Figure 4: Comparison of the 1965 master plan and the current Regional Express Rail network
2.2 First strategy: intermediate municipalities The first strategy focuses on intermediate cities. As stated before, the RER network have been implemented with the aim of connecting "new towns" located 30 km away from the histor- ical center of Paris. As they link the five new towns of the metropolitan area to the historical core, commuter trains happen to cross municipalities located in-between. These municipalities are exogenously treated and can be compared to similar untreated municipalities in order to estimate the causal impact of railway infrastructures. In addition, RER implementation mainly consist in upgrading existing lines and stations. Only some existing stations have been upgraded to RER stations, while others remain served only by commuter train (see figure 1). Finally, intermediate stations have been transformed in RER stations only if they were located in-between the histori- cal city center and new economic centers (new towns, airports, business district) or for technical reasons.
In the end, our first strategy selects municipalities located in-between Paris and other subcen- ters, which were already connected to the commuter rail network in 1969. The treatment group includes municipalities within 500 meters of a station and the control group includes municipalities within 500 meters of a commuter rail station out of the RER network. The treatment can be con- sidered as "quasi random" across the two groups since the new network aimed at connecting new economic centers to Paris and not specifically these municipalities. Termini stations are logically excluded from both groups as they are explicitly targeted by the RER policy. The treatment is clearly not exogenous in that case. By termini station, we mean the historic city of Paris and mu- nicipalities which are part of a "new town", host an airport or the business district of "La Défense" (see figure 5). In addition, it is impossible to find a proper counterfactual for these municipalities. For that reason, they are also excluded for the second set of estimations. Indeed, there is no new
8 Cergy−Pontoise
Roissy
La Défense PARIS Marne−la−Vallée
St.−Quentin−en−Yvelines
Orly
Airport Evry New town La Défense Metro line
Melun−Sénart
Figure 5: Excluded municipalities for estimations
town, airport, historic city center or business district which have not been connected to the RER in the Paris metropolitan region. Municipalities connected to the underground network are also removed not to interfere with the effect of the RER.
Considering the RER network have progressively spread over Paris metropolitan region, the treatment group enlarges over time while control group becomes smaller (see figure 6). We also see there are very few municipalities in the control group close to the center of the region after 1990. Therefore, it seems impossible to use our identification strategy thereafter. That is the reason why our estimations focus principally on the 1975-1990 period. We also note that the control group municipalities are smaller in population, total job and firm number (see table 1). We address this issue by computing a log-linear matching model.
2.3 Second strategy: initial plan and actual network The second strategy uses the substantial differences between the initial plan presented in 1965 and the actual network (see figure ). From this perspective, municipalities can be divided in three categories. Firstly, the control group includes municipalities not equipped with a RER but which should have been according to the initial project. The 1965 project planned hundreds of kilometers of new railways. Because of the 1970s crisis, the central Government gave priority to upgrading existing lines and limited railway construction to the most necessary cases. For exogenous bud- getary issues, these municipalities did not accommodate a new station. Secondly, we include municipalities equipped with a RER station but which should not have been equipped according to the initial plan in the treatment group. Because of the 1965 draft revision, some existing lines have benefited from a quite unexpected improvement. As existing lines had been built in the 19th
9 1990 − 1st Strategy 1990 − 2nd Strategy
PARIS PARIS
2005 − 1st Strategy 2005 − 2nd Strategy
PARIS PARIS
Legend − 1st Strategy Legend − 2nd Strategy Control group Control group Treatment group Treatment group Commuter rail line Initially unplaned line RER line Canceled project Conform to initial plan
Figure 6: Control and treatment group for both strategies
10 Table 1: Comparison of control and treatment groups Group 1 Group 2 Treated Untreated Treated Untreated Population1975 7547 20921 12684 19532 Firms1975 221 524 353 499 Employment1975 1611 4852 2804 4477 timeP aris 1975 79 64 69 66 dP aris 37 20 19 23 Number of municipality 216 107 36 65
century, there was no mean to manipulate the route of the RER lines for the Government. In addition, urban planning policies were highly centralized until the 1982 decentralization law. Lo- cal authorities had a limited influence on the RER project definition, route choice thus has been driven by technical and budgetary reasons. That is why we consider the treatment was driven by exogenous technical reasons. The third group includes municipalities municipalities for which the initial plan and the current network coincide. Obviously, we do not use them for our estimations because of endogeneity reasons. The coherence between plan and realization proves they were intentionally equipped with a RER station (see figure 6).
Note that we apply the same restriction to the sample as for the first group: municipality served by the underground, new towns and airports are excluded. There are less municipalities selected for the second strategy (see table 1) but there are less differences between control and treatement group.
2.4 The econometric method The aim of this study is to estimate δ the causal effect of a RER station opening in a munici- pality on employment, population and firm location. In the previous section, we describe two sets of treatment and control groups for which RER opening can be regarded as exogenous. In this section, we describe a difference-in-difference matching method to compute this causal estimate. We consider the municipality growth ∆ log(Yit+1) = log(Yit+1) − log(Yit ) is driven by the opening of a RER station between t and t + 1 and some initial parameters Xit . For population and em- ployment data, as we mostly uses census data, we compute a long-difference estimator on only one period, so there is no need to estimate time fixed effects. Our baseline estimates covers the 1990-1975 period. Initial condition variables are required because long-difference estimator only address constant over time disparities. While it is clear that the local growth rate varies according to the distance to Paris, the other transportation infrastructure or the initial density in employment and population of the municipality.
∆ log(Yit+1) = δ∆RERit+1 + βXit + αt + it For firm location choice, data are available for each year. For easy comparison, we keep the same period with a slightly different specification. The depend variable is the difference in the number of new firms in the municipality i between year 1975 and t + 1 and we consider year 1975 for initial conditions. In that case, time fixed effects are required.
11 Because initial condition variables may be not enough to account for the gap between control and treatment group, we add a matching method to our difference-in-difference framework. We follow Imbens and Wooldridge (2008) and we weight observations with the propensity score. It is estimated with a logit model as the probability for a municipality to get a RER station given the density in population and employment in 1968, the proximity to a highway, the distance to Paris and the cardinal position relative to Paris5. Thank to this matching method, municipalities which have a very low probability to be connected to the RER network account for little in the estimations. For example, the control group includes many small municipalities that are located at the rural fringes of the metropolitan region that may bias the treatment effect. The estimation of the propensity score is presented in the appendix (see table 9 page 23).
The central assumption of difference-in-differences is that each control group would have grown in the same proportions as the treatment groups in absence of the treatment. To test for the common trend assumption, we provide a placebo test. To do so, we restrict the treatment group to municipalities that received a RER station between 1975 and 1990 by excluding the few mu- nicipalities that were treated between 1969 and 1990. We apply the econometric model to the 1968-1975 period to be sure there is no ex-ante trend gap between treatment and control group. The placebo test is successful as we do not find any significant difference between the control and treatment group for both strategies before 1975 (see table 7 page 19).
Thanks to our identification strategies, there might be no unobserved phenomenon correlated to the treatment. In a word, the whole gap we find between control and treatment group is due to the treatment. However, the fundamental assumption of difference-in-differences could be partially rejected for another reason. There may be actually two effects of RER station opening: attractiveness and displacement. If employment increases in a treated municipality due to a dis- placement effect, it necessarily implies a decrease for other municipalities. If employment increases due to attractiveness effect, untreated municipalities would not be affected. In that case, the RER network increases the total firm number of the region. Obviously, the reality is in between. To discriminate between the two effects in the case of firms, Schmidheiny and Brülhart (2011) sug- gest to use a nested logit firm location model can separate the respective share of relocation and attractiveness. The estimation of such a model requires individual firm data and outside options of firms (for example, the rest of France and other European countries). Because we do not have such data, we can isolate neither the effect of the relocation of firm or population between treated and untreated municipalities within Paris metropolitan area, nor the attraction of external firms or inhabitants which would not have located in the metropolitan area without the RER construction. In facts, the control group could have been negatively affected by the treatment and could have evolved differently in absence of the RER commissioning. Consequently, the result we get could be overestimated and the true treatment effect lies between the half of the estimate and the provided value.
2.5 The treatment variable The Regional Express Rail have improved commuter train service in many ways. That is the reason why we introduce three types of treatment variables RERi,t to take into account all the improvement aspects. The first type corresponds to the presence of a Regional Express Rail station in a municipality. More precisely, the variable corresponds to the total surface of the municipality
5Cardinal position is a key factor to compare growth trends as South and West parts of the Paris metropolitan region are wealthiest that North and East.
12 which is located within 500 meters of a RER station, normalized to one for a single station.6 The second type of treatment variable is the travel time to Paris. The travel time to Paris is the mean of the 20 travel times to the 20 boroughs of Paris. Between 1969 and 2005 it decreased by 5.3 minutes for municipalities connected to the Regional express rail while it decreased only by 2 minutes in municipalities left apart from the new network. The third type of variable is ac- cessability index. It measures the accessibility Ait of a given municipality i at time t to a given opportunity Xjt (population, enterprises, etc.) in other municipalities j given the travel time by train dij between i and j. We tried three types of functions f suggested by the literature: ex- −α −α ponential with f (Xjt , tij , α) = exp(tij Xjt ), inverse f (Xjt , tij , α) = tij Xjt or indicator function f (Xjt , tij , α) = 1(dij 6 α) × Xjt . The indicator function offers a simple interpretation: it is the total amount of opportunity X reachable within α minutes but the choice of α is difficult. We retained the inverse function, which is the most widespread, with α = 1, since it gave similar results.
X Ait = f (Xjt , tij , α) with α > 0 j The RER variable is crossed with distance to Paris to estimate the heterogenous effect of transportation with respect to location.
3 Results
3.1 Treatment effect estimation Table 2 shows how important it is to estimate a well-specified model for transportation effect estimation. The results of our two identification methods are reported in columns (4) and (5): a RER station opening causes an increase between 8 and 15 percent of employment of a municipality between 1975 and 1990. The two methods yield slightly different results. According to the first strategy, the RER effect on employment is positive for places located within 30 km from Paris. According to the second strategy, the effect on employment is limited to municipalities located between 15 and 30 km from Paris. In a certain way, the Regional Express Rail contributed to job decentralization in the Paris metropolitan region. Indeed, it fostered the location of economic activities in the most developing places given their distance from Paris. Besides, we find a stronger treatment effect with the second strategy. This could be due to the fact that the first strategy compares municipalities connected to the RER network to municipalities with a commuter rail sta- tion whereas the second strategy compares them to municipality without any mass transit. With this in mind, it seems logical the gap between control and treatment is greater according to the second strategy.
The first column presents the most naive way to estimate the parameter of interest. It includes all the municipalities of the Paris metropolitan region, with no control for initial conditions. The treatment effect is clearly overestimated. Control variables and matching decrease the treatment effect but displace it further away from Paris. Column (2) and (3) results show that ignoring endogeneity issues causes an overestimation of the interest parameter. It seems to indicate that the Regional Express Rail connected in priority economically dynamic municipalities. This is totaly in line with the stated objective of 1965 plan to connect Paris and new economics subcenters of
6The variable is divided by the surface of a disk of radius 500 meters, in such a way that it is equal to one if the whole station surroundings are located in the same municipality.
13 Table 2: Effect of RER on employment at the municipality level
∆ log Employ1975−90 (1) (2) (3) (4) (5) Intercept 0.203∗∗∗ −1.131∗∗ −1.496∗∗∗ −1.52∗∗∗ −1.826∗∗ (0.019) (0.477) (0.412) (0.442) (0.753) ∗∗ RER1975−90 × 1(dP aris < −0.021 0.004 0.017 0.093 0.016 15km) (0.034) (0.029) (0.023) (0.042) (0.069) ∗∗∗ ∗ ∗∗ RER1975−90 × 1(15km < 0.312 0.012 0.027 0.079 0.152 (0.095) (0.098) (0.052) (0.047) (0.061) dP aris < 30km) ∗∗ ∗∗ RER1975−90 × 1(dP aris > 0.083 0.146 0.152 0.048 0.138 30km) (0.054) (0.064) (0.059) (0.055) (0.116) ∗∗∗ Train1990 −0.003 −0.002 −0.081 −0.044 (0.029) (0.021) (0.029) (0.038) ∗∗∗ ∗∗∗ ∗∗ Highway1990 0.318 0.216 0.197 0.263 (0.081) (0.064) (0.079) (0.161) log(Surface) 0.019 0.131∗∗∗ 0.13∗∗∗ 0.163∗∗ (0.029) (0.03) (0.033) (0.068) ∗∗∗ ∗∗∗ 1(15km < dP aris < 0.431 0.136 0.079 −0.079 30km) (0.061) (0.051) (0.056) (0.102) ∗∗ 1(dP aris > 30km) 0.16 −0.113 0.067 −0.116 (0.081) (0.086) (0.075) (0.181) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ log(Employment1975) −0.32 −0.173 −0.215 −0.228 (0.039) (0.032) (0.034) (0.052) ∗∗∗ ∗∗ ∗∗∗ ∗ log(Popu1975) 0.354 0.098 0.141 0.129 (0.045) (0.042) (0.047) (0.066) Number of observations 1250 1239 1233 319 99 Sample Complete Complete Complete Strategy 1 Strategy 2 Weighted by prop. score XXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
the metropolitan region together.
Table 3 indicates that the positive effect of the Regional Express Rail is also valid for firms. We apply our methodology to both firm stock and flow and we get estimates of a similar magnitude. Surprisingly, the effect of a station opening on the location of new firm is the strongest for places located at more than 30 km from Paris. We do not get any similar result for employment and firm stock. It suggests that RER stimulates firm creation and relocation in this places but does not cause any lasting effect on economic activity. Besides, we find no effect on population at all. It indicates that firms may be more selective than inhabitants when they choose their location. We notice that the coefficient associated with presence of a highway is also insignificant for population while it is for firms.
Table 4 shows our estimates broken down by industry. We note that treatment effect are not always consistent across groups. Concerning the first four regressions, strategy one logically yields the more accurate estimates since it has three times more observations. Because of these discrepancies, result by industry must be interpreted carefully. They show the RER affects build- ing and service industry within 15 km from Paris. However manufacturing firms, which probably require more space to set up, are sensitive to the commitment of a RER station further from Paris.
Table 6 isolates the effect of RER on foreign-owned firms. The effect is positive and very strong to other firms within 30 km from Paris. We even notice a negative effect further away from Paris, which suggests that there is a displacement effect. Even if we find discrepancies between
14 Table 3: Effect of RER on population and firms at the municipality level
∆ log ∆ log ∆ log Firm location 1975−90 Firms1975−90 Population1975−90 (1) (2) (3) (4) (5) (6) Intercept −3.72∗∗∗ −2.644∗∗∗ −1.658∗∗∗ −1∗ 0.265 0.487 (0.199) (0.213) (0.34) (0.533) (0.238) (0.366) RER1975−90 × 1(dP aris < −0.015 0.013 0.032 0.009 −0.005 −0.016 15km) (0.015) (0.018) (0.024) (0.038) (0.02) (0.031) ∗∗ ∗∗∗ ∗∗ ∗∗ RER1975−90 × 1(15km < 0.026 0.08 0.055 0.106 −0.013 0.004 (0.013) (0.019) (0.026) (0.041) (0.03) (0.057) dP aris < 30km) ∗ ∗∗ RER1975−90 × 1(dP aris > 0.07 0.08 0.027 0.081 −0.04 −0.066 30km) (0.042) (0.039) (0.062) (0.101) (0.026) (0.065) ∗∗∗ ∗ ∗∗∗ Train1990 −0.029 −0.022 −0.055 −0.042 −0.008 0.024 (0.011) (0.012) (0.021) (0.026) (0.016) (0.019) ∗∗∗ ∗∗∗ ∗∗∗ Highway1990 0.092 0.105 0.133 0.1 −0.012 −0.085 (0.022) (0.039) (0.05) (0.1) (0.036) (0.056) log(Surface) 0.132∗∗∗ 0.049∗∗ 0.093∗∗∗ 0.082∗ 0.037∗∗ 0.044 (0.014) (0.019) (0.025) (0.042) (0.018) (0.029) ∗∗ ∗ ∗ 1(15km < dP aris < 0.04 0.035 0.088 −0.02 0.071 −0.008 30km) (0.02) (0.032) (0.046) (0.076) (0.037) (0.052) ∗∗∗ ∗ 1(dP aris > 30km) −0.162 −0.016 0.045 −0.095 0.077 −0.001 (0.049) (0.065) (0.058) (0.143) (0.045) (0.107) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ log(Popu1975) 0.366 0.374 0.335 0.243 −0.083 −0.112 (0.024) (0.032) (0.033) (0.054) (0.013) (0.03) ∗∗∗ ∗∗∗ log(Firm location1975) −0.904 −0.824 (0.019) (0.024) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ log(Firms1975) 0.413 0.337 −0.438 −0.363 (0.028) (0.038) (0.034) (0.047) Number of observations 1663 1007 321 99 321 99 Time fixed effects XX Sample Strategy 1 Strategy 2 Strategy 1 Strategy 2 Strategy 1 Strategy 2 Weighted by prop. score XXXXXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
the two strategies, these results suggest that the proximity of Paris is stronger for foreign-owned firm.
Table 5 provides interesting elements on competition for space. Sadly, we do not observe neither population income nor housing prices. However, we argue that the proportion of executives and professionals in both employment and population provides a good approximation. This cat- egory includes a significant proportion of qualified workers and wealthy households in the French classification of occupations. Whereas we did not find any effect of the RER on population, we highlight a significant impact on the share of executives and professionals in population within 15 km from Paris. It suggest that well connected areas well-served by public transport become more attractive for households that are likely to pay more for their housing. However, we do not find any evidence of such effect on the share of executives and professionals in employment.
If we compare this results with previous ones, we notice that we find no clear evidence that the RER induced a strong increase in the gross number of employment, firm or inhabitants within 15 km of Paris. Nevertheless, we observe that certain types of activity or inhabitants are likely to respond to improved accessibility in these places. Municipalities located within 15 km of Paris happened to be already well developed in the 1960s. In the first group, population grew by 3 %
15 Table 4: Effect of RER on employment by industry at the municipality level
∆ log ∆ log ∆ log Building 1975−90 Manufacturing1975−90 Services1975−90 (1) (2) (3) (4) (5) (6) Intercept −2.775∗∗∗ −3.562∗∗ −2.64∗∗∗ −3.572∗∗ −1.491∗∗∗ −1.483∗∗ (1.025) (1.468) (0.785) (1.677) (0.482) (0.702) ∗∗ ∗ ∗ RER1975−90 × 1(dP aris < 0.239 0.229 0.073 0.052 0.073 −0.005 15km) (0.097) (0.136) (0.074) (0.124) (0.04) (0.065) ∗ ∗∗ RER1975−90 × 1(15km < 0.009 −0.041 0.168 0.163 0.068 0.174 (0.079) (0.11) (0.087) (0.161) (0.047) (0.069) dP aris < 30km) RER1975−90 × 1(dP aris > −0.088 −0.085 0.014 0.117 0.046 0.087 30km) (0.153) (0.296) (0.146) (0.321) (0.077) (0.149) ∗∗∗ ∗∗ Train1990 −0.061 −0.094 −0.186 −0.115 −0.059 −0.027 (0.056) (0.063) (0.065) (0.113) (0.029) (0.033) ∗∗ ∗ ∗∗∗ Highway1990 0.208 0.253 0.283 0.521 0.192 0.189 (0.171) (0.26) (0.127) (0.268) (0.07) (0.132) log(Surface) 0.093 0.141 0.23∗∗∗ 0.259∗∗ 0.114∗∗∗ 0.114∗ (0.07) (0.115) (0.061) (0.123) (0.032) (0.063) ∗ 1(15km < dP aris < 0.139 0.233 −0.03 0.008 0.102 −0.095 30km) (0.124) (0.207) (0.113) (0.239) (0.057) (0.09) 1(dP aris > 30km) 0.19 0.149 −0.114 −0.09 0.117 −0.012 (0.158) (0.399) (0.141) (0.388) (0.076) (0.2) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ log(Popu1975) 0.345 0.282 −0.044 0.068 0.155 0.159 (0.081) (0.103) (0.106) (0.177) (0.049) (0.059) ∗ ∗∗ ∗∗∗ ∗∗ log(Employment1975) 0.196 0.337 0.149 0.102 0.157 0.216 (0.119) (0.154) (0.122) (0.236) (0.053) (0.097) ∗∗∗ ∗∗∗ log(Emp building1975) −0.656 −0.731 (0.142) (0.197) log(Emp −0.264∗∗∗ −0.335∗∗ (0.065) (0.137) manufacturing1975) ∗∗∗ ∗∗∗ log(Emp services1975) −0.369 −0.434 (0.067) (0.101) Number of observations 303 96 299 98 317 99 Sample Strategy 1 Strategy 2 Strategy 1 Strategy 2 Strategy 1 Strategy 2 Weighted by prop. score XXXXXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
and employment grew by 40 % between 1968 and 1990 for both control and treatment group, while they grew respectively by 78 % and 40 % further away from Paris. It means that there was not much space for new firms or inhabitants at the opening of the RER. What we see in this type of municipalities is a replacement effect. People who may have a higher willing to pay for the proximity with Paris pull out others when a station opens, namely foreign-owned firms, building firms, executive and professionals. This question would need to be linked with regulation of land use, but we do not have any data to account whether it explains these areas do not densify. Fur- ther away from Paris, there might be less competition for space as the urban development is less advanced. Under these conditions, the gross number of firms can rise. The results on population suggest that inhabitants may be the less willing to pay category for the proximity with Paris. The increasing size of Paris requiers all places within 30 km from Paris to be occupied, whatever their accessibility. That would be why, we find no significant differences between control and treatment group for population. Poorly accessible areas would be filled by poorer households. In all cases, we need more results on the composition of population to confirm this last hypothesis.
16 Table 5: Effect of RER on the share of executives and professionals
∆ Share of executives and professionals1975−90 in population in employment (1) (2) (3) (4) Intercept 0.336∗∗∗ 0.364∗∗∗ 0.07 0.009 (0.088) (0.092) (0.061) (0.118) ∗∗ ∗∗ RER1975−90 × 1(dP aris < 15km) 0.017 0.018 0.017 0.02 (0.007) (0.008) (0.014) (0.019) ∗ RER1975−90 × 1(15km < dP aris < 30km) 0.014 −0.005 0.008 0.008 (0.008) (0.009) (0.007) (0.008) RER1975−90 × 1(dP aris > 30km) −0.002 −0.005 −0.006 0.011 (0.01) (0.018) (0.006) (0.015) Train1990 0.006 0.011 −0.001 0.003 (0.004) (0.007) (0.005) (0.004) ∗ ∗ Highway1990 −0.003 −0.02 0.013 0.023 (0.011) (0.011) (0.009) (0.013) log(Surface) −0.002 0.002 0.004 0.006 (0.005) (0.007) (0.004) (0.007) 1(15km < dP aris < 30km) −0.004 0.001 −0.009 −0.017 (0.009) (0.011) (0.01) (0.014) 1(dP aris > 30km) −0.015 −0.031 −0.017 −0.037 (0.012) (0.021) (0.013) (0.025) ∗∗∗ ∗∗∗ ∗∗∗ ∗ log(Popu1975) −0.028 −0.04 −0.02 −0.015 (0.009) (0.011) (0.006) (0.008) ∗ ∗ ∗∗ log(Employment1975) 0.01 0.014 0.013 0.003 (0.006) (0.008) (0.006) (0.007) ∗∗∗ ∗∗∗ Share of exec. and prof. in pop.1975 0.024 0.029 (0.004) (0.007) ∗∗ Share of exec. and prof. in emp.1975 0.003 −0.021 (0.006) (0.01) Number of observations 312 99 286 96 Sample Strategy 1 Strategy 2 Strategy 1 Strategy 1 Weighted by prop. score XXXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
3.2 Robustness checks Table 7 presents a set of placebo tests. It estimates the treatment effect before and more than a decade after the commissioning of the RER. We find no significant effect of RER on employment neither for the 1968-1975 period nor after 1990. It suggests that the common trend assumption is valid. It also shows that the RER did not induce significant anticipation effects on firms. We would need more frequent data to ensure that there in not treatment impact in the last preceding years before a station opening.
Table 8 tests whether alternative treatment variables yield similar conclusions. Instead of the surface of the municipality located 500 meters from a RER station, columns (1) and (2) use a dummy for the presence of a RER station. The estimates are lager than in the baseline regressions. As stated before, it is very crucial to introduce the travel time in our estimations. The opening of a station is not sufficient to assess the quality of the network connection. Column (3) uses the variation of travel time with mass transit between 1975 and 1990. As many municipalities of the second control group are not provided with any RER or commuter train station, it is not possible to use the time variable with the second strategy. According to our estimates, when travel time to Paris decreases by one minute, employment increases by 3.3 percent in intermediate municipalities in terms of distance from Paris. There is no significant effect for other types of municipalities, which is in line with the results we obtain with the other treatment variable. Column (4) report the regression with our accessibility index. We did not find any significant effect for this type of
17 Table 6: Effect of RER on the number of foreign-owned firms
∆ log(Number of foreign-owned firms)1975−90 (1) (2) Intercept −6.152∗∗∗ −5.796∗∗∗ (1.405) (2.053) ∗∗ RER1975−90 × 1(dP aris < 15km) 0.291 0.016 (0.116) (0.179) ∗∗ RER1975−90 × 1(15km < dP aris < 30km) 0.163 0.599 (0.168) (0.259) ∗ RER1975−90 × 1(dP aris > 30km) −0.332 −0.318 (0.18) (0.44) Train1990 −0.108 −0.1 (0.09) (0.133) Highway1990 0.38 0.592 (0.231) (0.498) log(Surface) 0.345∗∗∗ 0.496∗∗ (0.118) (0.199) ∗∗ 1(15km < dP aris < 30km) −0.081 −0.873 (0.234) (0.407) 1(dP aris > 30km) −0.068 −0.588 (0.27) (0.607) ∗ log(Employment1975) 0.268 0.05 (0.141) (0.256) log(Popu1975) 0.154 −0.063 (0.251) (0.462) log(Firms1975) −0.208 0.035 (0.205) (0.279) ∗∗∗ ∗∗∗ log(Foreign-owned firms1975) −0.461 −0.463 (0.086) (0.165) Number of observations 167 67 Sample Strategy 1 Strategy 2 Weighted by prop. score XX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
variable so far.
18 Table 7: Effect of RER on employment - Placebo tests ∆ log ∆ log Employment1968−75 Employment1990−2009 (1) (2) (3) (4) Intercept −0.884∗ −1.863∗∗ −0.918∗∗ −1.206 (0.485) (0.922) (0.43) (0.763) RER1975−90 × 1(dP aris < 15km) −0.001 −0.1 −0.037 −0.037 (0.055) (0.095) (0.038) (0.046) RER1975−90 × 1(15km < dP aris < 30km) 0.047 0.041 0.057 0.09 (0.048) (0.076) (0.044) (0.073) RER1975−90 × 1(dP aris > 30km) 0.081 0.084 0.063 0.073 (0.091) (0.187) (0.052) (0.095) ∗∗∗ Train1990 −0.11 −0.077 −0.015 −0.019 (0.036) (0.055) (0.026) (0.025) ∗ Highway1990 0.118 0.1 0.008 −0.053 (0.067) (0.091) (0.054) (0.101) log(Surface) 0.123∗∗∗ 0.221∗∗∗ 0.076∗∗∗ 0.118∗∗∗ (0.037) (0.075) (0.026) (0.045) 1(15km < dP aris < 30km) 0.03 −0.196 −0.063 −0.107 (0.077) (0.141) (0.05) (0.076) 1(dP aris > 30km) −0.117 −0.406 −0.064 −0.154 (0.102) (0.247) (0.062) (0.137) ∗ log(Employment1968) 0.091 −0.016 (0.053) (0.078) ∗∗ log(Population1968) −0.15 −0.106 (0.064) (0.098) ∗∗∗ ∗∗∗ log(Employment1990) −0.125 −0.138 (0.039) (0.047) ∗ log(Population1990) 0.089 0.067 (0.048) (0.063) Number of observations 319 99 321 99 Sample Strategy 1 Strategy 2 Strategy 1 Strategy 2 Weighted by propensity score XXXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
19 Table 8: Effect of RER on employment - Alternative treatment variables
∆ log Employment1975−90 (1) (2) (3) (4) Intercept −1.499∗∗∗ −1.978∗∗ −1.361∗∗∗ −1.319∗∗∗ (0.45) (0.772) (0.471) (0.467) ∗ 1RER 1975−90 × 1(dP aris < 15km) 0.09 0.017 (0.047) (0.08) ∗∗∗ ∗ 1RER 1975−90 × 1(15km < dP aris < 30km) 0.151 0.207 (0.053) (0.117) 1RER 1975−90 × 1(dP aris > 30km) 0.001 0.081 (0.061) (0.092) ∆timeP aris 1975−90 × 1(dP aris < 15km) −0.006 (0.005) ∗∗ ∆timeP aris 1975−90 × 1(15km < dP aris < 30km) −0.033 (0.016) ∆timeP aris 1975−90 × 1(dP aris > 30km) 0.004 (0.019) ∆ log(Accessibility)1975−90 −0.117 (0.184) Train1990 −0.049 −0.011 −0.005 −0.006 (0.03) (0.038) (0.027) (0.027) ∗∗ ∗ ∗ ∗∗ Highway1990 0.197 0.273 0.112 0.125 (0.078) (0.162) (0.058) (0.06) log(Surface) 0.131∗∗∗ 0.172∗∗ 0.098∗∗∗ 0.094∗∗∗ (0.033) (0.07) (0.031) (0.031) ∗ 1(15km < dP aris < 30km) 0.031 −0.127 −0.02 0.084 (0.056) (0.111) (0.073) (0.046) ∗ ∗ 1(dP aris > 30km) 0.081 −0.073 0.147 0.127 (0.077) (0.122) (0.082) (0.071) ∗∗∗ ∗ ∗∗∗ ∗∗∗ log(Popu1975) 0.137 0.12 0.194 0.183 (0.045) (0.066) (0.043) (0.043) Number of observations 319 99 226 242 Sample Strategy 1 Strategy 2 Strategy 1 Strategy 1 Weighted by prop. score XXXX Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
20 4 Conclusion
References
Abhijit Banerjee, Esther Duflo, and Nancy Qian. On the road: Access to transportation infras- tructure and economic growth in china. NBER Working Papers 17897, National Bureau of Economic Research, Inc, 2012.
Marianne Bertrand, Esther Duflo, and Sendhil Mullainathan. How much should we trust differences- in-differences estimates? The Quarterly Journal of Economics, 119(1):249–275, 2004.
Maggie Xiaoyang Chen and Michael O. Moore. Location decision of heterogeneous multinational firms. Journal of International Economics, 80(2):188–199, 2010.
Leonard K. Cheng and Yum K. Kwan. What are the determinants of the location of foreign direct investment? the chinese experience. Journal of International Economics, 51(2):379–400, 2000.
Bernard Collardey. Les trains de banlieue, tome II de 1938 à 1999. Editions La Vie du Rail, 1999.
Pierre-Philippe Combes, Gilles Duranton, Laurent Gobillon, Diego Puga, and Sébastien Roux. The productivity advantages of large cities: Distinguishing agglomeration from firm selection. Econometrica Forthcoming, 2012.
Cletus C. Coughlin and Eran Segev. Location determinants of new foreign-owned manufacturing plants. Journal of Regional Science, 40(2):323–351, 2000.
Matthieu Crozet, Thierry Mayer, and Jean-Louis Mucchielli. How do firms agglomerate? a study of fdi in france. Regional Science and Urban Economics, 34(1):27–54, 2004.
James C. Davis and J. Vernon Henderson. The agglomeration of headquarters. Regional Science and Urban Economics, 38(5):445–460, 2008.
Fabrice Defever. Functional fragmentation and the location of multinational firms in the enlarged europe. Regional Science and Urban Economics, 36(5):658–677, 2006.
Dave Donaldson. Railroads of the raj: Estimating the impact of transportation infrastructure. Nber working papers, National Bureau of Economic Research, Inc, 2010.
Gilles Duranton and Diego Puga. Micro-foundations of urban agglomeration economies. In J. V. Henderson and J. F. Thisse, editors, Handbook of Regional and Urban Economics, volume 4, chapter 48, pages 2063–2117. 2004.
Gilles Duranton and Matthew A. Turner. Urban growth and transportation. Review of Economic Studies, forthcoming, 2012.
Stephen Gibbons, Teemu Lyytikäinen, Henry Overman, and Rosa Sanchis-Guarner. New road infrastructure: the effects on firms. SERC Discussion Papers 0117, Spatial Economics Research Centre, LSE, 2012.
Keith Head and Thierry Mayer. Market potential and the location of japanese investment in the european union. The Review of Economics and Statistics, 86(4):959–972, 2004.
Adelheid Holl. Manufacturing location and impacts of road transport infrastructure: empirical evidence from spain. Regional Science and Urban Economics, 34(3):341–363, 2004a.
21 Adelheid Holl. Transport infrastructure, agglomeration economies, and firm birth: Empirical evi- dence from portugal. Journal of Regional Science, 44(4):693–712, 2004b.
Guido M. Imbens and Jeffrey M. Wooldridge. Recent developments in the econometrics of program evaluation. Nber working papers, National Bureau of Economic Research, Inc, 2008.
Sung Jin Kang and Hong Shik Lee. The determinants of location choice of south korean fdi in china. Japan and the World Economy, 19(4):441–460, 2007.
Xuepeng Liu, Mary E Lovely, and Jan Ondrich. The location decisions of foreign investors in china: Untangling the effect of wages using a control function approach. The Review of Economics and Statistics, 92(1):160–166, 2010.
Philippe Martin and Carol Ann Rogers. Industrial location and public infrastructure. Journal of International Economics, 39(3-4):335–351, 1995.
Thierry Mayer, Isabelle Mejean, and Benjamin Nefussi. The location of domestic and foreign production affiliates by french multinational firms. Journal of Urban Economics, 68(2):115– 128, 2010.
Guy Michaels. The effect of trade on the demand for skill: Evidence from the interstate highway system. The Review of Economics and Statistics, 90(4):683–701, 2008.
Loriane Py and Fabrice Hatem. Internationalisation et localisation des services : une analyse secto- rielle et fonctionnelle appliquée aux firmes multinationales en europe. Économie et Statistique, 426(1):67–95, 2009.
Kurt Schmidheiny and Marius Brülhart. On the equivalence of location choice models: Conditional logit, nested logit and poisson. Journal of Urban Economics, 69(2):214–222, 2011.
Vanessa Strauss-Kahn and Xavier Vives. Why and where do headquarters move? Regional Science and Urban Economics, 39(2):168–186, 2009.
Stephen Ross Yeaple. Firm heterogeneity and the structure of u.s. multinational activity. Journal of International Economics, 78(2):206–215, 2009.
22 A Appendix
Table 9: Estimation of the propensity score Intercept −8.954∗∗∗ (3.116) ∗∗∗ log(Employment1968) −0.051 (0.275) ∗∗∗ log(Population1968) 0.938 (0.312) ∗∗∗ 1(dP aris < 15km) −0.646 (0.385) ∗∗∗ 1(dP aris > 30km) −1.296 (0.354) North-West −1.316∗∗∗ (0.312) North-East −1.255∗∗∗ (0.361) South-East −1.875∗∗∗ (0.347) log(Surface) 0.093∗∗∗ (0.197) ∗∗∗ Highway1990 0.27 (0.41) Number of observations . Standard errors in brackets - Significance levels: ∗∗∗ 1%, ∗∗ 5%, ∗ 10%
23