Policy Research Working Paper 9990 Spatial Misallocation, Informality, and Transit Improvements Evidence from Mexico City Román D. Zárate Development Economics Development Research Group March 2022 Policy Research Working Paper 9990 Abstract This paper proposes a new mechanism to explain resource show that transit improvements reduce informality by 7 misallocation in developing countries: the high commut- percent in areas near the new stations. The paper develops ing costs within cities that prevent workers from accessing a spatial model that accounts for the direct effects of infra- formal employment. To test this mechanism, the paper structure in perfectly economies and allocative efficiency. combines a rich collection of microdata and exploits the Changes in allocative efficiency driven by workers’ realloca- opening of new subway lines in Mexico City. The findings tion to the formal sector amplify the gains by 20–25 percent. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at rzaratevasquez@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Spatial Misallocation, Informality, and Transit Improvements: Evidence from Mexico City∗ Román David Zárate† World Bank Keywords : Informality, allocative efficiency, urban transit infrastructure. ∗ I am extremely grateful to my advisors Ben Faber, Cecile Gaubert, and Andrés Rodríguez-Clare for their con- tinuous support and guidance in this project. I also want to thank my discussants Clare Balboni, Gabriel Ulyssea, and Alejandro Molnar, and seminar participants at the Dallas Fed, McGill University, the World Bank, ITAM, Uni- versidad de los Andes, Universidad del Rosario, PUC Rio, Vancouver School of Economics, Nottingham University, the Macroeconomics Conference at Oxford University, the Online Urban Economics Seminar, the NBER SI in Urban Economics, the Cities and Development Workshop, Hitotsubashi University, and the 2021 ASSA meetings for very useful suggestions. I also want to thank David Atkin, Kirill Borusyak, David Card, Andrés González-Lira, Marco González-Navarro, Patrick Kennedy, Chris Severen, Joaquín Klot, Isabela Manelici, Pablo Muñoz, Mathieu Pede- monte, Darío Tortarolo, Nick Tsivanidis, Jose P. Vásquez, Román Andrés Zárate, and Isabel Hincapie for very helpful comments and discussions. I thank the National Institute of Statistics and Geography (INEGI) and especially Dr. Natalia Volkow for granting me access to the data. Financial support from the Clausen Center at UC Berkeley is gratefully acknowledged. The views expressed are only those of the author and they do not reflect the views of INEGI. † World Bank, DEC TI. E-mail: rzaratevasquez@worldbank.org. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1 Introduction Poor transportation infrastructure is a common characteristic of cities in developing countries. For instance, in Mexico City, it takes a typical low-skilled worker approximately two to three hours to commute to work in the center of the city. In recent decades, governments around the world have spent billions of dollars on infrastructure projects to facilitate commuting. Recent research exam- ines the aggregate gains from public transit improvements, assuming perfectly efficient economies. However, perfectly competitive models may fail to capture key features of developing economies, where labor market frictions and other economic distortions are salient.1 In this paper, I study the economic impacts of transit infrastructure in Mexico City, while considering both its direct effects in perfectly efficient economies and the role played by distortions on allocative efficiency. Labor market informality is one of the most significant sources of distortions in low- and middle- income countries, with important implications for aggregate efficiency. Within developing countries, 50 to 60 percent of total employment is informal. Informal firms are less productive than formal ones, avoid paying taxes, and do not make social security contributions to their workers.2 As a result, the informal sector creates input wedges that cause factor misallocation, which ultimately lowers aggregate total factor productivity (TFP) (Banerjee and Duflo, 2005; Hsieh and Klenow, 2009; Restuccia and Rogerson, 2008). These intersectoral distortions between the formal and informal sectors imply that any policy or shock that impacts informality may have first-order effects on aggregate welfare through an allocative efficiency margin.3 This study explores the link between transit improvements, informality, and aggregate efficiency at the city level. I test whether infrastructure projects that facilitate transit within a city improve allocative efficiency by reallocating workers from the informal to the formal economy. As a result, aggregate gains from these projects can be larger relative to those demonstrated by standard urban models that assume perfectly efficient economies. The core intuition is that in cities in developing countries, workers in remote locations prefer to work in low-paid informal jobs near their home rather than incurring the high cost of commuting to formal employment. Transit improvements may thus provide better access to formal jobs, leading to an expansion of the formal sector and a more efficient labor allocation. The paper makes two main contributions. First, I combine rich administrative microdata with a transit shock to provide new empirical evidence on the effect of urban transit improvements on worker reallocation across the formal and informal sectors. Second, I rationalize these results through the lens of a quantitative spatial model. To this end, I extend recent work (Ahlfeldt et al., 2015; 1 See e.g., Atkin and Khandelwal (2019) for a recent review of market distortions in the context of the gains from market integration, and Busso et al. (2012); Levy (2018) for the effect of distortions on total factor productivity in the Mexican context. 2 See Gollin (2002, 2008) and La Porta and Shleifer (2008, 2014) for the relationship between the prevalence of the informal sector and economic development. 3 See, e.g., McCaig and Pavcnik (2018) and Dix Carneiro et al. (2018) for the case of trade policies and their effect on the informal economy and the aggregate gains from trade. The former paper studies the effect of the Free Trade Agreement between the US and Vietnam, and the latter the impact of the Brazilian trade liberalization episode. 1 Allen et al., 2015; Tsivanidis, 2019) by adding intersectoral distortions and factor misallocation to an urban framework. Following Baqaee and Farhi (2020), I provide a formula that decomposes the welfare gains from transit developments into a “direct” effect and an allocative efficiency term. This latter term captures two different components: factor misallocation and agglomeration forces when they differ between the formal and informal sectors.4 Mexico City constitutes a relevant and informative case study for several reasons. First, it has a dense concentration of economic activity, accounting for around 8.9 million people. Second, the Mexican case is typical of developing countries, especially in Latin America, where more than 50% of the urban labor force and 70% of business establishments are informal.5 Furthermore, the city constructed a new primary subway line in the early 2000s, connecting remote areas in the north with the center of the city. The entire line was not initially planned by the Government and there were multiple delays in its construction, suggesting that the opening dates were uncorrelated with local demand and supply shocks. Moreover, Mexico City collects unique data that I use to estimate the impact of transit improvements on informality. Throughout, I use the standard definition of informality: a worker is informal if he/she does not receive social security benefits based on the contractual relationship with his/her employer. At the center of the analysis is a rich collection of administrative microdata. I observe the geog- raphy of jobs and worker residences for both the formal and informal sectors at the high granular census tract level. In the analysis, I use four main sources of data. First, I use confidential microdata from several reports of the Mexican Economic Census, covering the universe of business establish- ments located in Mexico City. Second, I use the microdata of the Mexican Population Census to determine the residence of both formal and informal workers. Third, I use detailed information on the transportation network in Mexico City, which I complement with transportation diaries (origin- destination survey data). Additionally, I use the 2015 Intercensal Survey to construct commuting and trade flows at the city level for both sectors. I also use standard household survey data to calibrate some of the parameters of the model. In the first part of the paper, I document three empirical findings that suggest a negative rela- tionship between the accessibility of jobs and informality. First, I exploit cross-sectional variation among informal vs. formal workers to show that informal workers spend less time commuting, and work closer to home relative to their formal counterparts. For instance, informal workers spend 40% less time commuting on average. This implies that informal workers are more sensitive to commuting costs than formal workers. These findings are robust to controlling for different sets of fixed effects and individual characteristics. Second, I compare the residence of informal and formal workers. I document that formal jobs concentrate in the west and center of the city, where most economic activity takes place. By contrast, 4 See Bartelme et al. (2019) for recent work that studies the effect of optimal policies when agglomeration exter- nalities differ across sectors or locations. 5 See Perry et al. (2007) and Ulyssea (2018) who document informality rates in Latin America. In this region, the informal economy varies from 35% in Chile to 80% in Peru. 2 most informal workers reside in the east and the periphery of the city. Third, I exploit the construction of a new subway line that connected remote locations with the center of Mexico City to provide causal evidence that transit infrastructure leads to a decrease in informality rates. Specifically, I estimate a series of difference-in-differences specifications that use variation in access to new transit. These specifications control for initial characteristics of census tracts and capture changes in informality after the transit shock in locations close to the new subway line. The key identification assumption is that the opening dates of these new commuting links were unrelated to other local demand- or supply-side shocks that affected locations near the new line. This assumption is supported by the decades-long planning horizon in which part of the line was included, and several unexpected and multi-year delays in the opening schedule. I further corroborate this assumption by documenting no apparent pre-trends among the most affected locations in the preceding periods. The main finding is that informality rates decrease in locations close to the new subway stations. I find that the ratio of formal to informal residents increases by approximately 6%-7% in locations close to the new stations after the shock. Similarly, Workers’ informality rates decrease by 2 to 4 percentage points after the construction of the new line, and firms’ informality rates decrease by 1 to 3 percentage points. These estimates represent a 6.2% decrease in workers’ informality rates, using the average informality rate in the baseline year as a benchmark. I also construct bounds of the effect to control for the sorting of workers using retrospective questions. In particular, even under the extreme assumption that all workers who moved from the central areas to the treated locations were formal in the pre-period, the effect is still 4.5%. To check the robustness of the difference-in-differences specification, I compare the new line with similar planned metro lines that were not completed over this period for unrelated reasons, using an expansion plan from 1980. Reassuringly, this robustness check yields similar estimates to the baseline specification. Another potential concern with identification is a change in the composition of households in areas close to the new stations. To examine this, I also estimate the difference-in- differences specification using household characteristics as dependent variables. I find that the transit shock did not lead to changes in the composition of households based on observable characteristics.6 Also, to control for sorting, I also estimate bounds of the effect, assuming that all workers who moved to the locations that experience the shock were formal. To calculate and decompose the welfare gains from transit improvements, I build a quantitative model with multiple sectors and wedges that captures the three empirical findings. The model allows me to quantify the aggregate effects of new infrastructure, while considering its additional impact on factor allocation. Following Baqaee and Farhi (2020), I provide a formula that decomposes the welfare effects of any trade/commuting costs shock into two different components: a “direct” effect term and an allocation term.7 Intuitively, the sign of this additional impact will depend on whether 6 Moreover, the quantitative model accounts for employment and location decisions within the city. Thus, I consider the change in household characteristics after the transit shock based on unobservables. 7 This approach provides the main intuition for the allocation channel. Since it is a first-order approximation. To 3 the shock reallocates workers to sector-locations with larger wedges or stronger agglomeration forces.8 The logic is similar to that in Hsieh and Klenow (2009) who show that sectors with larger wedges are too small relative to the first-best allocation. If a shock reallocates workers to firms bearing higher distortions, the aggregate welfare effects are larger. I calibrate the model using structural relationships. The key parameter to estimate is the labor supply elasticity across sectors, which governs the reallocation of workers from the informal to the formal sector. I follow Tsivanidis (2019) and calculate measures of market access for residents and firms by sector. To recover this key elasticity, I exploit variation across locations after the shock by running a triple-difference estimator that associates changes in labor allocation between the formal and informal sectors with changes in market access. I find that the estimates for the labor supply elasticity parameter, around 1.9, are consistent with the theoretical assumptions of the model and the data. Intuitively, if a transit shock connects workers to better formal jobs relative to informal jobs, workers reallocate from the informal to the formal sector, generating additional welfare gains. Moreover, to calibrate the wedges, I use two different approaches that yield similar results. First, I follow Hsieh and Klenow (2009) and use the inverse of the labor and capital share relations. Under the assumption that all firms within the same sector use the same production function, differences in these shares capture the wedges. Second, I use the analysis from Levy (2018) that documents differences in taxes, subsidies, and other distortions between formal and informal establishments in the Mexican economy. I assume a constant wedge for formal firms considering all these distortions and a wedge of zero for the informal ones. Next, I quantify and decompose the welfare gains from the transit shock by varying trade/- commuting costs in the GE model. I find that the allocative efficiency margin drives a significant fraction of the total gains. I am able to run counterfactuals from an initial equilibrium inverting the model and recovering scale parameters. I compute the counterfactuals, using the estimates of the key elasticities and the initial equilibrium conditions (Dekle et al., 2008). The results suggest that the new subway line increased welfare by around 1.8%. I find that the direct effects explain approximately 79% of the total gains, while the reallocation of workers from informal to formal firms explains 18% and the remaining 2% are driven by differences in agglomeration between the two sectors.9 The counterfactual analysis also suggests that the reductions in commuting and trade costs account for a similar amount of the total gains. In terms of the cost-benefit analysis, the allocative efficiency margin increases net welfare by a considerable margin. According to official documents from the Government of Mexico City (hence- forth the Government), the net present value of the total cost of a subway line with 20 km and 20 capture actual changes in welfare, the approach rests on the assumption that any reduction in commuting costs is infinitesimal. Accordingly, I compute the counterfactuals using percentage changes. 8 The third term arises in the presence of heterogeneous agglomeration externalities or trade imbalances as in Fajgelbaum and Gaubert (2020). In the case of the efficient economy, I am assuming trade balances. In the inefficient economy, the labor wedges create trade imbalances, and the third term arises. 9 I compute two counterfactuals. The first one allows workers to migrate within the city. The second one holds constant the population in each census tract. The results of both counterfactuals are very similar. 4 stations is approximately 0.72% of the total GDP of Mexico City. Since line B increases welfare between 1.7% and 1.9%, this represents a net gain of around $2.5 USD per every dollar spent on the infrastructure. This gain would be lower if we didn’t consider the allocative efficiency margin in a perfectly efficient economy. For example, in the case of migration, the reallocation of workers from the informal to the formal sector increases the average real income net of the total cost by 26%. I run other counterfactuals in which I simulate different policies that the Government can imple- ment to reduce informality rates. The results suggest that transit infrastructure can be an effective policy tool to reduce informality by connecting informal workers with formal jobs. For example, to reduce informality rates by 0.5% at the aggregate level, the Government needs to reduce the fixed cost of entry to the formal economy by more than 7% or increase the fixed cost of entry to the informal economy by more than 10%. Similarly, I show that connecting informal workers with formal jobs is more efficient than implementing de-agglomeration policies that reallocate firms to the outskirts. I also show that in the case of constructing central transit lines, the allocative efficiency margin would explain a lower fraction that line B. Overall, the findings suggest that it is important to consider the role of the allocative efficiency margin in the optimal allocation of infrastructure. Recent papers such as Fajgelbaum and Schaal (2017), Balboni (2019), and Santamaría (2020) have estimated the infrastructure misallocation in spatial general equilibrium models. My results suggest that when a social planner decides where to allocate infrastructure, there are also first-order effects driven by the resource misallocation compo- nent that are economically important. Related Literature This paper contributes to different strands of the literature. The first is the economic geography and urban economics literature, which has assessed the economic impacts of urban infrastructure. The second is the macro-development literature, which has studied the main drivers of the informal economy, including the effect of allocative efficiency on TFP. This latter strand is related to a large literature on international economics that has estimated the impact of trade reforms on allocative efficiency in the presence of domestic distortions. First, a new strand of literature has explored the impact of transit infrastructure within cities (Ahlfeldt et al., 2015; Baum-Snow, 2007; Gonzalez-Navarro and Turner, 2018; Heblich et al., 2018; Monte et al., 2018; Tsivanidis, 2019). For example, Tsivanidis (2019) assesses the welfare and distributional effects of a new bus rapid transit system in Bogotá, and Heblich et al. (2018) study the economic consequences of the subway in London. My paper adds to this literature by examining the effect of transit infrastructure on allocative efficiency. I depart from standard urban economic models by adding distortions and resource misallocation.10 This paper also contributes to a literature on the role of factor misallocation in lowering ag- 10 Another type of distortion in the context of an urban model is studied by Pérez Pérez (2018) who assesses the impact of minimum wage on aggregate employment and commuting patterns across US cities. 5 gregate TFP (Banerjee and Duflo, 2005; Hsieh and Klenow, 2009; Restuccia and Rogerson, 2008). These studies have shown that the dispersion in distortions across firms and sectors generates factor misallocation, and more so in developing than in advanced economies. In the particular case of Mexico, Busso et al. (2012) show that if workers reallocate from the informal to the formal sector by eliminating wedges, TFP increases by approximately 50%. Other studies have aimed to understand the main causes of the large levels of resource misallocation in developing countries. Some of the primary explanations consist of regulations, markups, and the wedges caused by the informal sector. Similarly, other papers such as Fajgelbaum et al. (2019) and Hsieh and Moretti (2019) have shown that state taxes and housing restrictions generate spatial misallocation in the US. Third, my work also contributes to a strand of the international economics literature that studies gains from trade through the allocative efficiency channel. This literature was recently reviewed by Atkin and Khandelwal (2019), who discuss the role of distortions on the aggregate gains from market integration. Most of these articles have explored the response of markups to trade liberalization episodes or changes in infrastructure (Arkolakis et al., 2019; Asturias et al., 2016; Edmond et al., 2015; Holmes et al., 2014; Hornbeck and Rotemberg, 2019). Similar to my paper, some studies have ecki, 2017), and others the effect of analyzed the effect of intersectoral distortion on welfare (Świ¸ trade on informality (Dix Carneiro et al., 2018; McCaig and Pavcnik, 2018; McMillan and McCaig, 2019). While this literature focuses on trade reforms that affect labor demand, my paper examines the impact of commuting and urban trade on aggregate productivity. Other studies, such as Moreno-Monroy and Posada (2018) and Suárez et al. (2016), have also explored the relationship between commuting and informality. They argue that the high commuting cost to a formal job faced by a large part of the population increases informality rates in developing countries. My paper investigates this relationship by providing empirical evidence on the relationship between informality and transit infrastructure. Moreover, through a quantitative model, it measures the economic impact of infrastructure on allocative efficiency by analyzing factor reallocation to the formal sector. The rest of the paper is organized as follows. Section 2 introduces the setting of my study in Mexico City and describes the transit shock. Section 3 presents the reduced-form evidence of the effect of commuting on informality. Section 4 develops an urban quantitative model with multi- ple sectors and intersectoral distortions. Section 5 estimates the main parameters of the model. Section 6 quantifies and decomposes the welfare gains from transit improvements and run other counterfactuals. Section 7 concludes. 6 2 Institutional Context 2.1 Transit System In the second half of the twentieth century, Mexico City had severe public transport problems, with congested main roads and highways, particularly in the downtown area. In 1967, the Government established a decentralized public office to build and operate a rapid transit system of underground trains to facilitate public transportation in Mexico City. Two years later, on September 4, 1969, the Government inaugurated the first line. Today, the system has grown into 12 lines with 195 stations, for a total length of 128.4 miles. The subway is the largest in Latin America and the second-largest system in North America after the New York City Subway. The Plan Maestro 1985-2010 guided the expansion of the subway. It set the mobility goals that the transport system needed to satisfy over the long run, based on best practices in urban development and the operational constraints of the project. The Plan Maestro 1985-2010 underwent some modifications from what the Government had initially planned. For example, Line B was originally Line 10 and experienced extensive changes (Ramírez et al., 2017). These modifications responded mainly to changing demand patterns for transportation in Mexico City, which forced the Government to redesign some lines. Part of my empirical strategy is to compare the unplanned modifications to the subway lines with the original and un-executed plans. In my empirical strategy, I exploit the construction of line B. This line had the distinct feature of connecting informal workers in remote areas with jobs in the central business district (CBD) of Mexico City. It was inaugurated in 2000, and most of it was initially planned as part of Plan Maestro 1985, which reduces potential endogeneity concerns between the opening of the new stations and local demand/supply shocks. Moreover, the construction of the line also experienced multiple delays given changes in the regulatory framework and the 1994 financial crisis.11 The line is approximately 20 kms long and has 21 stations. It connects the metropolitan area of the city with some adjacent municipalities in Mexico State, such as Ecatepec de Morelos and Ciudad Nezahualcoyot. These areas are characterized by high poverty rates, low education, and high informality rates.12 As a result, line B has the distinct feature of connecting informal workers with formal jobs. To date, it is the line with the fourth-highest number of passengers in the network. The total cost of this line, including the net present value of service operations, maintenance, and other overheads, was approximately $2,900 million in 2014 USD dollars, which represents 0.7% of the total GDP of Mexico City. Figure 1 depicts a map of the Mexico City subway system in 2000, highlighting the lines that I use in my empirical strategy. Line B (purple) connects the northeastern area, including locations in the State of Mexico, with the center of the city. I also use line C and line 12 for robustness checks. Line C (green) was planned as a feeder line in the early 2000s, similar to line B; however, 11 The initial plan of the Government was to finish the line in 1997. However, they finished the construction of the entire line in 2002. 12 In the Appendix, I relate census tract characteristics before the shock to line B to show this result. 7 the Government never constructed it. Line 12 (red) is the newest subway line in Mexico City and was opened in 2012. 2.2 Informality Following Busso et al. (2012); Kanbur (2009), and Levy (2018), I use two definitions of informality. The first is the standard definition and is based on whether firms comply with labor regulations. A worker is defined as informal if the firm does not pay social security taxes.13 These workers can be salaried or non-salaried workers. The second definition of informality covers self-employed workers and family members that work in a household business. The latter definition is a more restrictive one, as it includes only the non-salaried workers of the first group.14 As in most developing countries, informality in Mexico is a significant problem. It affects 57% of the total workforce and 78% of firms (INEGI). Figure A1 in the Online Appendix compares informality rates (using the standard definition) in countries in Latin America and the Caribbean to the average of the OECD. Informality rates in the entire region are very high. The average across the region is 50%, which is much higher than the OECD average of 17%. Relative to other countries in the region, Mexico has one of the highest informality rates, and the difference is even more significant when we compare Mexico to other countries with a similar income level, such as Argentina or Colombia.15 The presence of the informal sector and the fact that informal firms avoid paying taxes create a labor wedge across establishments. According to recent estimates, a firm that fully complies with salary regulations is expected to pay social security taxes amounting to 18%-33% of a worker’s wage (Busso et al., 2012; Levy, 2018) and 20% on sale taxes.These wedges create distortions across firms that decrease welfare and TFP. Figure A2 in the Online Appendix plots the size and productivity distribution of different definitions of formal and informal firms in the Mexican context. Informal firms are smaller and less productive than formal firms. However, due to the presence of labor wedges (social security taxes), informal firms are larger and formal firms smaller, relative to a social optimum.16 As a result, reallocating workers from the informal to the formal sector may lead to productivity gains that impact welfare. Different studies have examined the gains from removing the informal sector in Mexico, finding that TFP would increase by approximately 200% in a world without these distortions (Busso et al., 2012). In the next section, I show how informality rates are unequally distributed across the city. On the one hand, most formal firms are usually located in the center. On the other, informal workers usually live in the periphery and have poor access to formal employment. 13 Social security benefits include health care, savings for retirement, social benefits for recreation, and invalidity allowances. 14 The second group is a subset of the first group. 15 I do not observe the second definition of informality in other countries. 16 Busso et al. (2012) and Levy (2018) study the formal vs. informal sector in the Mexican context, and show that wedges are larger for formal firms. 8 3 Data and Motivating Findings 3.1 Data My primary unit of observation is the urban census tract (Area Geoestadística Básica in the Mexican micro-data). I use a sample of approximately 3,500 census tracts from 116 different neighborhoods and 24 different municipalities, 16 municipalities of which are in Mexico City and 8 of which are adjacent municipalities from the State of Mexico. The first source of information is standard GIS data on the location of the transportation network and the new transit subway lines. I also use data on roads and highways in Mexico City to calculate commuting times for different transportation modes using the network analysis toolkit from ArcMap. By merging these datasets, I can estimate commuting/trade costs before and after the transit shock using a weighted average of travel times across the different transportation modes. The second source of data is the Mexican Economic Censuses collected by INEGI. This is an establishment level data set that provides standard information such as sales, value added, number of workers, salaried workers, social security, and other outcomes. This census is carried out every five years starting in 1994. I am able to define the informal sector at the firm level using social security payments as discussed in Section 2. I categorize firms and workers in four different groups based on labor market regulations. I also calibrate wedges for each location and sector using wage bill, sales, and social security payments.17 The third source of information is the Mexican Population Census. This census is carried out every ten years, and INEGI provided me with the data since 2000. With this information, I am able to calculate the number of informal, formal, and total residents in each location. In 2000, the Population Census also reported other variables such as household income and job characteristics the week before the census interview. Moreover, I use the 2015 Intercensal Survey that provides information on the workplace, residence, and transportation mode of formal and informal workers at the municipality/locality level. This data allows me to observe commuting flows in Mexico City for each sector. I also use the 2017 Origin-Destination Survey collected in the commuting zone area of Mexico City. I use this data for two purposes. First, I infer trade flows across the city using trips to restaurants and other types of shops at different hours of the day. Second, I discuss some motivational findings on commuting patterns. Finally, I complement my results with standard household survey data from the Encuesta Na- cional de Ocupación y Empleo (ENOE). I calibrate some of the parameters of the model using this data. 17 For the Economic Census, I observe data for periods before and after the transit shock, which allows me to test for parallel trends in my main specification. 9 3.2 Empirical Facts In this section, I discuss three empirical findings that show a negative relationship between informal- ity rates and the accessibility of formal jobs in Mexico City: 1) informal workers are more sensitive to commuting costs and spend less time commuting; 2) informal workers are located in areas in which they have poor access to formal employment; and 3) informality rates decrease with transit improvements that connect informal workers to formal employment. 3.2.1 Cross-sectional Variation Finding 1: Informal workers spend less time commuting and work closer to their home relative to formal workers. To reach the first finding, I use the 2015 Intercensal Survey. With this data, I observe the residence and workplace and average commuting time of each worker at the municipality level. Exploiting cross-sectional variation, I compare the average commuting time and the workplace decisions of informal vs. formal workers. I adopt the standard definition of informality based on the contractual relationship of the worker, I also restrict the sample to individuals who worked the week before the census interview. I run the following linear probability model to test whether informal workers spend less time commuting: yi = β0 + β1 Informali + γXi + γl(i) + γn(i) + γm(i) + i , (3.1) where yi is a dummy variable that takes the value of 1 if individual i commutes to a different municipality than the one in which he/she resides, whether he/she works in the central business district (CBD) of Mexico City, or whether their average commuting time is within some window of time (i.e., 16 to 30 minutes); Xi is a vector of individual characteristics that includes age, education, gender, relationship to head of household, and a dummy variable indicating whether the individual has an African or indigenous background; γl(i) and γn(i) are origin and destination fixed effects; γm(i) is a transportation mode fixed effect to compare informal vs. formal workers that use the same transportation mode; and, i is the error term of the regression. Figure 2 depicts the point estimate and confidence interval of a linear probability model. I relate the probability that the average commuting time of a worker is within some window of time with a dummy variable that takes a value of 1 if the worker is informal. As the figure shows, informal workers spend less time commuting than formal workers. For instance, the first bar shows that informal workers are more likely to work from their home relative to formal workers by 13 percentage points. Similarly, informal workers are more likely than formal workers to spend 15 minutes commuting. On the other hand, formal workers are more likely to spend 30, 60, or 120 minutes commuting. To provide more evidence of this result, table B1 in the Online Appendix reports the results for 10 the dummy variables of whether the worker commutes to another municipality; and whether he/she works in Mexico City. The results imply that informal workers spend less time commuting. For instance, the probability of commuting to a different municipality, decreases on average, between 8.0 and 25.0 percentage points for informal workers. Similarly, informal workers are less likely to work in the CBD between 4.0 and 9.0 percentage points. In the fourth column, I show that differences in transportation modes are not driving these effects. Overall, the results from Table B1 and Figure 2 suggest that informal workers spend less time commuting. One potential interpretation of this result is that informal jobs are easier to substitute across locations than formal ones.18 Finding 2: Most formal jobs are located in the central areas of the city, while most informal workers reside in the outskirts. The second finding is that most formal jobs are available in the center and west of the city, while informal workers reside in other, less connected areas. As a consequence, workers that cannot afford the high rents in the center of the city, live in outlying areas with poor access to formal jobs. Figure 3 presents a heat map of informality rates in Mexico City and adjacent municipalities in the State of Mexico in terms of jobs and the residence of workers. Panel A in figure A10 in the Appendix shows that the west and the center of Mexico City have the highest level of economic activity.19 Combining these two figures, we see that informality rates are lower in the west and the center of the city than in the east and the periphery of the city. This suggests that workers who live in remote locations usually have poorer access formal employment. 3.2.2 Difference-in-Differences Specification Finding 3: Informality rates decline with transit improvements that improve market access of formal employment to informal workers. I now exploit the construction of line B of the subway in Mexico City by estimating a series of difference-in-differences specifications. I compare locations close to the new subway line with loca- tions in the rest of Mexico City and test whether those that improved their market access experienced a change in informality rates after the transit shock while controlling for initial characteristics. One feature of line B is that it connects remote locations in the State of Mexico, close to Ecatepec de Morelos, with the city’s center. The identification assumption is that the opening of the new stations is uncorrelated with local demand/supply shocks. The fact that most of the line was planned decades earlier makes this assumption plausible. Moreover, since the construction of infrastructure may be endogenous (Redding and Turner, 2015), I include a set of covariates as controls to compare similar areas. Another potential concern to the identification is sorting given by a change in the residents that prefer to work in the formal sector. I show in the next section that household characteristics 18 This interpretation would be later corroborated by estimating gravity equations. 19 Panel B in figure A10 shows that the labor wedge in these locations is higher. 11 are not correlated with the opening of line B. Furthermore, I also estimate lower bounds of the effect using retrospective questions, and in the quantitative framework, I consider this channel by allowing migration within the city. I use both jobs’ and workers’ informality rates as dependent variables. I test first for changes in informality in terms of the locations in which the workers live. I use data from the Population Censuses and estimate the following specification relating the transit shock to the change in the ratio between formal and informal workers: ∆ (ln LiF − ln LiI ) = β Ti + γXi + δs(i) + i , (3.2) where Lis is the number of individuals that live in census-tract i and sector s, Ti is one of four dif- ferent treatment variables: log distance in meters, log distance in walking minutes using the network of roads, a dummy variable indicating whether the closest station is within the 10th percentile of the Euclidean distance, and a dummy variable whether the closest station is within 25 minutes, δs(i) are state or municipality fixed effects,20 and Xi is a vector of census-tract characteristics that include distance controls such as: the area in square kilometers, distance to other stations of public transit, a central business district dummy variable, and some productivity measure in the baseline year in which I include value added per worker and the number of firms to capture how good is the location in terms of jobs. This equation relates the transit shock to the log of the ratio between formal and informal workers.21 I estimate equation 3.2 for the pool of workers and for different groups based on skills.22 Table 1 reports the results for different specifications of equation 3.2, while Figure A3 in the Online Appendix depicts the three-point estimates of my preferred specification for the pool of all workers, low-skilled workers, and high-skilled workers, respectively. Overall, the results imply that locations close to the new subway line experienced a decrease in workers’ informality rates. In particular, the ratio of formal to informal individuals increased between 3.0% and 6.9% after the shock. These results are robust to different specifications, for example, to the use of different definitions of the treatment variable or to the use of different sets of fixed effects or controls. In addition, in panels C and D, I control for the change in workers’ composition in terms of skills and report the results only for low-skilled workers. The estimates are very similar to the ones found for the entire pool of workers. For instance, the ratio between formal and informal low-skilled workers increased on average between 4.0% and 7.1%. Moreover, in panels E and F, I report the results restricting the sample to the areas not located in the CBD of Mexico City. The effects should be larger in these locations since more informal workers live in these areas. Overall, I find larger effects for this specification; the ratio between formal and informal workers increased by almost 10% in these areas. 20 For the municipality fixed effects specifications, I classify locations in the State of Mexico into four different groups: northwestern, northeastern, west central, and east central for a total of 20 municipalities. 21 Equation 3.2 corresponds to a structural relationship that I will derive from the model in section 5. 22 One caveat of this specification is that I cannot test for parallel trends due to data constraints because I cannot observe the location of informal/formal residents before the 2000 Census. 12 Moreover, in table 2 I estimate line B’s effect on the overall log number of individuals and disentangle the effect from the previous regression between formal and informal workers. In panel A, I report the results for the pool of workers, while panels B and C report the results for the number of formal and informal workers. The dependent variable in the first and third columns is the log number of workers. On the one hand, the point estimates suggest that the effect is very small on the number of individuals. For instance, it is only 1.7% in the case of the pool of workers, and 2.2% for low-skilled. On the other hand, the second and fourth columns show the estimates for the log number of formal workers. The results suggest that the locations affected by the shock experienced an increase in formal workers between 3% and 6%. This effect is larger than the estimate on the number of individuals. Finally, the third and sixth column reports the results for the log number of informal workers. In the case of municipality fixed effects, the number of informal workers decreased by around 3% in the locations that experienced the shock. There are two main potential concerns of the interpretation of this effect. First, line B may be endogenous; I address this concern in section 3.2.3 by comparing line B with planned lines. Second, worker sorting maybe explaining the results. I address this concern in section 3.2.4. Second, I use data from the Economic Censuses and test whether the shock also generated an indirect effect affecting the “treated” location in terms of jobs. I estimate the following specification to study whether areas close to the new subway lines experienced changes in workers’ informality rates: yi,t = βτ Ti + δi + δs(i),t + γt Xi + i,t , (3.3) τ =1994 where yi,t is one of the outcomes of interest of census tract i at moment t. I estimate equation 3.3 for the following outcomes: the share of informal workers and the share of informal firms; Ti is one of the four different treatment variables; δi are census tract fixed effects, and δs(i),t are state-time or municipality-time specific trends, γt · Xi are census-tract characteristics-time-specific trends that include the distance controls, i,t , is the error term of the regression. The coefficients of interest are the parameters βτ , and the baseline year is 1994. Since the line was built in 2000, the placebo for parallel trends corresponds to 1999. I compute the standard errors with clusters at the census-tract level. Figure 4 and table B4 in the Online Appendix report the point estimates for the main outcome, the share of informal workers. I find that workers’ informality rates decrease in locations near line B after the transit shock. I also find evidence of parallel trends since the point estimate is small and not significant in 1999. On average, informality rates decrease between 2.0 and 4.0 percentage points in locations that experienced the shock. The results are similar using the standard definition of informality or a stricter definition of informality that considers only informal and non-salaried workers that do not have an actual contract with the establishment. Moreover, these effects are robust to the use of the Euclidean distance, the walking distance using the network of roads, or 13 dummy variables indicating whether locations are close to the new stations within some range (i.e., 2100 meters or 25 minutes). Furthermore, in columns five to eight, I include municipality-time fixed effects and the results hold, suggesting that even after I compare locations to those within the same municipality, census tracts closer to new stations experienced a change in informality rates after the shock.23 . Overall, the results suggest that informality rates decrease between 5.0 and 8.0 percent after the transit shock. Table B5 in the Online Appendix reports the results for the share of informal firms. The results are similar to the ones for the share of informal workers. In particular, after the transit shock, informality rates decrease between 1 and 2.5 pp., which corresponds to a 2 to 3 percent decrease in informality rates when the mean in 1999 is used as a baseline. There are some issues with parallel trends since there are small effects but significant for 1999.24 In the next two sections, I test the robustness of my results and show that in terms of observed covariates, there is a negligible change in the composition of households, which is a potential concern of my identification strategy. I also construct bounds using retrospective questions that ask where were you living before. 3.2.3 Robustness Checks For the robustness checks, I compare locations close to line B of the subway with locations near subway expansions that the Government planned to build in the 1980s or actually built years later. In particular, panel b of Figure 1 plots a map of Mexico City highlighting the three lines that I will compare in this section: Line B, which is the infrastructure project that I’m studying; line C, a feeder line, similar to line B, that was to connect northwestern locations in the State of Mexico with the center of Mexico City, but was never built; and Line 12, which is the latest subway line, opened in 2012. I estimate the same difference-in-differences specification from equations 3.3 and 3.2. The only difference is that the treatment variable corresponds to a dummy variable indicating whether the centroid of the census tract is within some buffer zone of line B (i.e., 1500 meters), and similarly, the control group consists of locations within some buffer zone of line C and/or line 12. I run these regressions for four different buffers: 1500, 2000, 2500, and 3000 meters. Figure A4 in the Online Appendix depicts the point estimates for the log of the ratio between formal and informal workers from equation 3.2. I find a similar pattern to the previous results. The log of the ratio between formal and informal workers increases by approximately 10% when I compare treated locations with census tracts close to the other two lines. As shown, in the graph, this finding is robust to the use of different buffer zones and is very stable. 23 I also find similar results restricting the sample to census tracts with a centroid that is farther than 600 meters from one of the new stations. I also show that the results are similar if the dummy variable is constructed using a walking range of 20 minutes (Figure A11) 24 One reason that may explain the effects in 1999 is that the Government announced the construction of the line in 1994. 14 In addition, Figure A5 in the Online Appendix depicts the main result from these regressions. I plot the coefficients for the most restricted definition of workers’ informality. The main finding is that there is a negative relationship between informality rates and transit improvements when locations that experienced the shock are compared with census tracts close to lines that were planned in the 1980s. For instance, informality rates for workers decrease on average between 4.0 and 11.0 percentage points, which is a more significant effect than the one found previously. This effect corresponds to a decrease of approximately 15%, using the mean of the control group before the shock. In most of the specifications, I also find parallel trends, suggesting that after the shock treated locations experienced changes in informality. 3.2.4 Households’ Composition and Lower Bound Effects A potential concern regarding the identification strategy from the previous section is that locations close to the new subway line might experience a change in the composition of households due to worker sorting.25 For example, high-skilled workers that would prefer to work in the formal sector might migrate to these census tracts and, as a result, there would be a decrease in informality rates that could explain my findings. Ideally, I would deal with this issue by using a multi-year panel of workers before and after the shock. Unfortunately, no such panel is available. I deal with this concern by comparing household characteristics before and after the shock. The goal is to show that at least in terms of observable covariates, there was no change in households’ composition. For that purpose, I run the same specification in equation 3.2 on household character- istics, such as the high-skilled share of workers on the left-hand side. Table 3 reports the results, including all the set of controls. On average, I find that household characteristics in locations close to line B were not affected by the shock relative to other areas in Mexico City. For example, the point estimates for the share of high-skilled workers, the number of kids, or the household size are not significant. On the other hand, the coefficients that are significant are very small. For example, the student share’s point estimates imply that the locations affected by the shock experienced a slight increase of 0.4 percentage points. Overall, this finding suggests that at least in terms of observable characteristics, there is no change in households’ composition due to the transit shock that can bias my estimates.26 Furthermore, I also estimate lower bounds of the effect using retrospective questions. In par- ticular, INEGI asks the state in which the person was living before in the census. According to 25 In the model, I am allowing for changes in terms of unobserved characteristics since it allows for migration within the city. However, the model only assumes one type of worker and, therefore, I also analyze changes in households’ composition in terms of unobserved characteristics. 26 This result corroborates the findings of other papers in the Mexican context. For example, Gonzalez-Navarro and Quintana-Domeque (2016) exploit a random allocation of street asphalting in peripheral neighborhoods in Veracruz. The authors follow individuals for two years and find a negligible reallocation of households across locations in the city. Similarly, Hernández-Cortés et al. (2021) find negligible reallocation effects exploiting subways and BRT expansions in Mexico City. Moreover, it is related to other papers that have shown that there are high migration costs in developing countries. 15 this variable, 1.27% of the population living in the treated locations resided in Mexico City before. Around 971 thousand people were living in this area before the shock; this means that approxi- mately 12,500 people moved from Mexico City to the treated municipalities. Then, I can estimate a lower-bound effect under the extreme assumption that all the migrants were formal. In particular, removing these people from the specification yields a lower bound of the impact. Figure 5 plots the results of this lower bound for different values of the population that moved and decided to live in the treated areas after. For the value of 1.27%, there is a lower bound of 4.6%. Then, even under this extreme scenario, the ratio of formal to informal employment increases in the locations that experience the transit shock. 4 Model In this section, I present a quantitative model to assess first-order aggregate welfare effects of transit infrastructure on allocation. The model is based on recent work by Tsivanidis (2019), Monte et al. (2018), Heblich et al. (2018), and Ahlfeldt et al. (2015). My model extends this framework by adding intersectoral wedges and resource misallocation. The main theoretical result is a formula from a first-order approximation that decomposes the total change in welfare after a transit shock into three different components: the first is a “direct” effect term, and the second is an allocation term that can be decomposed into a resource misallocation term and an agglomeration externality term. This formula is similar to the general case from Baqaee and Farhi (2020) of GE models on changes in productivity. In the model, I assume that there are three groups of agents in the economy: workers denoted by L, house owners denoted by H , and commercial floor space owners denoted by Z .27 4.1 Preferences There is a mass of N locations in the economy that are indexed by n and i. There is a mass of LL workers that operate in 2 sectors indexed by s ∈ I, F , where I and F represent the informal and formal sectors respectively. The utility function takes a standard Cobb-Douglass form. Consumers obtain utility from a composite consumption good and housing. The utility function of worker ω is: α 1−α Cnisω Hnisω Unisω = · d− 1 ni · nisω , α 1−α where C is consumption, H is housing, the parameter α is the expenditure share on the consumption good, dni is an iceberg commuting cost to move from location n to i, and is an idiosyncratic shock 27 The focus of the paper is efficiency. In the Appendix, I generalize the results to consider different group of workers such as high- and low-skilled workers. Intuitively, the results are isomorphic if preferences for the formal and informal sector come from the scale parameters of Fréchet shocks, or if the commuting and labor supply elasticities differ between the two groups, and low-skilled workers prefer to work in the informal sector. 16 to worker ω . After solving the maximization problem, the indirect utility of worker ω living in location n and working in sector s and location i is wis d− 1 ¯ ni nisω (1 + t) Vnisω = αr 1− α , (4.1) Pn n where wis is the wage per efficiency unit in location i, and sector s, Pn is the price index of the ¯ is a proportional tax rebate from the Government. consumption good, rn is the rent for housing, and t In the Online Appendix, I show the results when the rebate is only given to formal workers. The term nisω is an idiosyncratic utility shock that is drawn from a nested Fréchet or extreme-value type II distribution H (·),   κ η  κ θs −θs H ( ) = exp − Bn  Bns  , with η < κ < θs ∀s.    nis n s i Each worker receives a one-time shock and makes three decisions, one for each nest: 1) location to live, 2) sector (formal or informal), and 3) workplace.28 . In the Online Appendix, I derive the model when the shock is to efficiency units instead of utility units. The parameters η, κ, and θs measure productivity dispersion across locations, sectors, and workplaces respectively and capture the notion of comparative advantage.29 On the other hand, the parameters Bn capture specific amenities that attract residents to each location n. I assume that these parameters are fixed over time. I allow the third parameter θs to differ across sectors to capture the fact that productivity differences across locations are larger in the formal sector, or in other words, that formal jobs are more difficult to substitute across locations than informal jobs. This parameter also represents the labor supply elasticity with respect to commuting costs conditional on working in sector s. The estimation θF < θI implies that workers in the informal sector are more sensitive to commuting costs, and thus, prefer to work close to their residence, as documented in Section 3. From the properties of the Fréchet distribution, the probability of living in location n and working in (i, s) is −αη −(1−α)η η κ Bns Wns θs −θs Bn Pn rn Wn |n wis dni λnisL = −αη −(1−α)η η κ θs −θs , (4.2) Bn Pn rn Wn s Bns Wns |n i wi s dni n λnL λnsL|n λnisL|ns 28 I am assuming that the idiosyncratic shock is to utility, but another possibility is to assume that the shock is to earnings. From a welfare point of view this assumption does not have any implications. In the Appendix, I consider a version of the model with Fréchet shocks to earnings and efficiency units. 29 Different articles have assumed a similar structure to analyze the allocation of workers across sectors. For example, Lagakos and Waugh (2013) study selection in the agricultural sector in developing countries using this kind of shock; Hsieh et al. (2019) study the allocation of talent in the past 50 years across different occupations in the US, and Galle et al. (2017) study the distributional implications of trade given that workers have idiosyncratic productivities for sectors. 17 κ = κ θs θs −θs where Wn s Wns |n is a wage index from location n, and Wns|n = i wis dni is a wage index from location n and sector s. This probability can be decomposed into three terms as in Monte et al. (2018). First, there is the probability of living in n; second, the probability of working in s conditional on living in n; and third, the probability of working in i conditional on living in n and operating in sector s. Note that i λnis|ns = 1, s λns|n = 1, and n λn = 1. Using again the properties of the Frechet distribution, I equate the expected ex-ante utility of a worker to the following constant: 1 η ¯L ≡ E[max Unis −αη −(1−α)η η U nis ] = Bn Pn rn Wn γη , (4.3) n ˜ is hired by (i, s) is equal to the where γη is a constant term.30 Then, the total amount of labor L amount supplied by all locations and is given by ˜ is = L ¯ L. λnis · L (4.4) n ¯n ≡ Thus, the average income received by workers that reside in n is y i,s λnis wis . 4.2 Production of the Composite Good Similar to Miyauchi et al. (2020), preferences for the composite good take a standard CES form of different varieties x across sectors and locations.31 It is described by a two-nested CES structure. In the first nest, consumers choose between sectors, and in the second nest, they choose between varieties j within each sector:32 ξ σs ξ−1 ξ−1 σs −1 σs −1 ξ σs Cn = Cns , Cns = xnisj dj , s i j where the parameter ξ captures the elasticity of substitution across sectors and the parameters σs capture the elasticity of substitution across varieties within sectors. Note that the lower nest parameter varies across sectors, hence, agglomeration externalities differ between the two sectors generating an additional allocation effect. In principle, we should expect σF < σI to capture that trade flows in the informal sector are more sensitive to trade costs and that agglomeration externalities are larger in the formal sector. I will estimate these parameters by estimating gravity equations. The price index Pn in location n, and the price indices for each sector Pns take the usual 30 The term γη = Γ(1 − 1/η ) and Γ(·) is the gamma function. This is the usual constant that arises after integrating the pdf from the Fréchet distribution. 31 Recent work on the public finance literature has shown that consumers, especially on the lower-income levels have preferences for varieties in the informal sector (Bachas et al., 2020). 32 The CES preferences can be micro-founded using extreme value-type distributions as in the literature that has studied the demand of heterogeneous consumers for a set of differentiated goods (Anderson and de Palma, 1992). For example, Miyauchi et al. (2020) uses this procedure. 18 CES functional form: 1 1 1−ξ 1−σs 1−ξ 1−σs Pn = Pns , Pns = pnisj dj , (4.5) s i j where pnisj is the price charged by firm j in (i, s) to consumers in n. I model the production of each good and the market structure as in the new economic geography literature (Helpman, 1995; Krugman, 1991). Firms compete monopolistically. To produce a variety a firm must incur both a constant variable cost and a fixed cost. Both costs use labor and commercial floor space with the same factor intensity across firms, which implies that the production function is homothetic. The variable cost varies with the productivity from location i and sector s, and it is represented by Ais . The total cost of producing xij units of variety j in location i and sector s is: xisj Γisj = Fs + (wis [1 + tisL ])βs (qi [1 + tisZ ])1−βs , (4.6) Ais where wis is the wage per efficiency unit in (i, s), qi is the price of commercial floor space, and Fs is a fixed cost that varies by sector to capture that the number of firms in the informal sector is larger. In the case of commercial floor space, both sectors face the same price. Finally, I add exogenous wedges represented by tisL and tisZ . These parameters represent taxes and subsidies in each sector and location (i.e., payroll taxes), and they imply that the marginal revenue of labor is not equalized across firms deviating from the optimum. Informal firms avoid paying these taxes, first generating dispersion in TFPR and then lowering TFP. I model informality in a different way relative to recent papers such as Ulyssea (2018) and Dix Carneiro et al. (2018).33 However, it captures the main differences between the formal and informal economy. First, differences in TFP captured by the parameter Ais and differences in the input intensity captured βs . Profit maximization implies that the equilibrium price is the standard constant mark-up in trade models over marginal cost. Firms also face iceberg trade costs τni to sell goods. In the empirical analysis, I assume that these trade costs also change after the transit shock. The price charged by firms in i to location n is σs τni (wis [1 + tisL ])βs (qi [1 + tisZ ])1−βs pnisj = . (4.7) σs − 1 Ais The zero-profit condition implies that the equilibrium output of each variety is constant across firms that operate in the same location and sector and is given by ¯is = Ais Fs (σs − 1). xisj = x (4.8) 33 In section D.4 of the Online Appendix, I consider a version of the model in which firms endogenously decide to operate in the formal vs. informal sectors following the logic from these studies. Moreover, firms also determine the location to operate in the city. 19 Aggregate payments to labor and commercial floor space, including taxes, are constant shares of the total revenue in location i and sector s. These shares are captured by βs and 1 − βs respectively:34 ˜is = (1 − βs )Yis . ˜ is = βs Yis , qi (1 + tisZ )Z wis (1 + tisL )L (4.9) From these expressions, I construct the labor demand. 4.2.1 Expenditure Shares The assumption of CES preferences implies a standard gravity relationship for bilateral trade flows in goods between locations for each sector. Using the CES demand, the price indices from equation 4.5, and the fact that all firms from (i, s) charge the same price, the share of location n’s expenditure on goods produced in (i, s) is: 1 1−ξ 1−σ 1−σs Pns Mis pnis 1−σs πnis = 1−ξ · 1−σ , with Pns = Mis pnis , (4.10) s Pns i Mi s pni s i πns πnis|s where Mis is the total number of firms in location i and sector s, πns is the share of expenditure in goods from sector s, and πnis|s is the expenditure share on goods from i conditional on consuming goods from sector s. Finally, since all firms within the same location and sector choose the same amount of labor and commercial floor space units, the total number of firms in each location i and sector s in equilibrium is a function of the aggregate amount of labor and commercial floor space:35 ˜ 1−βs ˜ βs Z ˜s L β is is Mis = , (4.11) σs Fs ˜s is a constant term that varies by sector. The fact that consumers have a love of variety where β (LOV) and that there is free-entry imply that there are agglomeration externalities for each sector. 1 As mentioned above, these agglomeration externalisties are captured by the elasticity σs −1 . Since the elasticity within the second nest varies by sector, agglomeration externalities generate an additional first-order effect as in Bartelme et al. (2019). 4.3 Housing and Commercial Floor Space ˜ that produce residential housing and ˜ , and Z I assume that there are two additional industries: H commercial floor space respectively. Both of these sectors are non-tradable goods (τniH ˜ → ˜ = τniZ ∞ ∀n = i) and operate under perfect competition in all locations. The only factors of production 34 Total revenue Yis = n απns πni|s Xn , where Xn is the expenditure from location n. 35 This model is akin to the perfectly competitive case in which there is a single firm in all locations and sectors, there is perfect competition and there are agglomeration externalities for each sector and location described by Ais = A˜is · L ˜ (1−β )γs , where γs = 1 . ˜ βγs Z is is σs − 1 20 of these sectors are the group of agents H , and Z . The former supplies units to residential housing, and the latter to commercial floor space. The production function for both sectors is linear in labor. There is no commuting for both groups, therefore, they only supply units where they live, which means that dniH = dniS → ∞ ∀n = i. The indirect utility of worker ω from group ν where ν ∈ {H, Z } from living in location n is: ¯nνω ≡ Bn wnν · nνω , U (4.12) Pnα · r 1−α n where νω is an idiosyncratic shock drawn from a Fréchet distribution with dispersion parameter ην , and location parameter Tiν , wnν is the wage per efficiency unit of group ν in location n. I assume that ην → 1, where ν ∈ {H, ˜ S˜}, this assumption replicates the specific factor case. Hence, the supply of residential and commercial floor space is perfectly inelastic and is fixed. Finally, from the production function of housing and the assumption of perfect competition, the price of housing in location n is rn = wnH , and the price of floor space is qi = wiZ . Using equation 4.9, which relates payments to labor and commercial floor space in terms of total revenue from (i, s), the equilibrium condition to clear the market of commercial floor space in each location i is ˜ is (1 − βs )(1 + tisL )wis L ˜i = qi Z . (4.13) s βs (1 + tisZ ) This equation equates the supply of commercial floor space described by the left-hand side to the demand by firms described by the right-hand side. Similarly,the residential floorspace market clearing condition is ˜ n = (1 − α)Xn , rn H (4.14) where Xn is total expenditure from location n, which I will explain later. This expression equates the total supply of housing to total demand. 4.4 Government Budget Constraint As mentioned above, the Government collects taxes and gives a rebate to households captured by ¯. I assume that the rebate is proportional to household income instead of a lump-sum so that the t Government does not distort migration decisions. This rebate is given by the following expression: ˜is = t ˜ is + tisZ qi Z tisL wis L ¯· Xn . (4.15) i,s n This equation equates the income of the government from the left-hand side to total expenditure on the right-hand side. I proceed to close the model by finding an expression of total expenditure in 21 each location. 4.5 Goods and Labor Market Clearing I now derive the equilibrium conditions for the goods market-clearing conditions. I analyze the expression first for total expenditure from location n, and then, for total revenue from (i, s). From equation 4.4, the total labor income received by agents of type g ∈ {L, H, Z } in location n is ˜ i,s wis Lnisg . Then, taking into account the proportional rebate from the government to households, yields that the total expenditure from location n: Xn = (¯ ¯). yn Ln + qn Zn + rn Hn ) (1 + t (4.16) On the other hand, the labor demand comes from consumer preferences and the production function. By the properties of the CES preferences, total revenue of location i and sector s, Yis , is given by: Yis = α πnis Xn . (4.17) n Finally, equating labor demand and labor supply, the goods market-clearing condition to close the model is: ˜ is = αβ wis (1 + tisL )L πnis Xn . (4.18) n This equilibrium condition implies that total payments to workers including taxes is equal to a fraction β of total revenue, where total revenue is a function of expenditures from all locations. ¯ create trade imbalances since aggregate Note that taxes tisL , tisZ , and the proportional rebate t expenditure is no longer equal to aggregate income in each location n. 4.6 Equilibrium The general equilibrium of the model is described by the following vector of endogenous variables: x = {wis , qi , rn , y ˜is , Ln }, ˜ is , Z ¯n , Wns , Pis , L ¯ given a set of exogenous parameters: and a constant U ¯ L A = {dni , τni , Ais , Bn , L, ¯Z , Z ¯H , L ˜ i , tisL , tisZ , Fs , θs , κ, η, σs , ξ, α, βs }, ˜i , H that solve the following system of equations: workplace and sector choice probabilities from equation 4.2; residence choice probabilities from equation 4.2; price indices from equations 4.5 and 4.7; total expenditure from equation 4.16; goods market clearing described by equation 4.18; commercial floor 22 space market clearing described by equation 4.13; housing market clearing described by equation 4.14; labor market clearing; and the Government budget constraint from equation 4.15. To assure that the equilibrium is unique, I assume the standard conditions for uniqueness in this class of GE models (Allen et al., 2015). Agglomeration externalities should be lower than congestion 1 forces. The parametric condition is (1 − βs ) > σs −1 ∀s. I proceed to analyze the effect of transit shocks on welfare using a first-order approximation. 4.7 Welfare Decomposition To aggregate welfare at the city level, I assume a social planner that takes a utilitarian perspective. Then, the aggregate welfare function is: ¯L + ωH U ¯ = ωL U U ¯S , ¯H + ωS U (4.19) where ωg represents the weights that replicate the efficient allocation of the economy.36 This equation suggests that aggregate welfare is a weighted average of the ex-ante utility of the three different types of agents in the economy. Let’s define L as an allocation of factors of production given a set of exogenous parameters A. ¯ achieved by the allocation L. I’m interested in the effect Specify U (A, L) as the welfare function U of shocks on aggregate welfare. By a first-order approximation, the total change in welfare of any trade/commuting shock is: ¯ ¯ ¯ = ∂ ln U d ln A d ln U + ∂ ln U dL . (4.20) ∂ ln A ∂L “Direct” effect Allocation/Agglomeration Equation 4.20 suggests that the effect of any shock can be decomposed into two different terms: a direct effect term that considers just changes in exogenous parameters as iceberg commuting costs dni or trade costs τni , and a first-order allocation term. This second term captures allocation from two different forces: wedges and differences in agglomeration externalities between the two sectors.37 38 ¯ ¯ For the parametric case of my model, these weights solve the following expressions: ωL 36 ¯ U UL = αβ , ωZ¯ U UZ = α(1 − β ), ¯ ωH UH and U ¯ = (1 − α). 37 The agglomeration externality component captures distortions from differences in markups or differences in pref- erences for love of variety across the two sectors. 38 This formula applies in the general class of urban models for any wedge, such as, variable market power across firms in product or labor markets. In the Appendix, I show this result. 23 Under the assumptions of the model described above, the explicit solution for this expression is: “Direct” effect = −αβ λnisL · d ln dni − α (βs λnL + (1 − βs )λnZ ) πnis · d ln τni (4.21a) n,i,s n,i,s ¯ tisL − t ¯ tnsZ − t ˜ ˜ns Allocation = α βs 1+t ¯ λnisL · d ln Lnis + (1 − βs ) 1+t¯ λnsZ · d ln Z (4.21b) n,i,s n,s βs 1 + tisL ˜ is + (1 − βs ) 1 + tisZ ˜is . Agglomeration = ¯ dL ¯ dZ (4.21c) i,s σs − 1 1+t i,s σs − 1 1+t The first term corresponds to a Hulten (1978) or “direct” effect term that comes from an envelope argument. It suggests that under the case of perfectly efficient economies, the cost time-saving approach captures the welfare effect of any trade/commuting shock. For instance, to measure the welfare gains from a transit improvement, it is sufficient to know the value of jobs in each link between n and i.39 This is the cost time-saving formula used by Train and McFadden (1978) to evaluate reductions in commuting costs. This implies that if the goal is to understand the aggregate gains, in the case in which the shock to commuting costs is very small, all the nominal effects cancel out. The second term captures changes in allocative efficiency. It suggests that if workers reallocate to sectors and locations with higher wedges, there is an increase in welfare. Hence, a transit shock may have an additional first-order impact in the presence of distortions. Intuitively, the sign depends on whether workers reallocate to firms with larger wedges. Firms that pay higher taxes have higher values of TFPR, while firms that do not pay taxes have very low values. Thus, if workers move to the firms with higher TFPR, the dispersion of TFPR decreases and the new equilibrium gets closer to the first-best allocation. Finally, the last term represents agglomeration externalities. This component arises only in the presence of externalities that differ between the two sectors as in BCDR or trade imbalances as in FG. This term captures the effect of these externalities on aggregate TFP and welfare. In my case, agglomeration externalities differ between the two sectors, and wedges and transfers create trade imbalances, so the third term also shows up in the formula. This component depends on two margins: differences in agglomeration externalities, and the wedge. Intuitively, if workers reallocate to the sector with bigger externalities, there are larger increases in welfare. For the wedge, the argument is similar to the second term. Firms that are paying higher taxes are small relative to the first-best due to trade imbalances; hence, reallocating workers to these firms increases welfare. I show the derivation of this formula in Section D.1 of the Appendix. I also generalized this result for different groups of workers and a general utility and production function by solving the social planner problem in Section D.2. The only assumptions for this derivation are that the utility function, production function, the consumption good aggregator, and the efficiency unit aggregator 39 In his seminal work, Hulten (1978) considers productivity shocks and shows that to measure their effect on GDP, it is sufficient to know the share of sector s on value added, or the so-called Domar weights. 24 are homogeneous of degree one.40 Most of the literature whose primarily goal is to measure the welfare gains from transit infras- tructure within cities has focused on the first term and direct effects, by assuming that there are no wedges in the economy and that it operates under perfect competition. I contribute to this literature by analyzing the effect of transit improvements on the second and third margins.41 5 Empirical Strategy and Estimation In this section, I describe the main empirical strategy and estimation of the main parameters. This section is divided into four parts: parametrization of commuting and trade costs; estimation of trade and commuting elasticities; estimation of the labor supply elasticity across sectors -κ-, and model inversion to recover the fundamentals of the economy such as technological and amenity parameters. 5.1 Trade and Commuting Costs For the counterfactual analysis, I parametrize commuting costs as in the urban economics literature (Ahlfeldt et al., 2015; Heblich et al., 2018; Tsivanidis, 2019). I assume that both iceberg commuting and trade costs are parametrized using the following expressions: dni = exp(δd timeni ), (5.1a) τni = exp(δτ timeni ), (5.1b) where timeni is the average travel time in minutes across different transportation modes of moving from location n to location i.42 The main objects of interest are the parameters δd , and δτ that transform travel times to iceberg costs. I estimate these parameters from a nested logit specification using the 2017 Origin-Destination Survey. I use trips to from home to work and vice-versa to estimate δd , and trips to restaurants, outlets, and retail shops to obtain the parameter δτ . The estimation is based on the following choice model. A worker ω is choosing between different transportation modes to travel from n to i. These transportation modes are grouped into different nests, for example public or private nests denoted by G . Denote the set of transportation modes in g , by Υg . The indirect utility of choosing transportation model m ∈ Υg ⊂ G is: 40 Holmes et al. (2014), Świ¸ecki (2017), and Asturias et al. (2016) use a similar formula using hat algebra. 41 Since this formula applies to the case in which the change in commuting/trade costs is infinitesimal, for the counterfactual analysis, I estimate and decompose the change in welfare using percentage changes and exact hat algebra. 42 I calculated a weighted average of travel times across the different transportation modes using each transportation mode’s aggregate share for commuting and consumption from the travel survey data. Hence, in terms of workers’ utility, the assumption is that transportation modes’ preferences take a Cobb-Douglas form. This is a conservative assumption. For example, in the case of CES or random idiosyncratic shocks, workers will substitute more other modes of transportation for the subway after the transit shock. 25 Vnimω = δ timenim + γm + ψnigω + (1 − λg ) nimω , where Vnimω is the indirect utility of worker ω if he/she chooses transportation mode m to travel from n to i. This is the classic framework that Berry (1994) studies. The parameter δ measures the sensitivity of the decision of the worker/consumer to the average time she spends on moving across locations.43 The parameter γm captures preferences for transportation mode m relative to a baseline mode; in my case, I normalize γbus to zero. For example, γcar captures preferences for car relative to buses, which can include the price of a car, or the stress of driving in a complicated city such as Mexico City. The variable ψ is common to all transportation modes for worker/consumer ω within group g and has a distribution function that depends on λ ∈ (0, 1). This latter parameter measures the correlation of errors within each nest. If this parameter is zero we are in the standard multinomial logit case. Finally, nimω is an idiosyncratic shock to worker ω of choosing m. The error term of this equation is ψnigω + (1 − λg ) nimω which is drawn from an extreme value-type I distribution. Table B6 shows the main result after estimating the nested logit specification. The first column reports the results for commuting, and the second column reports the results for trade trips. I obtain a value for δd of -0.009, which is consistent with previous findings from the literature (Ahlfeldt et al., 2015). The point estimate for δτ is -0.013, which is also consistent with the literature. On the other hand, in terms of preferences, when people go to work, the most preferred transportation mode is car, whereas, when they travel to restaurants or retail shops, the most preferred transportation mode is walking. The last two rows report the average iceberg commuting and trade costs across locations in Mexico City before and after the transit shock. On average, after Line B of the subway opens, commuting costs drop by 4.7%, and trade costs by 4.5%.44 5.2 Commuting and Trade Elasticities Commuting Elasticities: To estimate the commuting elasticities, I use the 2015 Intercensal Sur- vey. In this survey, workers report the municipality of their residence and workplace, and I am also able to define formal and informal workers by using employment and social security information. From the model, it is easy to derive the following gravity equation relating commuting flows across municipalities and iceberg costs: ln λnism|nsm = βs ·timenim + γism + γnsm + nism , (5.2) δd ·θs 43 I also estimate a heterogeneous δ between the formal and informal sector for the consumption trips. However, I do not find significant differences between the two coefficients, δI = −0.0128, and δF = −0.0114. These results are available upon request. However, I cannot distinguish between the formal and informal sectors for the working trips since I construct the commuting flows using the 2015 Intercensal survey. 44 I restrict the sample to trips that only use one transportation mode or two transportation modes + walking. 26 where the subindex m corresponds to one of four different transportation modes: car, metro or metrobus, bus, and walking; λnism|ns is the share of workers that commute to location i from location n working in sector s using the transportation mode m; timenim is the average commuting time across municipalities n, i using m; γnsm are origin-transportation-sector fixed effects; γism are destination-transportation-sector fixed effects, and nism captures the measurement error observed in the data of this gravity equation. The goal is to recover the parameters θs after knowing βs and δd described in the previous section. The parameter θs captures how sensitive workers are to commute in the formal/informal sector. From the evidence in Section 3, the expected result is that θI > θF , suggesting that informal jobs are easier to subsitute across location. I estimate this equation via the Poisson regression by pseudo maximum likelihood (PPML) to include the zero commuting flows between municipalities. Given the set of fixed effects, the identification comes from comparing the workplace decision of workers that use the same transportation mode and live (work) in the same municipality and sector, but work (live) in different places. Panel A in Table 4 reports the results. As expected, there is a negative relationship between commuting flows and the average commuting times. I find that the commuting elasticity in the formal sector is 3.11, and in the informal sector it is approximately 4.66. These values are consistent with the theoretical assumptions, and they confirm that informal workers are more sensitive to commuting costs than formal workers. Trade Elasticities: To estimate the trade elasticities, I use the 2017 OD Survey focusing on data on trips to different establishments. I restrict the sample to trips to restaurants, retail shops, and factory-outlets. I assume that people move across the city and spend their income on different consumption goods. To estimate a different trade elasticity for the informal and formal sectors, I use the fact that most informal establishments in Mexico correspond to restaurants and retail shops, while most formal establishments are manufacturers, as Figure A6 shows (Levy, 2018). I estimate the following gravity equation relating trade flows π s across municipalities (trips) with iceberg trade costs: ln πnism|sm = βs ·timenim + γism + γnsm + nism , (5.3) δτ ·(σs −1) where the different parameters represent the same variables as in equation 5.2. The identification comes from comparing trips to locations that use the same transportation mode and whose origin (destination) is the same, but in which individuals are moving to (from) a different municipality. I estimate this equation via PPML to include zero trips across locations. The goal is to recover the parameters σs . These parameters represent the elasticity of substitution across varieties for each sector. They measure how sensitive are trade flows to trade costs when people move across the city to buy different goods. In addition, according to the monopolistic model, they also represent 1 agglomeration externalities given by σs −1 . I allow these externalities to differ by sector, generating additional welfare effects from workers’ reallocation. One expected result is that σI > σF , indicating 27 that agglomeration forces are larger in the formal sector. The intuition for this result is that informal varieties are more substitutable than formal ones, and as a result, agglomeration externalities in the informal sector are lower. Panel B in Table 4 describes the main results for this estimation. As in all gravity equations, trade flows decrease with commuting times. The estimate of σs is consistent with the results from the previous literature. In particular, the elasticity of substitution in the informal sector is 6.94, and in the formal sector it is 5.39, suggesting that agglomeration externalities are 0.16 in the informal sector, and 0.22 in the formal sector. 45 5.3 Labor Supply Elasticity across Sectors In this section, I estimate the main equation from the model to recover the labor supply elasticity across sectors, κ. This parameter governs the reallocation of workers from the informal to the formal economy. I build market access measures following Tsivanidis (2019). According to the model, these measures represent the wage index for each sector. Hence, they capture whether workers obtained better access to formal jobs relative to informal jobs after the transit shock. For this estimation, I calculate travel times across the different census tracts in Mexico City with and without Line B of the subway using the network analysis toolkit from Arcmap. I compute travel times for three different transportation modes: car, walking, and the public transit system. I calibrate speeds for different types of roads and the public system using random trips from Google Maps. Table C1 describes the values obtained for each category and each mode of the transportation system.46 With the commuting times at hand, I define the commuter market access (CMA) for location 1 θs n and sector s as CMAns = Wns . This is an index of the accessibility of jobs in location n to employment in sector s. Following Tsivanidis (2019) and Donaldson and Hornbeck (2016), I can solve the following system of equations to compute MA measures for both firms and workers specific to each sector and location: ˜ is d−θs L Lns d− θs ni ni CMAns = , FMAis = , (5.4) FMAis n CMAns i ˜ is represents the total amount of labor hired by location i and sector s; Lns corresponds where L to the total number of workers that reside in location n and work in sector s; and FMAis is a firm market access measure that captures whether firms in i have good access to workers from sector s.47 45 These externalities are relatively large compared to the previous findings in the literature, where they are around 0.1. However, both numbers are still reasonable, especially in developing contexts. For example, Tsivanidis (2019) find that agglomeration externalities in Bogota are around 0.21, which is a larger value than those of previous findings. 46 Section C1 in the appendix explains the procedure. 47 Tsivanidis (2019) estimates these measures for Bogotá and shows that with data of commuting costs, and the number of residents and workers in each sector and location, the system of equation 5.4 has a unique solution. Another way to prove the existence and uniqueness of this system of equations is to apply the theorem from Allen et al. (2015). 28 After solving this system of equations, we can also recover the wage distribution from the market access approach. Figure A8 plots the wage distribution. The intuition of this system of equations follows the same logic as the case with only one sector. These measures capture whether residents from location n have good access to jobs from sector s, and similarly whether firms from location i have good access to labor. Figure A9 plots ventiles of the change in CMA for both sectors after the transit shock, holding constant the number of workers and residents. It is clear that locations close to the new subway line improved their market access to both formal and informal employment relative to other census tracts in Mexico City. Additionally, Figure 6 plots natural breaks of the change in CMA, taking the difference between the formal and informal sector. The figure shows that census tracts near line B experienced a larger increase in market access in the formal sector. As a consequence, workers in these census tracts obtained better access to formal jobs relative to the informal sector reallocating to firms with higher TFPR. I exploit this variation to estimate the labor supply elasticity parameter across sectors. From the structure of the model, I derive a log-linear relationship between the commuter market access θs = CMA . Then, from equation measures and the wage indices for each sector. In particular, Wns ns 4.2, and similar to the reduced-form results from Section 3, I estimate the following labor supply equation that correlates the change in the ratio between formal and informal residents with the change in CMA measures over time and across sectors: 1 1 ∆ ln LnF,t − ∆ ln LnI,t = κ ∆ ln CMAnF,t − ∆ ln CMAnI,t + βXn + γs(n) + nt , (5.5) θF θI where ∆ corresponds to the difference between 2000 and 2010; LnF,t , and LnI,t is the total number of residents that live in location n and work in the formal and informal sectors respectively; and γs(n) is a municipality or state fixed effect. I include a vector of controls Xn to capture specific trends that vary with initial characteristics. To recover κ, equation 5.5 is akin to a triple difference estimator. The first difference corresponds to time variation before and after the transit improvements, the second difference exploits heterogeneity of the treatment across locations, and the third difference uses variation in the market access measures across sectors. Equation 5.5 is a labor supply relationship and implies that people reallocate to the formal sector as they obtain better access to formal jobs relative to informal employment. As Figure 6 shows, Line B improved access to formal jobs for residents close to the new stations. It is important to mention that to estimate the parameter κ, the reallocation of workers wouldn’t bias the estimate of κ since the model allows for migration within the city, then according to the model, I can estimate κ comparing census tracts.48 The largest eigenvalue of this system of equations is 1. Thus, there is at most one strictly positive solution, up to scale with this system of equations. 48 In particular even if the people reallocate, the comparison to estimate κ needs to be across census tracts after 1 κ κ taking the ratio. The variable that determines who reallocates to the treated locations is Wn = (Wn I + WnF ) , and κ 29 One caveat with the estimation of equation 5.5 is that the change in CMA may capture other shocks in the economy that shifts the allocation of labor across sectors and locations. These shocks can change the decision of workers to operate in the formal or informal sector, thus, generating a correlation between the change in CMA and the error term. This generates a bias in the estimation of κ. To deal with this problem, I estimate equation 5.5 by two-stage least squares using two instruments. The first instrument is the change in the CMA measures when the number of residents and workers is held fixed, and the second instrument is the treatment dummy variable. The idea is to capture changes in commuting costs and clean the estimation from other economic shocks. For example, the treatment dummy variable captures changes in residents, employment, and market access in the treated census tracts only because of line B and not other economic shocks in the city. Table 5 reports the results for the labor supply elasticity across sectors. I obtained estimates of κ between 1.1 and 2.4. These estimates are consistent with the model and the commuting elasticities. The first two columns show the results for the OLS and the other four columns for the IV using each instrument separately. In my preferred specifications, which are the ones in columns 4 and 6, I obtained a point estimate between 1.5 and 2.4. For the counterfactuals, I take an average between these two numbers. Comparing the estimates from the 2SLS and OLS, it suggests that there were other shocks in the economy that created a downward bias for κ. For instance, these shocks reallocated workers from the informal to the formal sector generating a negative correlation between the change in the CMA measures and the error term.49 50 5.4 Labor and Capital Wedges Labor and capital wedges are a crucial parameter for the quantitative analysis. I follow the popular approach from Hsieh and Klenow (2009) and use the inverse of the wage bill and capital share to calibrate the distortions.51 From the profit-maximization condition, the inverse of the labor and commercial floorspace share paid by each firm is −1 −1 wis lis σs qis zis σs = (1 + tisL ) , = (1 + tisZ ) pis yis (σs − 1)βs pis yis (σs − 1)(1 − βs ) where wis lis is the wage bill, qis zis is the commercial floorspace payments, and pis yis are total sales or value-added. I can observe the left-hand side of this equation for each firm in the Economic Census, and use the average labor share βs and markups in each industry to calibrate the wedges. To aggregate from the firm level to the census-tract-sector cell, I take the mean of the inverse of the I am controlling for this variable after taking the ratio between the formal and the informal sectors since this wage index cancels out. Then, the reallocation of workers does not bias the estimation of κ since what matters is the share of formal to informal residents across census tracts even if there is migration. 49 I didn’t instrument the change in CMA with the two instruments since both of them capture changes in commuting cost because of the transit shock. 50 Relative to previous studies on estimating labor supply elasticities across sectors, such as Galle et al. (2017), Lagakos and Waugh (2013), and Berger et al. (2019), my estimates are similar. 51 Other papers such as Busso et al. (2012) and Levy (2018) that have explored the role of resource misallocation in Mexico also use the same method. 30 wage bill and capital share across firms in each cell. Figure A7 in the Online Appendix plots the labor-wedge distribution across locations for each sector in the baseline year. The wedges between the formal and informal sectors are very similar to the ones found by Busso et al. (2012). Formal firms face larger distortions. On average, the wedge in the formal sector is approximately 1.67 times the wedge in the informal sector. Furthermore, panel B of figure A10 in the Online Appendix shows the spatial distribution of labor wedges after I construct ventiles across locations. In places in the center of the city, where there is more economic activity and formal firms locate, wedges are larger. Moreover, for the counterfactual analysis, I also use a constant wedge for formal firms based on the work from Levy (2018) (Table 7.9). For the labor wedge, I use a conservative value of 0.95, and for the commercial floor space, a value of 0.75. These wedges include several distortions such as implicit taxes on salaried workers, regulations on dismissals and reinstatements, non-contributory social insurance, standard labor taxation like state payroll taxes, and firm taxation including REPECO and value-added taxes. 5.5 Other Parameters I calibrate other parameters of the model using simple moments of the data, or take them directly from the previous literature. I calibrate the expenditure share on housing using the ENOE and find, on average, a value of α = 0.75. Similarly, for the labor share, I use data from the Economic Census in 1999 and find a value of βI = 0.70, and βF = 0.6. To calculate the total amount of ˜ and commercial floor space Z housing H ˜ in each location, I use the area in square kilometers of buildings in each census tract from the Global Human Settlement Layer (GHSL) in 2000 weighted by the total number of employees and residents. To calibrate the fixed costs, I use the log-linear relationship between the total number of firms and the workforce in each sector from the model,and find FI = 0.15, and FF = 1.2. Section C.2 in the appendix specifies the details for this estimation. This result is consistent with the fact that for a firm, it is more difficult to produce in the formal sector. In addition, I use the estimate of the elasticity of substitution across sectors ξ = 2 from Edmond et al. (2015), which is similar to the estimates of other papers (Asturias et al., 2016). Also, I compute the counterfactuals using a value of η = 1.50 which is the lowest value of the migration elasticity that Tsivanidis (2019) finds for Bogotá, a similar context to Mexico City. This value is consistent with the assumption that η ≤ κ from the theoretical framework.52 5.6 Model Inversion In this section, I recover the fundamental parameters Bn , Bns , which capture differences in amenities that attract residents to each location and sector; and the parameters Ais , which represent differences 52 In section 6, I show that the results are robust to different values of the migration elasticity and the elasticity of substitution across sectors. 31 in productivity across locations. The argument is that knowing the key elasticities, and the number of workers and residents in each location and sector, I can identify the entire model from Section 4. Knowing these parameters, I can then compute trade flows and commuting flows and solve the counterfactuals using initial equilibrium conditions. I proceed in three steps. In the first step, I recover relative differences in amenities, and the wage distribution equating the labor supply to actual data. In the second step, I recover the productivity levels Ais equating the labor demand to the number of workers in the data. In the third step, I recover the amenity parameters Bn , equating the residents’ share in each location in the model to the data. Step 1: In a simultaneous step, I recover the entire wage distribution and the parameters Bns by equalizing the labor supply from equation 4.4 to the total number of workers in each sector and location from the data. I assume without loss of generality that BnI = 1. I identify BnF from the following relationship using the share of informal workers from the data in each location and the wages: BnF Wκ nF λnF |n = . BnF Wκ nF + Wκ nI I then identify the wage distribution by equalizing equation 4.4 to the number of workers from the data in the pre-period. Step 2: Using the vector of wages, I recover the productivity parameters Ais by solving the labor demand from equation 4.18. I solve for the vector of productivities, equating the labor demand implied by the model to the number of workers in each sector and location from the data. Step 3: With data on wages, and knowing the key elasticities, I can obtain the amenity parameters in each location Bn by equating the implied number of residents from the model with the number of residents from the data in the pre-period. In particular, I use λn from equation 4.2 in the model and equate it to the number of residents in the data. As a result, I then can compute trade flows for each sector across the city and solve for the counterfactuals using exact hat algebra as in Dekle et al. (2008). In Section D.3 of the Appendix, I provide the equilibrium conditions of the model with exact hat algebra. 6 Counterfactual Analysis This section describes the counterfactual analysis. To compute the welfare effects of Line B, I use the estimates of the key elasticities, and the commuting times with and without Line B. Then, I solve for the GE equilibrium before and after the shock. Regarding the distortions, I find a similar TFP effect as Busso et al. (2012), removing the wedges lead to gains of around 200%. I compute two different counterfactuals. The first one assumes that there is no migration within 32 the city and only solves the goods market-clearing condition. The second ones takes into account ¯ is the migration channel. I assume that the city is closed, so that the total number of workers L constant. I calculate changes in welfare and total output using percentage changes. To decompose the welfare effects into the three terms, I compute the equilibrium with and without the labor wedge, and for the agglomeration channel, I assume a different value of σs in the two sectors. Figure 7 plots the results for the different counterfactuals and Table 6 in the Online Appendix reports the numbers. Panel A and C holds the number of residents constant, while panel B and D add the migration margin decision. In panel C and D, I run the counterfactuals with a constant wedge for the formal sector. On average, Line B of the subway increased welfare between 1.7%-1.9%. Both changes in commuting and trade costs account for around 50% of the total gains. In terms of the welfare decomposition, I find that in the case in which the distortions are calibrated using the data, the “direct” effect term represents approximately 79% of the total gains, the reallocation of workers to the formal sector explains 18%, and the agglomeration externality component drives the remaining 2%. As a result, the allocation mechanism generated 26% additional gains relative to the standard case under the perfectly efficient economy. On the other hand, in the case in which I assume a constant wedge for the formal sector in the model, the direct effect explain a larger fraction of the total gains, 83%; the change in factor allocation explains 14%, and differences in external economies of scale between the two sectors explain 3%. The results are robust for different values of η , κ, and ξ .53 Relative to previous findings, and considering the size of my shock, these estimates are a bit higher. Nevertheless, these studies only considered changes in commuting costs and the direct effect. In my counterfactual, I’m analyzing changes in consumption costs and the allocative efficiency margin, which explains why the welfare effects are bigger. The project’s cost-benefit analysis implies that there was an increase of around 26% of real income net of the total cost at the aggregate level in the city. According to official documents from the Government, the total cost of Line B in 2000 was approximately USD 2,900 million in 2014, considering the net present value of maintenance, operational costs, and other overheads. This number represented approximately 0.72% of the total GDP of Mexico City in 2000. Then, in the benchmark case, line B generated an increase of around 2.59 USD per dollar spent on the infrastructure. This change would have been only 2.00 without considering the allocation mechanism. The new margin increased the effect on total welfare per dollar spent on the infrastructure by approximately 26%.54 For instance, if the city constructs a line or a road with a similar demand, but in places in which most of the workers are formal, the changes in welfare are smaller. The main takeaway from this analysis is that when policymakers assess the economic impact 53 These results are available upon request. 54 This number is obtained in the following way: in the perfectly efficient economy, the total gains are: 1.48% of the GDP, then the benefit per dollar spent on the project is 2.04 (1.47/0.72). By contrast, under the inefficient economy, the benefit is 1.86%, and the value per dollar spent on transit infrastructure is 2.59 (1.86/0.72). Thus, there was an increase of 26.3% relative to the perfectly efficient economy. 33 of transit infrastructure, it is critical that they consider other mechanisms that may affect welfare beyond common factors such as transportation demand. For example, when governments decide where to allocate future infrastructure, they should not only focus on connecting poor areas with efficient locations for distributional implications, but also for efficiency reasons. As this study shows, connecting informal workers with formal employment may generate additional welfare gains by reducing factor misallocation. Other policies In this section, I consider the effectiveness of other policies that the government can implement to reduce informality. I study two different types of policies. The first type consists of reductions in the entry fixed costs and the second type of placed based policies. Entry fixed costs First, I consider a policy in which the Government reduces the entry fixed cost of formal firms or increases it for informal firms. These policies are akin to making it easier for entrepreneurs to start a formal business in Mexico City (i.e., reducing red tape or bureaucracy) or to increase government regulations that make it more difficult for informal firms to enter the market. According to the reduced form estimates, Line B of the subway led to a decrease in informality rates at the aggregate level by 0.4%. Figure 8 plots the effectiveness of different policies that change the entry fixed cost for both formal and informal firms. Panel A plots the results for different rates decreasing the entry fixed cost for formal firms, and panel B simulates an increase in the informal entry fixed cost for different values. There are three main takeaways from this analysis. First, according to the model, it is more effective to reduce the entry fixed cost of formal firms relative to increasing the entry fixed cost of informal firms. For example, to decrease informality rates by 0.4% at the aggregate level, the government can lower the formal fixed cost by 5-8%, but it needs to increase the informal fixed cost by more than 8%. This suggests that it is more effective to focus on policies that benefit formal firms than to harm informal firms. Second, as the target of the Government increases, it becomes more effective to reduce the formal fixed cost relative to increasing the informal fixed cost. Third, the results suggest that transit infrastructure that connects informal workers with formal employment can be a useful tool to reduce informality rates. For example, if the government wants to generate similar results at the aggregate level, it needs to change the fixed cost by a substantial proportion. Overall, the findings imply that transit lines can be an excellent tool to reduce informality rates by giving better access to formal jobs to workers that live in remote areas compared to other types of policies that the government can implement. 34 Place-based policies For the second set of policies, I study whether place-based policies that reallocate formal firms in the city can effectively increase welfare and reduce informality rates. The intervention consists of increasing the commercial floor space employed by formal firms in different parts of the city. I consider two sets of policies; the first one consists of increasing the commercial floor space in the center of the city, and the second one in the outskirts. Figure 9 plots the locations in which the Government implements the policy; in total, there around 250 treated census-tracts in both parts of the city. The goal is to compare policies that reallocate formal firms to the outskirts vs. transit shocks that connect informal workers with formal jobs. Figure 10 plots the results of the intervention. In panel A, the Government increases commercial floor space in the central locations, and in panel B in the remote areas. The results suggest that it is more effective to intervene in the central locations than in the outskirts. For instance, if the Government increases the commercial floor space by around 40% in the central areas, the policy generates similar welfare gains to the transit shock that I studied. On the other hand, as shown in panel B, it is very ineffective to reallocate firms to the outskirts. Even if the Government increases the commercial floor space by a substantial proportion, 60%, it only increases welfare by 0.18%, which is significantly lower than the one obtained by the new subway line that connected informal workers with formal jobs. Moreover, in the latter case, the allocative efficiency margin and the externality component explain a very small fraction of the total gains. There are two main explanations for this result. The first one is that since most of the formal firms locate in the CBD, the agglomeration forces are minimal in the outskirts. The second one is that these locations are very unproductive in terms of the productivity scale parameters, especially for formal firms. Hence, reallocating firms to the outskirts generate negligible welfare gains. In general, the results imply that it is more effective to connect informal workers with formal jobs by transit lines than to move formal firms to the city’s remote areas through de-agglomeration policies. 7 Conclusion This paper has examined the welfare gains from transit improvements in developing countries, con- sidering the allocative efficiency margin. The mechanism that it studies is whether workers reallocate from the informal to the formal sector. I find that transit infrastructure that facilitates commuting may generate additional welfare gains by improving the market access of the informal labor force to formal employment. From an empirical perspective, the paper exploits a transit shock in Mexico City that connected poor and remote areas with the center of the city. The main finding is that informality rates decrease in the locations that experienced the shock relative to other areas in the city. This result implies that workers reallocated to firms with higher TFPR, thereby increasing welfare to a larger extent 35 than the predictions under perfectly efficient economies. On the quantitative side, the paper departs from the standard efficiency case in urban models that have studied the economic impact of transit infrastructure. The model extends the classic framework by adding wedges and resource misallocation. The paper quantifies the gains from transit infrastructure and finds that allocative efficiency drives approximately 17%-25% of the total gains. The results from this study are informative to policymakers in several aspects. First, it is critical that when they analyze the cost-benefit and opportunity cost of a project, they take into consideration other first-order effects that are driven not just by direct effects through the classic approach of transportation demand. These projects can have an additional economic impact through an allocative efficiency margin. For example, policymakers should consider whether the population that resides in the potential connected areas work in the informal or formal economy. The results suggest that even if a government is not concerned about distributional aspects, connecting poor areas with high-efficiency locations can generate larger gains than transit developments that link locations with a similar composition of workers through this new margin. Moreover, the results are informative on other public policy issues in urban areas. Programs that segregate informal workers and poor individuals in cities in developing countries, combined with high commuting costs, can increase the extent of resource misallocation, lowering both aggregate efficiency and TFP. Hence, governments must make decisions based on an analysis that considers all the first-order components that may affect welfare. 36 References Ahlfeldt, G. M., Redding, S. J., Sturm, D. M., and Wolf, N. (2015). The Economics of Density: Evidence From the Berlin Wall. Econometrica, 83:2127–2189. Allen, T., Arkolakis, C., and Li, X. (2015). On the Existence and Uniqueness of Trade Equilibria. Working Paper, Yale University. Alvarez, F. and Lucas, R. J. (2007). General equilibrium analysis of the Eaton-Kortum model of international trade. Journal of Monetary Economics, 54(6):1726–1768. Anderson, S. P. and de Palma, A. (1992). The logit as a model of product differentiation. Oxford Economic Papers, 44(1):51–67. Arkolakis, C., Costinot, A., Donaldson, D., and Rodríguez-Clare, A. (2019). The Elusive Pro- Competitive Effects of Trade. Review of Economic Studies, 86(1):46–80. Asturias, J., García-Santana, M., and Ramos, R. (2016). Competition and the Welfare Gains from Transportation Infrastructure: Evidence from the Golden Quadrilateral of India. Working Papers 907, Barcelona Graduate School of Economics. Atkin, D. and Khandelwal, A. (2019). How distortions alter the impacts of international trade in developing countries. Working Paper 26230, National Bureau of Economic Research. Bachas, P. J., Gadenne, L., and Jensen, A. (2020). Informality, Consumption Taxes and Redistri- bution. Policy Research Working Paper Series 9267, The World Bank. Balboni, C. (2019). In harm’s way? infrastructure investments and the persistence of coastal cities. Unpublished paper, MIT. Banerjee, A. V. and Duflo, E. (2005). Growth Theory through the Lens of Development Economics. In Aghion, P. and Durlauf, S., editors, Handbook of Economic Growth, volume 1 of Handbook of Economic Growth, chapter 7, pages 473–552. Elsevier. Baqaee, D. R. and Farhi, E. (2020). Productivity and Misallocation in General Equilibrium. The Quarterly Journal of Economics, 135(1):105–163. Bartelme, D., Costinot, A., Donaldson, D., and Rodríguez-Clare, A. (2019). External Economies of Scale and Industrial Policy: A View from Trade. Working paper, UC Berkeley. Baum-Snow, N. (2007). Did highways cause suburbanization? The Quarterly Journal of Economics, 122(2):775–805. Berger, D. W., Herkenhoff, K. F., and Mongey, S. (2019). Labor market power. Working Paper 25719, National Bureau of Economic Research. 37 Berry, S. T. (1994). Estimating discrete-choice models of product differentiation. The RAND Journal of Economics, 25(2):242–262. Busso, M., Fazio, M. V., and Algazi, S. L. (2012). (In)Formal and (Un)Productive: The Productivity Costs of Excessive Informality in Mexico. Research Department Publications 4789, Inter-American Development Bank, Research Department. Dekle, R., Eaton, J., and Kortum, S. (2008). Global Rebalancing with Gravity: Measuring the Burden of Adjustment. IMF Staff Papers, 55(3):511–540. Dix Carneiro, R., Goldberg, P., Meguir, C., and Ulyssea, G. (2018). Trade and Informality in the Presence of Labor Market Frictions and Regulations. Working Paper, Yale University. Donaldson, D. and Hornbeck, R. (2016). Railroads and American Economic Growth: A “Market Access” Approach. The Quarterly Journal of Economics, 131(2):799–858. Edmond, C., Midrigan, V., and Xu, D. Y. (2015). Competition, Markups, and the Gains from International Trade. American Economic Review, 105(10):3183–3221. Fajgelbaum, P. and Schaal, E. (2017). Optimal Transport Networks in Spatial Equilibrium. Working Paper, UC Los Angeles. Fajgelbaum, P. D. and Gaubert, C. (2020). Optimal Spatial Policies, Geography, and Sorting*. The Quarterly Journal of Economics, 135(2):959–1036. Fajgelbaum, P. D., Morales, E., Serrato, J. C. S., and Zidar, O. (2019). State Taxes and Spatial Misallocation. Review of Economic Studies, 86(1):333–376. Galle, S., Rodríguez-Clare, A., and Yi, M. (2017). Slicing the Pie: Quantifying the Aggregate and Distributional Effects of Trade. NBER Working Papers 23737, National Bureau of Economic Research, Inc. Gollin, D. (2002). Getting Income Shares Right. Journal of Political Economy, 110(2):458–474. Gollin, D. (2008). Nobody’s business but my own: Self-employment and small enterprise in economic development. Journal of Monetary Economics, 55(2):219–233. Gonzalez-Navarro, M. and Quintana-Domeque, C. (2016). Paving Streets for the Poor: Experimental Analysis of Infrastructure Effects. The Review of Economics and Statistics, 98(2):254–267. Gonzalez-Navarro, M. and Turner, M. A. (2018). Subways and urban growth: Evidence from earth. Journal of Urban Economics, 108(C):85–106. Heblich, S., Redding, S. J., and Sturm, D. M. (2018). The Making of the Modern Metropolis: Evidence from London. NBER Working Papers 25047, National Bureau of Economic Research, Inc. 38 Helpman, E. (1995). The Size of Regions. Papers 14-95, Tel Aviv. Hernández-Cortés, D., Olica, P., and Severen, C. (2021). Modal Choice, Income, and Congestion in Mexico City. Working paper, Federal Reserve Bank of Philadelphia. Holmes, T. J., Hsu, W.-T., and Lee, S. (2014). Allocative efficiency, mark-ups, and the welfare gains from trade. Journal of International Economics, 94(2):195–206. Hornbeck, R. and Rotemberg, M. (2019). Railroads, Reallocation and the Rise of American Manu- facturing. Working paper, University of Chicago. Hsieh, C.-T., Hurst, E., Jones, C. I., and Klenow, P. J. (2019). The allocation of talent and u.s. economic growth. Econometrica, 87(5):1439–1474. Hsieh, C.-T. and Klenow, P. J. (2009). Misallocation and Manufacturing TFP in China and India. The Quarterly Journal of Economics, 124(4):1403–1448. Hsieh, C.-T. and Moretti, E. (2019). Housing Constraints and Spatial Misallocation. American Economic Journal: Macroeconomics, 11(2):1–39. Hulten, C. R. (1978). Growth Accounting with Intermediate Inputs. The Review of Economic Studies, 45(3):511–518. Kanbur, R. (2009). Conceptualizing Informality: Regulation and Enforcement. Working Papers 48926, Cornell University, Department of Applied Economics and Management. Krugman, P. (1991). Increasing Returns and Economic Geography. Journal of Political Economy, 99(3):483–499. La Porta, R. and Shleifer, A. (2008). The Unofficial Economy and Economic Development. Brookings Papers on Economic Activity, 39(2 (Fall)):275–363. La Porta, R. and Shleifer, A. (2014). Informality and Development. Journal of Economic Perspec- tives, 28(3):109–126. Lagakos, D. and Waugh, M. E. (2013). Selection, agriculture, and cross-country productivity differ- ences. American Economic Review, 103(2):948–80. Levy, S. (2018). Under-Rewarded Efforts: The Elusive Quest for Prosperity in Mexico. Interamerican Development Bank. McCaig, B. and Pavcnik, N. (2018). Export Markets and Labor Allocation in a Low-Income Country. American Economic Review, 108(7):1899–1941. McMillan, M. S. and McCaig, B. (2019). Trade liberalization and labor market adjustment in botswana. Working Paper 26326, National Bureau of Economic Research. 39 Miyauchi, Y., Nakajima, K., and Redding, S. (2020). Consumption Access and Agglomeration: Evidence from Smartphone Data. Working paper, Princeton University. Monte, F., Redding, S. J., and Rossi-Hansberg, E. (2018). Commuting, Migration, and Local Employment Elasticities. American Economic Review, 108(12):3855–3890. Moreno-Monroy, A. I. and Posada, H. M. (2018). The effect of commuting costs and transport subsidies on informality rates. Journal of Development Economics, 130(C):99–112. Pérez Pérez, J. (2018). City minimum wages. Unpublished paper, Brown University, http://jorgeperezperez. com/files/Jorge_Perez_JMP. pdf (viewed July 19, 2018). Perry, G. E., Maloney, W. F., Arias, O. S., Fajnzylber, P., Mason, A. D., and Saavedra-Chanduvi, J. (2007). Informality : Exit and Exclusion. Number 6730 in World Bank Publications. The World Bank. Ramírez, S. B., Rojo, M. H., and Gault, D. A. (2017). Decisiones e Implementación en la Con- strucción de las Primeras Once Líneas de la Red del Metro en La Ciudad de México Hacia la Desorganización del Metro (1967-2000). Working Paper, Centro de Investigaci ón y Docencia Económicas CIDE. Redding, S. J. and Turner, M. A. (2015). Transportation Costs and the Spatial Organization of Economic Activity. In Duranton, G., Henderson, J. V., and Strange, W. C., editors, Handbook of Regional and Urban Economics, volume 5 of Handbook of Regional and Urban Economics, chapter 0, pages 1339–1398. Elsevier. Restuccia, D. and Rogerson, R. (2008). Policy Distortions and Aggregate Productivity with Het- erogeneous Plants. Review of Economic Dynamics, 11(4):707–720. Santamaría, M. (2020). The Gains from Reshaping Infrastructure: Evidence from the division of Germany. Working paper, University of Warwick. Suárez, M., Murata, M., and Campos, J. D. (2016). Why do the poor travel less? Urban structure, commuting and economic informality in Mexico City. Urban Studies, 53(12):2548–2566. ecki, T. (2017). Intersectoral distortions and the welfare gains from trade. Journal of Interna- Świ¸ tional Economics, 104(C):138–156. Train, K. and McFadden, D. (1978). The goods/leisure tradeoff and disaggregate work trip mode choice models. Transportation Research, 12(5):349 – 353. Tsivanidis, N. (2019). Evaluating the Impact of Urban Transit Infrastructure: Evidence from Bo- gotá’s Transmilenio. Working Paper, UC Berkeley. Ulyssea, G. (2018). Firms, Informality, and Development: Theory and Evidence from Brazil. Amer- ican Economic Review, 108(8):2015–47. 40 Figures Figure 1: Transit System (a) Line B (b) Other lines Notes: This figure plots a map of Mexico City with the transportation system. Panel (a) highlights the transit line -Line B- that I exploit in my main specification. On the other hand, panel (b) highlights the two lines that I use as a control group for the robustness checks. According to the transit expansion plan from 1980, line c -green line- was planned as a feeder line in the early 2000s, similar to line B. However, the Government of the city never constructed it. And line 12 -red line- is the latest subway line in Mexico City and was opened in 2012. The other lines correspond to the other subway lines of the actual system. 41 Figure 2: Commuting Time- Informal vs. Formal Notes: This figure plots the point estimate and 95th percentile confidence interval of a regression that relates the probability of commuting within some window of time with an informal dummy variable. The first bar reports the results for the category of non-commuting, the second bar if the worker spends on average between 1 to 15 minutes, the second bar between 16 to 30 minutes, the fourth bar between 30 to 60 minutes, the fifth bar between 60 to 120 minutes, and the sixth bar more than 120 minutes. The dark-blue bar does not include controls, while the light-blue bar includes individual controls and municipality fixed effects. Standard errors are computed with clusters at the municipality level. Figure 3: Spatial distribution of informality (a) Informal workers (b) Informal Residents Notes: This figure plots a map of Mexico City with the spatial distribution of informality rates. Panel (a) plots a heat map of workers’ informality rates by deciles in 1999. Panel (b) plots a heat map of residents’ informality rates by deciles in 2000. The main takeaway of this map is that in the middle-west and center of the city informality rates are lower than on the boundaries and east of Mexico City. As a result, informal workers that live in the outskirts have poor access to most of the formal employment, which is located in the center of the city. 42 Figure 4: Difference in Difference Results-Workers’ Informality Share (a) Informal workers (b) Informal and non-salaried workers Notes: This figure depicts the point estimates and 90th percentile confidence interval from the difference in difference specification relating workers’ informality rates with the transit shock. The treatment group are census tracts with centroids within a walking range of 25 minutes to stations of line B. The control group are census tracts in Mexico City. Panel (a) reports the results for the share of informal workers, and panel (b) for the share of informal and non-salaried workers. Standard errors are clustered at the census tract level. Figure 5: Lower-bound of Transit Infrastructure Impact Notes: This figure plots the results of the lower bound effect under the extreme assumption that all workers that move from the city to the outskirts were formal. The red line represents the point estimate of the main specification, the orange line the share of workers in the treated areas that live in the locations that experience the shock, and the blue line the difference-in-difference point estimate when the specification removes the people that moved. 43 Figure 6: Change in CMA across sectors Notes: This figure plots a heat map of Mexico City with the spatial distribution of the change in CMA across sectors after the transit shock. I construct natural breaks across locations by taking the difference between the formal and informal sector of CMA before and after the shock. Each color represents one of the natural breaks categories. Blue colors represent a very small change, while red color a very large change. From the figure, census tracts close to the new line got better access to formal employment relative to the informal sector. Thus, workers reallocate to the formal sector. 44 Figure 7: Counterfactual results (a) No migration (b) Migration (c) No migration-constant wedge (d) Migration-constant wedge Notes : This figure plots the counterfactual results. Panel (a) and (c) show the results for the counterfactual with no migration, and panel (b) and (d) for the counterfactual in which there is migration. In panel (a) and (b), I calibrate the distortions using value added and in panel (c) and (d) a constant wedge for the formal sector based on Levy (2018). 45 Figure 8: Counterfactual results-Fixed costs (a) Decrease formal costs (b) Increase informal costs Notes: This figure plots the counterfactual results for changes in the entry fixed cost for both formal and informal firms. Panel (a) shows the results for a counterfactual reducing formal fixed costs, and panel (b) for a counterfactual increasing informal fixed costs. The objective of the government is to reduce informality rates by 0.5%, which is the aggregate effect that I find from the transit shock. Figure 9: Place-based policies Notes: This figure plots a map of Mexico City with the locations in which the Government increases the commercial floorspace for formal firms. The central locations are in red, and the remote locations in blue. 46 Figure 10: Counterfactual results-Place based policies (a) Place-based policies central locations (b) Place-based policies remote locations Notes: This figure plots the counterfactual results for changes in the supply of commercial floor space for formal firms. Panel (a) shows the results for a counterfactual increasing commercial floor space in central locations, and panel (b) in remote areas. The objective of the government is to increase welfare by 1.84%, which is the effect from the transit line. 47 Tables Table 1: Difference-in-Difference - Share of Informal Residents (1) (2) (3) (4) (5) (6) (7) (8) Outcome: ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) ∆(ln LF − ln LI ) Panel A: Continuous treatment measure-Pool of residents - ln distance 0.040*** 0.054*** 0.045*** 0.058*** 0.014* 0.030*** 0.018** 0.035*** (0.007) (0.008) (0.008) (0.008) (0.008) (0.008) (0.009) (0.009) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.162 0.248 0.162 0.248 0.230 0.300 0.230 0.301 Panel B: Treatment dummy variable-Pool of residents Ti 0.038** 0.069*** 0.033** 0.067*** 0.024 0.068*** 0.016 0.064*** (0.016) (0.016) (0.016) (0.016) (0.018) (0.016) (0.018) (0.017) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.156 0.241 0.156 0.240 0.230 0.300 0.230 0.300 Panel C: Continuous treatment measure-Low skilled residents -ln distance 0.049*** 0.056*** 0.053*** 0.060*** 0.017* 0.032*** 0.021** 0.036*** (0.008) (0.008) (0.008) (0.008) (0.009) (0.008) (0.010) (0.009) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.137 0.230 0.138 0.230 0.203 0.281 0.203 0.282 Panel D: Treatment dummy variable-Low skilled residents Ti 0.051*** 0.071*** 0.046*** 0.069*** 0.027 0.068*** 0.019 0.065*** (0.017) (0.016) (0.017) (0.016) (0.019) (0.017) (0.019) (0.017) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.130 0.222 0.130 0.221 0.202 0.281 0.202 0.281 Panel E: Continuous treatment measure-Outskirt area -ln distance 0.072*** 0.088*** 0.079*** 0.095*** 0.037*** 0.050*** 0.041*** 0.055*** (0.009) (0.009) (0.010) (0.010) (0.010) (0.009) (0.011) (0.010) Observations 2,171 2,171 2,171 2,171 2,171 2,171 2,171 2,171 R-squared 0.199 0.279 0.200 0.279 0.279 0.338 0.280 0.338 Panel F: Treatment dummy variable-Outskirt area Ti 0.076*** 0.138*** 0.066*** 0.131*** 0.062*** 0.110*** 0.048** 0.099*** (0.021) (0.019) (0.022) (0.019) (0.023) (0.021) (0.024) (0.022) Observations 2,171 2,171 2,171 2,171 2,171 2,171 2,171 2,171 R-squared 0.185 0.264 0.184 0.262 0.277 0.336 0.276 0.335 Distance Meters Minutes Meters Minutes Meters Minutes Meters Minutes Dist.+Prod. Controls X X X X X X X X Population Controls X X X X State FE X X X X Municipality FE X X X X Notes: This table reports the results of a regression relating changes in the share of informal residents in each location with the line B of the subway. Panel A reports the results for the continuous treatment measures and the pool of residents, panel B for the treatment dummy variables and the pool of residents, panel C for the continuous treatment measure and low-skilled workers, panel D for the treatment dummy variables and low skilled workers, panel E for the continuous treatment measure on the locations that are not in the CBD, and panel F for the treatment dummy variable on the locations that are not in the CBD. In the first four columns, I include state-time fixed effects, and in the fifth column to the eight column municipality-time fixed effects. The regressions are weighted by the population in 2000. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 48 Table 2: Difference-in-Difference - Log individuals (1) (2) (3) (4) (5) (6) Outcome: ∆ ln Li ∆ ln LiF ∆ ln LiI ∆ ln Li ∆ ln LiF ∆ ln LiI Panel A: Pool of workers Ti 0.017* 0.057*** -0.010 -0.006 0.030*** -0.034*** (0.009) (0.010) (0.013) (0.009) (0.011) (0.013) Observations 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.310 0.417 0.177 0.365 0.458 0.251 Panel B: Low-skilled workers Ti 0.022*** 0.067*** -0.002 -0.001 0.039*** -0.026** (0.008) (0.011) (0.013) (0.009) (0.011) (0.013) Observations 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.440 0.489 0.220 0.465 0.513 0.280 Panel C: High-skilled workers Ti 0.008 0.022 -0.014 -0.011 0.004 -0.038** (0.013) (0.014) (0.017) (0.012) (0.013) (0.017) Observations 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.446 0.442 0.375 0.497 0.492 0.427 Controls X X X X X X State fe X X X Municipality fe X X X Notes: This table reports the results of a regression relating changes in the log of the number of individuals in each location and sector with the line B of the subway. Panel A reports the results for the pool of workers, panel B for low-skilled workers, and panel C for high-skilled workers. In the first three columns, I include state-time fixed effects, and in the fourth column to the sixth column municipality-time fixed effects. The first and fourth column reports the results for the overall number of individuals, the second and fifth column for individuals in the formal sector, and the third and sixth column for workers in the informal sector. The regressions are weighted by the population in 2000. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 49 Table 3: Change in covariates after the transit shock (1) (2) (3) (4) (5) (6) (7) (8) Outcome: Number of kids Household size - ln distance 0.008 -0.004 0.015 0.011 (0.042) (0.052) (0.013) (0.014) Ti -0.026 -0.049 0.024 0.011 (0.091) (0.104) (0.030) (0.029) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.038 0.038 0.076 0.076 0.060 0.060 0.076 0.076 Outcome: Male dummy Age - ln distance 0.000 -0.000 -0.008 0.031 (0.000) (0.000) (0.021) (0.026) Ti -0.001 -0.002*** 0.008 0.107* (0.001) (0.001) (0.050) (0.057) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.255 0.255 0.273 0.273 0.137 0.137 0.179 0.180 Outcome: High-skilled share Student share - ln distance -0.000 -0.001 -0.002** -0.001 (0.001) (0.001) (0.001) (0.001) Ti -0.002 -0.001 -0.006*** -0.004** (0.002) (0.002) (0.002) (0.002) Observations 3,192 3,192 3,192 3,192 3,192 3,192 3,192 3,192 R-squared 0.244 0.244 0.310 0.310 0.142 0.143 0.174 0.175 Controls X X X X X X X X State FE X X X X Municipality FE X X X X Notes: This table reports the results of a difference-in-difference specification relating changes in household composi- tion and covariates with the transit shock. The odd columns report the results for the continuous treatment variable, and the even columns for the treatment dummy variable. The regressions are weighted by the population in 2000. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 50 Table 4: Gravity Equations-Commuting and Trade Elasticities (1) (2) Formal sector Informal sector Panel A: commuting Outcome ln λniF ln λniI Minutes -0.028*** -0.042*** (0.003) (0.005) Observations 2,257 2,280 R-squared 0.535 0.518 Implied θ 3.11 4.66 Panel B: Trade Outcome ln πniF ln πniI Minutes -0.059*** -0.078*** (0.004) (0.005) Observations 2,128 2,108 R-squared 0.406 0.497 Implied σ 5.39 6.94 Origin -Transportation mode FE X X Destination -Transportation mode FE X X Notes: This table reports the results of a gravity equation relating commuting and trade flows at the municipality level with the average time for four different transportation modes: car, bus, metro or metrobus (brt), and walking. I estimate this regression via the PPML method to include the zeros. The first column presents the results for the formal sector the second column for the informal sector. Standard errors are clustered at the municipality of origin level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 51 Table 5: Estimation of Labor Supply across sectors (1) (2) (3) (4) (5) (6) OLS OLS IV1 IV1 IV2 IV2 Outcome: ∆t,s ln Ls ∆t,s ln Ls ∆t,s ln Ls ∆t,s ln Ls ∆t,s ln Ls ∆t,s ln Ls κ 0.179** -0.383*** 1.142*** 1.516*** 1.490*** 2.387*** (0.078) (0.102) (0.142) (0.222) (0.351) (0.643) Observations 3,192 3,192 3,192 3,192 3,192 3,192 Adjusted R-squared 0.237 0.296 0.247 0.302 0.238 0.295 FS1 FS1 FS2 FS2 Outcome: ∆t,s CMA ∆t,s CMA ∆t,s CMA ∆t,s CMA ∆t,s CMA 1.981*** 1.407*** (0.093) (0.070) Ti 0.047*** 0.028*** (0.004) (0.003) Observations 3,192 3,192 3,192 3,192 Adjusted R-squared 0.588 0.852 0.523 0.822 F-stat 588.91 708.68 77.06 59.95 Controls X X X X X X State FE X X X Municipality FE X X X Notes: This table reports the results of the estimation of the labor supply elasticity to recover the parameter κ, which governs the reallocation from the informal to the formal sector. The dependent variable is the change in the log ratio between formal and informal workers. The independent variable is the change in CMA across sectors. The first two columns show the results for the OLS. The third and fourth column displays the results of a two-stage least square estimation using as an instrument the change in CMA across sectors and holding constant the number of workers and residents. The fifth and sixth column display the results of a two-stage least square estimation using as an instrument a dummy variable indicator of whether the centroid of the census tract is within a 25 minute walking range. The odd columns include state fixed effects, and the even columns include municipality fixed effects. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 52 ˆ = X /X Table 6: Counterfactual Results X (1) (2) (3) (4) (5) (6) Panel A: Percentage change in welfare-Distortions No Migration Migration %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Total change 0.95% 0.77% 1.86% 0.92% 0.73% 1.75% Decomposition Pure Effect 91.94% 65.36% 79.16% 91.70% 65.51% 79.27% Allocation 7.17% 31.17% 19.20% 7.21% 30.54% 18.64% Agglomeration 0.89% 3.46% 1.64% 1.08% 3.95% 2.09% Panel B: Percentage change in welfare-Constant wedge %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Total change 0.93% 0.71% 1.76% 0.91% 0.68% 1.68% Decomposition Pure Effect 93.60% 71.26% 83.71% 92.72% 70.70% 82.90% Allocation 5.62% 23.55% 14.01% 6.29% 23.61% 14.39% Agglomeration 0.78% 5.19% 2.28% 0.99% 5.69% 2.71% Notes : This table reports the counterfactual results for the line B of the subway. The first and fourth column considers only change in commuting costs, the second and fifth column changes in trade costs, and the third and sixth column considers changes in both type of iceberg costs. The first three columns presents the results for the counterfactual with no migration, and the second three columns for the counterfactual in which I allow for migration in the model. Panel A reports the results for welfare with the calibrated distortions a lá Hsieh & Klenow (2009), and panel B for welfare with a constant wedge in the formal sector based on Levy (2018). The first row describes the results considering the total change. While, the other rows decompose the total change into the different components. The second row shows the percentage explained by the direct effect, the third row by the allocative efficiency margin, and the fourth row by the agglomeration externality component. 53 Online Appendix: Spatial Misallocation, Informality, and Transit Improvements A Additional Figures 1 B Additional Tables 8 C Data and Quantification Appendix 12 C.1 Calibration of Speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 C.2 Calibration of Fixed Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 C.3 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 C.4 Additional Infrastructure Counterfactuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 C.5 Formal workers rebate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 C.6 Line 1 and 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 D Theoretical Appendix 20 D.1 Welfare Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 D.2 The problem of the social planner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 D.3 Equilibrium Conditions - Exact Hat Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 D.4 Model with ex-ante firm decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 0 A Additional Figures Figure A1: Informality Rates-Latin America and the Caribbean 80 60 Informality Rate 4020 0 ile y a il a a or bia p. o u la gua ia ay Ric uel tin a xic Per z ma liv Re vad ugu agu Ch Bra lom ara gen nez Me Bo ate sta m. Sal Par Ur Nic Co Ar Ve Gu Do Co El Notes: This figure plots informality rates across countries from Latin America and the Caribbean. The data source is the online appendix from Ulyssea (2018) that uses data from SEDLAC, an initiative from the World Bank and Universidad Nacional de la Plata. Informal workers are defined as those without social security. The orange line represents the average informality rate of countries from the OECD. The figure shows that informality rates in LAC are very high, and even within the region, Mexico is one of the countries with the highest informality rates. Figure A2: Firm size and Productivity Distribution-Economic Census 1999 .6 .4 .3 .4 Density function Density function .2 .2 .1 0 0 1 4 16 64 256 1024 4096 16384 65536 -5 0 5 10 Firm Size Sales per worker Legal and informal Illegal and informal Legal and informal Illegal and informal Mixed Legal and formal Mixed Legal and formal (a) Firm size (b) Productivity Notes: This figure plots the firm size and productivity distribution for the four different categories of firms: 1) Legal and informal 2) Illegal and informal, 3) Mixed, and 4) Legal and formal. I use the 2004 economic census. Panel (a) plots the firm size distribution and panel (b) the productivity distribution. Firm size is measured as the number of workers, and productivity as the logarithm of sales per worker. 1 Figure A3: Difference in Difference Results-Residents’ Informality Share Notes: This figure depicts the point estimate and 90th percentile confidence interval of a regression that relates the change over time in the log of the ratio between formal and informal residents with the transit shock. The treatment variable takes a value of 1 for census tracts with a centroid within a 25 minutes walking range. The first three bars show the results of a regression including distance and population controls with state fixed effects, and the second three bars report the results with municipality fixed effects. Standard errors are clustered at the census tract level. The dark-blue bar reports the results for the pool of workers, the middle-blue bar for low-skilled workers and the light-blue bar for high skilled workers. Line B increased the ratio of formal to informal residents on approximately 7% when I compare treated areas vs. the rest of Mexico City. Figure A4: Robustness Checks-Residents’ Informality Share Notes: This figure depicts the point estimate and 95th percentile confidence interval of a regression that relates the change over time in the log of the ratio between formal and informal residents with the transit shock. The treatment variable takes a value of 1 for census tracts whith a centroid within a buffer zone of the new subway line. The control group are locations within a buffer zone of line C or line 12. These were subway lines that the Government planned to build in the 1980s, but it didn’t construct in my period of analysis. I use different buffer zones: 1500, 2000, 2500, and 3000 meters. The first bar shows the results for 1500 meters, the second bar for 2000 meters, the third bar for 2500 meters, and the fourth bar for 3000 meters. Standard errors are clustered at the census tract level. 2 Figure A5: Robustness checks-Workers’ Informality Share (a) Buffer: 1500 meters (b) Buffer: 2000 meters (c) Buffer: 2500 meters (d) Buffer: 3000 meters Notes: This figure depicts the point estimates and 95th percentile confidence interval from the difference-in-difference specification using different buffers and different control groups. The treatment group is defined as census tracts that are within a buffer to the new stations. The control group are census tracts within a buffer to lines that the Government planned to build in 1980, but that were not constructed in my period of study. The outcome variable is the share of informal and non-salaried workers and the specification includes state-time fixed effects. Panel (a) reports the results for a buffer of 1500 meters, panel (b) for 2000 meters, panel (c) for 2500 meters, and panel (d) for 3000 meters. The blue line pools together as a control group the locations close to line C and line 12, the orange lines locations close to line C, and the green line are locations close to line 12. Standard errors are clustered at the census tract level. 3 Figure A6: Informal/formal sector by industry 1 0.15 0.53 0.46 0.85 .8 Informal and formal employment .6 0.54 0.47 .4 .2 0 Manufacturing Retail Service Formal Informal Source: Levy (2018) Notes: This figure plots the share of employment by industry between the formal and informal sector. The information comes from the book by Levy (2018), who uses the 2014 Mexican Economic Censuses. In his book, like this study, the author defines the informal and formal sector using the contractual relationship between the firm and the worker. An establishments is informal if it only hires non-salaried workers or if it does not provide social security to their workforce. Figure A7: Distribution of Labor Wedges by Sector Notes: This figure plots the distribution of the labor wedge by sector across the different census tracts. I follow Hsieh and Klenow (2009) to calculate the labor wedge for each sector-location cell using the inverse of the labor share. In particular the distortion is computed using the following relations w is Lis pis yis . The blue line depicts the labor wedge distribution for the formal sector, and the red line for the informal sector. The figure suggests that conditional on productivity, formal firms are too small relative to a perfectly efficient allocation since these firms have higher levels of total factor revenue productivity (TFPR). The marginal revenue product of labor does not equalize across firms. 4 Figure A8: Distribution of Wages by Sector Notes: This figure plots the wage distribution obtained from the market access measures and the number of workers in each census tract. According to the definition of firm market access, wis θs = Lis FMA− 1 is . The blue line depicts the wage distribution for the formal sector, and the red line for the informal sector. The model replicates that the formal sector pays a wage premium. This value is approximately 55% by comparing the wage median between the formal and informal sector. Figure A9: Change in CMA for each sector (a) ∆ CMA Formal Sector (b) ∆ CMA Informal Sector Notes: This figure plots a map of Mexico City at the census tract level with the spatial distribution of the change in CMA after the transit shock for each sector. I construct ventiles for the change in CMA across locations before and after the transit shock. Each color represents one quantile category. Blue colors represent a very small change, while red color a very large change. Panel (a) plots a heat map for the formal sector, and panel (b) for the informal sector. From the figure, it is clear that locations that experienced the shock and are close to the new stations got better access to both formal and informal employment. 5 Figure A10: Spatial Distribution of Productivity and the Labor Wedge (a) Productivity (b) Labor wedge Notes: This figure plots a map of Mexico City with the spatial distribution of productivity measured as value added per worker. I construct ventiles across locations after aggregating value added measures and the total number of workers. Each color represents one of the quantile categories. Census tracts in central areas have higher productivity measures. Figure A11: Difference in Difference Results-Workers’ Informality Share-20 minutes (a) Informal workers (b) Informal and non-salaried workers Notes: This figure depicts the point estimates and 90th percentile confidence interval from the difference in difference specification relating workers’ informality rates with the transit shock. The treatment group are census tracts with centroids within a walking range of 20 minutes to stations of line B. The control group are census tracts in Mexico City. Panel (a) reports the results for the share of informal workers, and panel (b) for the share of informal and non-salaried workers. Standard errors are clustered at the census tract level. 6 Figure A12: Difference in Difference Results-Residents’ Informality Share-20 minutes Notes: This figure depicts the point estimate and 90th percentile confidence interval of a regression that relates the change over time in the log of the ratio between formal and informal residents with the transit shock. The treatment variable takes a value of 1 for census tracts with a centroid within a 20 minutes walking range. The first three bars show the results of a regression including distance and population controls with state fixed effects, and the second three bars report the results with municipality fixed effects. Standard errors are clustered at the census tract level. 7 B Additional Tables Table B1: Informality and Commuting Patterns (1) (2) (3) (4) (5) Panel A: Probability of working in the same municipality of residence Outcome: Workplace municipality Workplace municipality Workplace municipality Workplace municipality Workplace municipality Informal -0.265*** -0.231*** -0.231*** -0.132*** -0.079*** (0.009) (0.008) (0.008) (0.006) (0.008) Observations 577,041 577,039 577,039 517,354 516,931 R-squared 0.069 0.098 0.123 0.215 0.465 Panel B: Probability of working in the CBD of Mexico City Outcome: Workplace-CBD Workplace-CBD Workplace-CBD Workplace-CBD Workplace-CBD 8 Informal -0.086*** -0.056** -0.059*** -0.037*** - (0.026) (0.024) (0.018) (0.011) - Observations 577,041 577,039 577,039 517,354 - R-squared 0.007 0.042 0.468 0.444 - Individual Characteristics X X X X Origin FE X X X Transportation Mode FE X X Destination FE X Notes: This table reports the results of a linear probability model relating the probability of working in the same municipality as the one in which the worker resides, and the probability of working in the CBD with a dummy variable that takes the value of 1 if the worker is informal. Panel A reports the results for working in the same municipality, and panel B whether the individual works in the CBD. Standard errors are clustered at the residence municipality level and reported in parentheses. ∗p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01. Table B2: Descriptive Statistics 1999 and 2000 Panel A: Outcomes Variable Mean Sd Min Max Share informal workers 60.25% 33.37% 0.00% 100.00% Share informal and non-salaried workers 43.47% 29.60% 0.00% 100.00% Share informal firms 84.15% 18.26% 0.01% 100.00% Share informal residents 46.68% 10.58% 1.00% 91.97% Share informal high-skilled residents 35.65% 7.47% 1.42% 76.99% Share informal low-skilled residents 50.31% 10.47% 1.07% 93.01% Panel B: Treatment Variables Variable Mean Sd Min Max Euclidean Distance to new stations (meters) 11223.33 6625.81 411.89 32838.87 Walking Distance to new stations (minutes) 124.70 73.62 4.58 364.88 Dummy variable (dist<2463) 10.74% 30.97% 0.00% 100.00% Dummy variable (minutes≤25) 10.00% 30.04% 0.00% 100.00% Notes: This table reports summary statistic of the main variables. Panel A presents the statistics for the outcomes of interests: workers’ informality rates from the Economic Census in 1999 and residents’ informality rates from the Population Census in 2000. Panel B for the different definitions of the treatment group that includes: the euclidean distance, the network walking distance, a dummy variable whether the centroid of the ageb is within buffer zone of 2000 meters to the new stations, and a dummy variable whether the centroid of the ageb is within a 25 minutes walking range. Table B3: Results: Census tract characteristics 1999 and 2000 vs. Treatment (1) (2) (3) (4) Outcome: ln Income High Skill Share Occupation share Informality Rates Ti -0.026*** -0.032*** -0.012*** 0.026*** (0.009) (0.008) (0.002) (0.007) Observations 3,193 3,193 3,193 3,193 R-squared 0.299 0.205 0.332 0.125 Distance controls X X X X State FE X X X X Notes: This table reports the results of a regression relating census tract characteristics with a dummy variable whether the centroid of the census tract is within a 25 minutes walking range. The first column reports the results for the log of income, the second column for the share of high-skilled workers, and the third column for the informality rate. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. 9 Table B4: Difference-in-Difference- Share of Informal Workers (1) (2) (3) (4) (5) (6) (7) (8) Outcome: Informal Informal Inf.-non salary Inf.-non salary Informal Informal Inf.-non salary Inf.-non salary Panel A: Continuous Treatment Measure -ln distancei x 1999 0.000 0.002 0.001 0.002 -0.001 0.001 -0.001 -0.000 (0.004) (0.005) (0.004) (0.004) (0.006) (0.006) (0.005) (0.006) -ln distancei x 1999 -0.008 -0.007 -0.012** -0.011** -0.015** -0.016** -0.018*** -0.019*** (0.006) (0.006) (0.005) (0.005) (0.007) (0.007) (0.006) (0.007) -ln distancei x 2004 -0.016*** -0.015** -0.016*** -0.014** -0.019** -0.020** -0.017** -0.017** (0.006) (0.006) (0.006) (0.006) (0.008) (0.008) (0.007) (0.008) Observations 11,504 11,504 11,504 11,504 11,504 11,504 11,504 11,504 R-squared 0.866 0.866 0.844 0.843 0.869 0.869 0.847 0.847 Panel B: Treatment Measure using the dummy variable Ti x 1999 -0.006 -0.010 -0.001 -0.006 -0.005 -0.010 -0.006 -0.013 (0.010) (0.010) (0.009) (0.009) (0.011) (0.011) (0.010) (0.010) Ti x 2004 -0.023* -0.026** -0.023** -0.029** -0.033** -0.036*** -0.028** -0.035*** 10 (0.013) (0.013) (0.011) (0.011) (0.013) (0.013) (0.012) (0.012) Ti x 2009 -0.036*** -0.035** -0.034*** -0.036*** -0.037** -0.036** -0.026* -0.029** (0.013) (0.014) (0.012) (0.013) (0.014) (0.014) (0.014) (0.014) Observations 11,504 11,504 11,504 11,504 11,504 11,504 11,504 11,504 R-squared 0.866 0.866 0.843 0.843 0.869 0.869 0.847 0.847 Mean outcome before the shock 0.582 0.582 0.415 0.415 0.582 0.582 0.415 0.415 Distance Measure Meters Minutes Meters Minutes Meters Minutes Meters Minutes Distance Controls X X X X X X X X State-Time FE X X X X Municipality-Time FE X X X X Notes: This table reports the results of a regression relating changes in the share of informal workers in each location with the line B of the subway. Panel A reports the results for the continuous treatment measures, and panel B for the dummy variables. In the first four columns, I include state-time fixed effects, and in the fifth column to the eighth column municipality-time fixed effects. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. Table B5: Difference-in-Difference- Share of Informal Firms (1) (2) (3) (4) (5) (6) (7) (8) Outcome: Informal Informal Inf.-non salary Inf.-non salary Informal Informal Inf.-non salary Inf.-non salary Panel A: Continuous Treatment Measure -ln distancei x 1999 -0.005** -0.005* -0.004 -0.003 -0.005** -0.004* -0.004 -0.003 (0.002) (0.002) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) -ln distancei x 1999 -0.009*** -0.007** -0.006* -0.004 -0.009*** -0.009*** -0.006** -0.005 (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) -ln distancei x 2004 -0.018*** -0.017*** -0.015*** -0.014*** -0.017*** -0.018*** -0.013*** -0.014*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.004) (0.004) (0.004) Observations 11,504 11,504 11,504 11,504 11,504 11,504 11,504 11,504 R-squared 0.884 0.883 0.906 0.906 0.892 0.892 0.910 0.910 Panel B: Treatment Measure using the dummy variable Ti x 1999 -0.014*** -0.013** -0.007 -0.006 -0.010** -0.009* -0.007 -0.005 (0.005) (0.005) (0.006) (0.006) (0.005) (0.005) (0.006) (0.006) 11 Ti x 2004 -0.022*** -0.020*** -0.015** -0.013** -0.017*** -0.015** -0.014** -0.012* (0.006) (0.006) (0.007) (0.007) (0.006) (0.006) (0.007) (0.007) Ti x 2009 -0.032*** -0.029*** -0.022*** -0.019*** -0.026*** -0.023*** -0.019*** -0.016** (0.006) (0.006) (0.007) (0.007) (0.006) (0.006) (0.007) (0.007) Observations 11,504 11,504 11,504 11,504 11,504 11,504 11,504 11,504 R-squared 0.883 0.883 0.906 0.906 0.892 0.892 0.910 0.910 Mean outcome before the shock 0.833 0.833 0.796 0.796 0.833 0.833 0.796 0.796 Distance Measure Meters Minutes Meters Minutes Meters Minutes Meters Minutes Distance Controls X X X X X X X X State-Time FE X X X X Municipality-Time FE X X X X Notes: This table reports the results of a regression relating changes in the share of informal workers in each location with the line B of the subway. Panel A reports the results for the continuous treatment measures, and panel B for the dummy variables. In the first four columns, I include state-time fixed effects, and in the fifth column to the eight column municipality-time fixed effects. Standard errors are clustered at the census tract level and reported in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. Table B6: Nested Logit - Iceberg Costs (1) (2) Costs: Commuting Trade Trips to Workplace Trips to Shops Minutes -0.009*** -0.013*** (0.000) (0.000) Metro -0.144* -0.596*** (0.079) (0.007) Metrobus -0.556*** -0.855*** (0.117) (0.100) Car -0.220** -0.603*** (0.095) (0.073) Walking -0.352*** 0.338*** (0.106) (0.049) λ public 0.484*** 0.484*** (0.033) (0.009) Observations 56,330 312,015 Trips 11,266 62,403 Iceberg cost before (mean) 4.095 7.217 Iceberg cost after (mean) 3.901 6.893 Notes: This table reports the results of a nested logit using the 2017 OD survey considering only trips that use one transportation mode. The first column reports the results to estimate commuting costs considering only trips from work to home or viceversa between 6am to 10am, and between 5pm to 9pm. The second column reports the results to estimate trade costs using trips to retail shops, outlets, and restaurants. I restrict the sample to trips after 1pm. C Data and Quantification Appendix C.1 Calibration of Speeds This section describes the calibration of speeds across different transportation modes. I use different sources of information. For the transportation network in Mexico City, I use data from the Government of the city.55 For the network of roads, I use information from the New York University digital archive in which they report different types of roads for each census tract in the commuting zone of Mexico City. The different roads include: autopistas, calles, viaductos, etc. I calibrate an average speed for each one of the roads. With this information I compute commuting times across census tracts in Mexico City using the Network analysis toolkit from Arcmap 55 The data can be found here. 12 (Tsivanidis, 2019). I compute these times for four different modes of transportation: walking, car, traditional buses, and the subway. I add five minutes in each station when I compute times for the public transit network, and three minutes when I compute travel times for “car” to capture the time spent in the parking lot. To compute commuting and shopping iceberg costs, I take an average of these times across the different modes. I calculate a matrix across census tracts of approximately 13 million observations. Table C7: Calibration of speeds using trip data from Google Maps Type Speed Panel A: Public transit system Subway Lines 601.24 m/min Metrobus 308.13 m/min Bus 216.67 m/min Walking 90.00 m/min Panel B: Types of roads for cars Autopista 752.03 m/min Avenida 266.84 m/min Boulevard 608.12 m/min Calle 198.56 m/min Callejón 69.643 m/min Calzada 169.98 m/min Carretera 623.38 m/min Cerrada 123.39 m/min Circuito 304.69 m/min Corredor 160.75 m/min Eje vial 273.98 m/min Pasaje 240.71 m/min Periférico 673.43 m/min Viaducto 399.99 m/min Notes: This table reports the calibration of speeds using trips from Google maps. The calibration uses 4,000 random trips. The information was downloaded with the command gmapsdistance in R that uses the Distance Matrix Api from Google. I computed these times between 8 am - 11 am and 5 pm - 8 pm under different traffic scenarios. To calibrate speeds for each mode and each type of road, I use random trips from Google Maps.56 I downloaded 4000 random trips between 8 am-11 am, and 5 pm-8 pm using the command gmaps distance in R that uses the Google Maps Distance Matrix Api. I use as an origin and destination, the closest vertex of each type of road or metro line. This tool has the feature that you can calculate times for different modes under several traffic scenarios: pessimistic, optimistic, or none and modes such as: walking, car, or the public transit network. Using this information, I calibrate speeds for each road and each line using the average time spent to 56 I did not calculate times across census tracts using Google Maps because the network analysis toolkit is much faster, and the command gmaps distance takes a lot of time. 13 move from one vertex to the other. Table C7 reports the average speed for each one of the roads and the public transit system. C.2 Calibration of Fixed Costs Table C8: Estimation of fixed costs (1) (2) (3) (4) Outcome: ln Mis ln Mis ln Mis ln Mis ln Lis 0.715*** 0.879*** 0.642*** 0.568*** (0.014) (0.077) (0.017) (0.054) γi 1.799*** 1.735*** 1.825*** 1.853*** (0.034) (0.036) (0.051) (0.046) ln wis Lis -0.154** 0.070 (0.069) (0.051) γ -1.005*** -0.559*** -0.598*** -0.810*** (0.055) (0.194) (0.091) (0.183) Observations 5,387 5,387 4,374 4,374 Adjusted R-squared 0.851 0.853 0.901 0.902 Implied FI 0.182 0.123 0.117 0.141 Implied FF 1.366 0.875 0.911 1.124 State FE X X Municipality FE X X Notes: This table reports the results of a regression relating the number of firms to the number of workers to recover the parameter β and the fixed costs FI and FF for the informal and formal sector respectively. The unit of observation is a sector-census-tract cell. The dependent variable is the log number of firms in each cell. Columns 1 and 2 include state fixed effects, while column 3 and 4 include census-tract fixed effects to control for the price of commercial floor space qi . Even columns add as a control the wage bill for each sector to control for the price per unit of commercial floor space. Standard errors are clustered at the municipality level and reported in parentheses. ∗p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01. In the model from section 4, in equilibrium, the optimal number of firms is ˜β ˜β Z ˜ −1 σ −1 L Mis = βFs s is is ˜is is the amount of labor and commercial floor-space units employed by location i and sector s, ˜ is and Z where L σs is the elasticity of substitution, and Fs is the entry fixed cost. Taking logs, I estimate the following equation relating the number of firms to the number of workers for both sectors in the baseline year. This estimation allows me to recover the parameters Fs : ln Mis = β ln Lis + (1 − β ) ln Zis − ln σs − ln Fs . (C.1) wis Lis γs qi 14 In some of my specifications, to control for Zis , I include the wage bill for each sector and location with a census-tract fixed effect to capture qi . Then, I estimate the following equation using the Economic Censuses in 1999, the omitted category is the formal sector, ln Mis = γ1 ln Lis + γ2 ln wis Lis + γi + γI + γ + is , (C.2) β 1−β where γi is the census-tract fixed effect, γI is an informal sector dummy variable, and γ is a constant term. ˜ + ln(σF FF ). From the optimal number of firms, we have that γI = ln(σI FI ), and γ = ln β Table C8 reports the results for this estimation for different specifications. I run the previous equation, including state and municipality fixed effects, and in the even columns, I control for the wage bill. I obtained that on average, the value of β ≈ 0.7. I also find that the entry fixed cost for a firm into the informal sector is approximately 0.15, and into the formal sector is 1.1. This means that the fixed cost to enter into the formal sector is more than five times the one to enter into the informal sector. This result is consistent with the fact that the average size in terms of workers of informal firms is lower, but that there are more informal firms in the economy. C.3 Algorithm In this section, I explain the main algorithm to solve for the general equilibrium model. The system of equations is described in section 4. The sub-index t represents simulation. The algorithm is based on Alvarez and Lucas (2007) and it is a contraction mapping. It is as follows: 1. Guess an initial vector of wages w0 , and number of residents in each location L0 . 2. Given a vector wt and Lt , compute the following equations: • Labor supply equations: θs −θs wis dni λnisL|ns = θs −θs (C.3) i wi s dni κ Wns|n θs θs −θs λnsL|n = κ , Wns |n = wis dni (C.4) s Wns |n i ˜ is = L ¯ L. λnis · L (C.5) n • Average income ¯n ≡ y λnis wis (C.6) i,s • Commercial floor space prices ˜ is (1 − β )(1 + tisL )wis L ˜i = qi Z , (C.7) s β (1 + tisZ ) 15 • Number of firms ˜L β ˜β Z˜ 1−β is is Mis = , (C.8) σs Fs • Expenditure shares 1 1−ξ −σ Pns Mis p1 nis 1−σs 1−σs πnis = 1−ξ · 1−σ , with Pns = Mis pnis , (C.9) s Pns i Mi s pni s i πns πnis|s • Government budget constraint ˜is ˜ is + tisZ qi Z tisL wis L i,s δ≡ ¯n Ln ny + qn Z n ¯= α·δ t (C.10) 1 + (1 − α) · δ • Aggregate Expenditure (1 + t¯) Xn = yn Ln + qn Zn ) . (¯ (C.11) ¯ α − t(1 − α) • Labor demand Yis = α πnis Xn . (C.12) n αβYis LDis = t (C.13) wis • Calculate the difference between labor demand and labor supply and the number of residents αβYis − wist (1 + t ˜ isL )Lis zw = (C.14) wist (1 + t ˜ isL )Lis −αη −(1−α)η η ˜t Bn Pn rn Wn ¯L Ln = −αη −(1−α)η η L (C.15) n Bn Pn rn Wn ˜ t ) − (0, Lt )|| < 3. If ||(zw , L tol then, the algorithm stops. Otherwise, update i i t+1 t wis = wis (1 + νw zw ) (C.16a) Lt+1 ˜ t + (1 − νL )Lt , = νL L (C.16b) n n n where νL , and νw are convergence parameter and tol is a tolerance value. 16 C.4 Additional Infrastructure Counterfactuals In this section, I report the results of two additional counterfactuals. In the first counterfactual, I assume that the Government only gives the rebate to formal workers instead of the entire population in the city. In the second counterfactual, I compute travel times without Lines 1 and 2 of the subway and show that the resource misallocation component only explains half of the total gains relative to line B. C.5 Formal workers rebate I now consider what are the welfare gains of line B under the assumption that the Government only gives the rebate to the formal workers. There are two main equations that change from the general equilibrium framework. First the labor supply equation from the formal sector is now given by: ¯)κ κ (1 + t BnF WnF λnF L|n = κ , (C.17) ¯)κ + BnI Wn κ (1 + t BnF WnF I ¯ is given by: and the new Government budget constraint that pin downs the value of t ˜is = t ˜ is + tisZ qi Z tisL wis L ¯L¯· λn λnF L|n λniF L|nF wiF . (C.18) i,s i,n We can solve the system of equations with these two new equations. Table C9 reports the results for the counterfactual of line B. Overall, the results are very similar to the baseline simulations. The welfare gains are between 1.7%-1.8%, and the resource misallocation component explains between 15% to 20%. 17 ˆ = X /X - Line B Table C9: Counterfactual Results X (1) (2) (3) (4) (5) (6) Panel A: Percentage change in welfare-Distortions No Migration Migration %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Panel A: Calibrated wedges Total change 1.00% 0.67% 1.79% 1.01% 0.67% 1.77% Decomposition Pure Effect 87.37% 75.70% 81.98% 83.60% 71.73% 78.32% Allocation 11.28% 14.51% 13.82% 14.94% 18.35% 17.20% Agglomeration 1.34% 9.79% 4.21% 1.45% 9.93% 4.48% Panel B: Percentage change in welfare-Constant wedge %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Total change 0.95% 0.64% 1.71% 0.94% 0.63% 1.65% Decomposition Pure Effect 92.13% 78.17% 86.19% 89.69% 76.38% 84.07% Allocation 7.04% 14.87% 10.94% 9.34% 16.22% 12.70% Agglomeration 0.83% 6.97% 2.88% 0.97% 7.39% 3.23% Notes : This table reports the counterfactual results for Line B of the subway with a rebate only to formal workers. The first and fourth column considers only change in commuting costs, the second and fifth column changes in trade costs, and the third and sixth column considers changes in both type of iceberg costs. The first three columns presents the results for the counterfactual with no migration, and the second three columns for the counterfactual in which I allow for migration in the model. Panel A reports the results for welfare with the calibrated distortions a lá Hsieh & Klenow (2009), and panel B for welfare with a constant wedge in the formal sector based on Levy (2018). The first row describes the results considering the total change. While, the other rows decompose the total change into the different components. The second row shows the percentage explained by the direct effect, the third row by the allocative efficiency margin, and the fourth row by the agglomeration externality component. C.6 Line 1 and 2 This section reports the result of a counterfactual, in which I remove lines 1 and 2 of the subway. These lines have the characteristic of connecting the central areas in the city. Figure C13 plots a map of the city highlighting lines 1 and 2. The interpretation of the counterfactual consists of an allocation without these lines, starting from a world with these lines. 18 Figure C13: Transit System-Lines 1 and 2 (a) Subway Lines Notes: This figure plots a map of Mexico City with the transportation system highlighting the first two lines of the subway: lines 1 and 2. In this counterfactual, I remove these lines to measure the effect on informality and welfare Table C10 reports the main findings. Overall, lines 1 and 2 lead to a real income increase of around 2.7%. However, the allocation component explains a lower fraction of the total gains relative to line B since it only explains 10% of the welfare gains. In contrast, in the case of line B, the reallocation of workers explains more than 20%. Then, Line B generated a larger reallocation of workers from the informal to the formal sector relative to the size of the shock. 19 ˆ = X /X - Lines 1 and 2 Table C10: Counterfactual Results X (1) (2) (3) (4) (5) (6) Panel A: Percentage change in welfare-Distortions No Migration Migration %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Total change 1.57% 1.16% 2.70% 1.56% 1.16% 2.69% Decomposition Pure Effect 96.71% 78.88% 89.27% 96.77% 78.87% 89.30% Allocation 2.93% 10.02% 5.87% 2.86% 9.80% 5.74% Agglomeration 0.36% 11.10% 4.86% 0.37% 11.33% 4.96% Panel B: Percentage change in welfare-Constant wedge %∆ Welfare ∆dni ∆τni ∆dni , ∆τni ∆dni ∆τni ∆dni , ∆τni Total change 1.56% 1.14% 2.67% 1.56% 1.14% 2.67% Decomposition Pure Effect 96.93% 80.50% 90.11% 96.88% 80.38% 90.05% Allocation 2.69% 7.25% 4.57% 2.71% 7.15% 4.54% Agglomeration 0.38% 12.26% 5.32% 0.41% 12.47% 5.41% Notes : This table reports the counterfactual results for Lines 1 and 2 of the subway. The first and fourth column considers only change in commuting costs, the second and fifth column changes in trade costs, and the third and sixth column considers changes in both type of iceberg costs. The first three columns presents the results for the counterfactual with no migration, and the second three columns for the counterfactual in which I allow for migration in the model. Panel A reports the results for welfare with the calibrated distortions a lá Hsieh & Klenow (2009), and panel B for welfare with a constant wedge in the formal sector based on Levy (2018). The first row describes the results considering the total change. While, the other rows decompose the total change into the different components. The second row shows the percentage explained by the direct effect, the third row by the allocative efficiency margin, and the fourth row by the agglomeration externality component. D Theoretical Appendix D.1 Welfare Decomposition In this section, I derive the formula for the welfare decomposition. I start with the perfectly efficient economy and then, I introduce labor wedges. As in the text, there are three group of agents: workers denoted by L, commercial floor space owners denoted by Z , and house owners denoted by H . The two latter groups do not commute. The indirect utility of agent ω is: dni wis niω Vniω = Bn β 1−β (D.1) rn Pn where wis is the wage per efficiency unit in location i, and sector s, and niω is an idiosyncratic shock drawn from a nested Fréchet distribution with dispersion parameters θs , and κ. By the properties of the Fréchet, the total amount of efficiency units d−1 L ˜ nis net of commuting costs provided by location n to location i-sector s is: ni wis d− 1˜ ni Lnis = λn λns|n λnis|ns y ¯ ¯n L, (D.2) 20 1 1 κ κ θs −θs θs ¯n ≡ ( where y s Bns Wns ) , and Wns ≡ i wis dni . From these expressions, the goods market clearing condition is the following system of equations: λn λns|n λnis|ns y ¯ = αβ ¯n L πns πnis|s y ¯ + qn λnZ Z + rn λnH H ¯n λn L (D.3a) n qn λnZ = α(1 − β ) πns πnis|s y ¯ + qn λnZ Z + rn λnH H . ¯n λn L (D.3b) n And the housing market clearing condition is: rn λnH H = (1 − α) y ¯ + qn λnZ Z + rn λnH H ¯n λn L (D.3c) And the average utility in each location is: ¯n ¯ n = Bn y U (D.4) α r −α Pn 1 n D.1.1 Social Planner There is a social planner maximizing welfare such that the market allocation replicates the perfectly efficient allocation. The problem of the planner consists to maximize: ¯ = ωL U ¯n +ωZ δn U ¯nZ +ωH δnZ U ¯nH , δnH U (D.5) n n n ¯L U ¯Z U ¯H U ¯L ωL U where ω and δ are weights that replicate the market allocation.57 As shown later U¯ = αβ . I am interested in a shock to commuting costs or trade costs. Then, by a first-order approximation, the effect of any shock is: ¯ = αβ d ln U ˜ nL (d ln y λ ¯n − αd ln Pn − (1 − α)d ln rn ) (D.6a) n + α(1 − β ) ˜ nZ (d ln qn − αd ln Pn − (1 − α)d ln rn ) λ (D.6b) n + (1 − α) ˜ nH (d ln rn − αd ln Pn − (1 − α)d ln rn ) , λ (D.6c) n ˜ nL ≡ ¯n λn y ˜ nH ) is the share ˜ nZ (λ where λ y ¯n λn is the share of total labor income in location n, and similarly, λ n of total income of commercial floor space (housing) in location n. Then, the change in the average income, and 57 ¯ = To replicate the market allocation, U ¯n Ln + qn Zn + rn Hn y n 21 the price index is: d ln Wn = λns|n λnis|ns d ln wis − λns|n λnis|ns d ln dni (D.7a) i,s i,s d ln Pn = πns πnis|s (βd ln wis + (1 − β )d ln qi ) + πns πnis|s d ln τni . (D.7b) i,s i,s From the goods market clearing condition and with some algebra manipulation then: ¯ = −αβ d ln U ˜ nL λns|n λnis|ns d ln dni λ (D.8a) n,s,i −α ˜ nZ + (1 − α)λ ˜ nL + α(1 − β )λ πns πnis|s αβ λ ˜ nH d ln τni . (D.8b) n,s,i This equation is the Hulten result. When the economy is perfectly efficient, the change in welfare is a weighted average of the change in the fundamentals. In this case, changes in trade and commuting costs. D.1.2 Labor wedge I now assume that firms face distortions. These wedges can be variable markups or taxes. As in HK, I am going to denote these wedges as taxes. In particular, the goods market clearing condition now is: ¯ λn λns|n λnis|ns y ¯ = 1 + t αβ ¯n L πns πnis|s y ¯ + qn λnZ Z + rn λnH H , ¯n λn L (D.9) 1 + tisL n ¯ is a rebate of the Government that can vary by location, or in the case of markups a portfolio where 1 + t that is rebate to households. The previous equation create trade imbalances and wedges across firms. Thus, there is an additional effect in the first-order approximation. This effect captures changes in wages and it is: ¯ = −αβ d ln U ˜ nL λns|n λnis|ns d ln dni λ (D.10a) n,s,i −α ˜ nZ + (1 − α)λ ˜ nL + α(1 − β )λ πns πnis|s αβ λ ˜ nH d ln τni (D.10b) n,s,i tisL − t¯ + αβ ˜ nL λns|n λnis|ns λ d ln wis (D.10c) 1+t ¯ n,i,s ¯). + d ln(1 + t (D.10d) It is easy to show that the change in the rebate is: tisL − t¯ ¯) = d ln(1 + t ˜ nL λns|n λnis|ns λ ˜ nis ). (d ln wis + d ln L 1+t ¯ n,i,s 22 Then, the total change in welfare is: ¯ = −αβ d ln U ˜ nL λns|n λnis|ns d ln dni λ (D.11a) n,s,i −α ˜ nZ + (1 − α)λ ˜ nL + α(1 − β )λ πns πnis|s αβ λ ˜ nH d ln τni (D.11b) n,s,i tisL − t¯ + αβ ˜ nL λns|n λnis|ns λ ˜ nis . d ln L (D.11c) 1+t ¯ n,i,s The third term captures agglomeration forces and it suggests that when workers reallocate to sectors-locations with larger wedges, there is an additional increase in welfare due to an improvement in allocative efficiency. D.1.3 Agglomeration forces Finally, there is an additional effect due to agglomeration forces. In my model this force comes from LOV. This additional effect also captures changes in allocative efficiency and it arises for two reasons. First, if agglomeration externalities differ between the two sectors as in BCDR, or because there are trade imbalances as in FG. In the presence of LOV or agglomeration forces, consumers benefit from lower prices as the sector-location becomes bigger. In particular, there is an additional change in welfare captured by: ¯ = ... + β ˜ nZ + (1 − α)λ ˜ nL + α(1 − β )λ ˜ is ˜ nH d ln λ d ln U πns πnis|s αβ λ 1 + σs n,i,s ¯ = ... + β 1 + tisL ˜ is , d ln U dλ (D.12) 1 + σs 1+t ¯ n,i,s ˜ is is the labor share in total income from sector s and location i. This additional term captures two where λ things. First, if workers move to sectors-location in which agglomeration externalities are larger, then there is an increase in total welfare, and second, if workers reallocate to sectors-location with larger wedges the effect of any shock on welfare is larger. 23 Combining the previous expressions, then, ¯ = − αβ d ln U ˜ nL λns|n λnis|ns · d ln dni λ (D.13a) n,i,s “Pure” effect commuting costs −α ˜ nZ + (1 − α)λ ˜ nL + α(1 − β )λ πns πnis|s αβ λ ˜ nH d ln τni (D.13b) n,i,s “Pure” effect trade costs   tisL − t¯ + αβ  ˜ nL λns|n λnis|ns λ ˜ nis  d ln L (D.13c) 1+t ¯ n,i,s Allocative efficiency 1 1 + tisg ˜ isg , + βg dλ (D.13d) σ −1 1+t ¯ n,i,s s g Agglomeration Forces which is the same expression as in the text. D.2 The problem of the social planner In this section, I find the equilibrium conditions for the problem of the social planner. I show two results. First, in the case in which the economy operates under perfect competition, the market allocation coincides with the ¯ is equal to the aggregate total expenditure or income of the economy, efficient allocation. Second, the variable U which is the main assumption from the previous section. There are different groups of workers indexed by g , sectors indexed by s and a mass of locations N indexed by n and i. Each group has a utility function Ug (cng , hng ), where cng represents the average consumption of a composite good in location n and hng is the average amount of housing in location n. This utility function is homogeneous of degree one. In the optimal allocation, workers are indifferent across locations and there are iceberg trade and commuting costs. The problem of the planner is to maximize the following welfare function: ¯ = λg · Ug U subject to i) spatial mobility constraints ¯g ∀n, g Ung Lng ≤ U ii) composite and housing feasibility constraints τni Qis ≤ Yis (Egis ) ∀i, s n,s Lng · cng ≤ C (Qn11g , ..., QnSN g ) ∀n ˜nhg ) ∀n Lng · hng ≤ Hn (E 24 iii) labor supply constraints ˜isg ≤ E d− 1 ni Enisg ∀i, s, g including the sector that produces housing n Eg (En11g , ..., EnSN g ) ≤ Ln,g ∀n, g iv) non-negativity constraints of commuting flows, trade flows, labor. v) Labor Market clearing ¯g Ln,g ≤ L n,g where Y is the production function, C (·) is a composite good aggregator across locations and sectors, in my case the nested CES; E (·) is a efficiency units aggregator, in my case the nested Fréchet; and Enisg are efficiency units provided from location n to i, s by group g .58 The other parameters represent the same variables as in section 4. The Lagrangian of the planning problem omitting the non-negative constraints is: L = Lg Ug − ωng Lng (Ug − Ug (cng , hng )) n,g − p∗ is ˜isg ) τni Qnis − Yis (E i,s n ∗ − Pn Lng cng − C (Q(Qn11g , ..., QnSN g )) n g − ∗ wisg ˜isg − E d− 1 ni Enisg i,s,g n ∗ − y ¯n,g (Eg (En11g , ..., EnSN g ) − Ln,g ) n,g − ∗ rn ˜nhg ) Lng hng − Hn (E n g − Ψg ( ¯ g ) + ... Lng − L g n ˜ihg , Lng , and Ug to maximize welfare. I proceed in two ˜isg , E The planner chooses cng , hng , Qnis , Enisg , E parts. First, I show the relationship between U ¯ and aggregate expenditure, and then, I show that the market allocation coincides with the efficient allocation. Then, I generalized the formula from the previous section using the goods market clearing condition. ξ −1 ξ−1 σs −1 σs −1 58 σs σs Recall that the CES aggregator from section 4 is Cn ξ ≡ s ξ Qns , where Qns ≡ i Qnis and the efficiency units κ κ θs θs κ−1 κ−1 θs − 1 θs − 1 aggregator is En ≡ s Ens , where Ens = i Enis . 25 Utility and Total Expenditure The F.O.C with respect to cng and hng is: ∂Ug ∗ ωng cng ≤ Pn cng ∀g ∂c ∂Ug ∗ ωng hng ≤ rn hng ∀g ∂h Since Ug (·) is homogeneous of degree one, then, ∗ ∗ Lng (Pn cng + rn hng ) = Lng ωng Ung (D.14) The LHS of equation D.14 is the aggregate expenditure Xng of group g who lives in location n. The F.O.C whit respect to Ug is: ωng Lng = Lg n Combining this equation with equation D.14, and the fact that in equilibrium Ung = Ug for all the locations in which Lng > 0 yield that: Lg Ug = Xng n ¯= Recall that U Lg Ug , thus, g ¯ =X U where X ≡ Xng is aggregate expenditure. At the aggregate level, total expenditure is equal to total income n,g ¯= then in the previous section U ˜n + rn Hn , which was the assumption for the theoretical result ¯n Ln + qn Z ny of the first-order approximation. Efficient Allocation Now, I show that the market allocation coincides with the efficient allocation. The F.O.C with respect to other variables is: ∗ ∂C [Qnis ] :Pn ≤ p∗ is τni (D.15a) ∂Qnis ˜isg ] :p∗ ∂Y ∗ [E is ˜ ≤ wisg (D.15b) ∂ Eisg ∂Eg [Enisg ] :wisg∗ d− 1 ni ≤ y ¯ng (D.15c) ∂Enisg ˜nhg ] :r∗ ∂H [E n ˜ ≤ wnhg (D.15d) ∂ Enhg y n ≤ Ψg [Lng ] :¯ (D.15e) 26 Equations D.15a to D.15d are the same as the utility and profit maximization conditions of the consumer’s and firm’s problem. In the particular case in which the function C (·) is the nested CES utility function from section 4, E (·) is the nested Frechet, and assuming that Y (·) is homogeneous of degree one, I can rewrite these conditions as: λnsg|n λnisg|ns y ∗ ¯ng ∗ Lng = wisg d− 1 ni Enisg ∗ ˜ wisg Eisg = βisg p∗ is Yis p∗ is Yis = αng πnis y ∗ ¯ng Lng n,g ∗ ∗ rn Hn = (1 − αng )¯ yng Lng , n,g where ∂Y Eisg ∂E isg βisg ≡ Yis ∂U g cng ∂cng αng ≡ Ung These are the same conditions as the market allocation from the previous section. Then, the market allocation is efficient in the case in which there are no wedges. We can generalize the welfare decomposition from the previous section for different groups of labor under the assumptions where the utility and production function is homogeneous of degree one. In particular, we can ¯ as: rewrite the change in U ¯ =− d ln U ˜ ng λnsg|n λnisg|ns · d ln dni αng βisg λ (D.16a) n,i,s,g “Pure” effect commuting costs − ˜ ng d ln τni πns πnis|s αng βisg λ (D.16b) n,i,s,g “Pure” effect trade costs   ¯ tisg − t + ˜ ng λnsg|n λnisg|ns αng βisg λ ˜ nisg  d ln L (D.16c) 1+t ¯ n,i,s,g Allocative efficiency βg 1 + tisg ˜ isg . + dλ (D.16d) σs − 1 1+t ¯ n,i,s,g Agglomeration Forces This result is similar to the one obtained by Baqaee and Farhi (2020) in GE models. However, this expression is in the context of an urban model in which firms face iceberg trade costs and workers face i) commuting costs, and ii) are indifferent to live across locations within the city. 27 D.3 Equilibrium Conditions - Exact Hat Algebra In this section, I solve for the equilibrium conditions and change in total welfare using exact hat algebra as in Dekle et al. (2008). I define the percentage change of a variable as: x ˆ= x x then, the change in the average utility is 1 η ˆ ¯= U ˆn λn P −αη −(1−α)η ˆ η ˆn r Wn , (D.17) n η −αη −(1−α)η Wn Pn rn where λn ≡ −η η is the share of residents in location n in the pre-period. The change in the price n Pn Wn and wage indices is given by the following expressions: 1 1−σs ˆns = −σs P ˆ1 πni|s p is (D.18a) i 1 1−ξ ˆn = 1−ξ ˆns P πns · P (D.18b) i 1 θs ˆ ns = W λnis|ns · w ˆis d− θs ˆ ni θs (D.18c) i 1 κ ˆn = W ˆ ns λns|n · W κ . (D.18d) s The change in the residence, sector, and workplace choice probability is: ˆn−η ˆ η ˆn−η ˆ η ˆn = P Wn P Wn λ − η η = (D.19a) ˆ ˆ ˆ ¯η n λn · Pn Wn U Wˆ nsκ ˆ ns W κ ˆ ns|n = λ = (D.19b) ˆκ ˆn k λnk|n · Wnk W ˆ nis|ns = w θs ˆ ˆis d−ni θs w θs ˆ ˆis d−ni θs λ θs ˆ = . (D.19c) λ nls | ns · wˆ d θs Wˆ ns θs l l nl And the change in the expenditure shares is: 1−ξ ˆns 1−ξ ˆns P P ˆns = π = 1−ξ (D.20a) ˆ 1−ξ ˆn k πnk Pnk P 1−σs 1−σs ˆnis p ˆnis p ˆni|s = π −σs = . (D.20b) ˆ1 l πnls|s pnls 1−σs ˆns P 28 The change in the average labor income and aggregate expenditure is: ˆ ¯n = y λY ˆ ˆ nis λnis|ns λns|n wˆis , (D.21) i,s λnis|ns λns|n wis where λY nis ≡ y ¯n . Then, the change in Xn is: X ˆ ¯t ˆ n = (1 + ˆ ¯n λ ) ωnL y ˆ n + ωnZ q ˆn , ˆn + ωnH r (D.22) ¯n Ln y qn Zn rn Hn where ωnL ≡ ¯n Ln +qn Zn +rn Hn , y ωnZ ≡ ¯n Ln +qn Zn +rn Hn , y and ωnH ≡ ¯n Ln +qn Zn +rn Hn . y Then, the goods market clearing condition using exact hat algebra for each location i and sector s is: w ˆis ˜ ni λ ˆ nis|ns λ λ ˆ ns|n λ ˆn = X πnis π ˆns π ˆn, ˆnis|s X (D.23) n n ˜ ni = λnis X = πns πnis|s Xn where λ λn , and πnis πn s πn is|s Xn . I compute the counterfactuals solving the previous system n is n of equations D.23. D.4 Model with ex-ante firm decision In this section, I present a version of the model in which firms decide whether to operate in the formal or in the informal sector. The model is based on Ulyssea (2018) and Dix Carneiro et al. (2018) in which firms that operate in the informal sector face a distortion that increases with size. There is a infinite mass of potential entrants that exit at an exogenous rate δs . The labor supply function takes the same form as in the main text. On the other hand, firms make two decisions. First, they decide whether to enter in the labor market and conditional on entry whether to operate in the formal or informal sector based on a pre-entry signal and a entry fixed cost. Second, firms decide the location in the city in which they are going to operate based on an extreme value type II shock. There is no production fixed cost. The total operational profits of firm ω that operate in location i and sector s, and sells to n is given by: 1−σs op 1 τni (wis [1 + tisL ])β (qi [1 + tisZ ])1−β σs −1 ˜ πnis (ω ) = Pn Xns , σs − 1 z (ω ) is (ω ) op op πis (ω ) = (1 − υis (ris (ω ))) πnis (ω ), n where z (ω ) is the pre-entry signal, is (ω ) is an idiosyncratic productivity shock of firm ω that varies across locations, and υis (ris (ω )) is a distortion that captures the probability of getting caught if firm ω operates in the informal sector. This probability increases with the size and revenue r(ω ) of firm ω . I assume that the idiosyncratic shocks are drawn from a Frechét distribution with shape parameter ψ and scale parameters Ais . Then, the share of firms with pre-entry signal z from sector s that operate in location i is: ψ Ais πi s(ω ) µis (z ) = ψ . (D.24) l Als πl s(ω ) With these assumptions, the expected value of entry for a firm with pre-entry signal z that operates in sector 29 s is:  1 ψ 1−σs ψ σ s −1 β 1−β z τni (wis [1 + tisL ]) (qi [1 + tisZ ]) σ s −1 ˜ Vse (z, wis ) =  (1 − υis (ris ))Ais Pn Xns  . (D.25) δs i n z (ω ) A firm decides to enter and operate in sector k if the following condition holds: e Vke (z, w) − Ek ≥ max{V− k (z, w ) − E−k , 0}. (D.26) Because the average expected profits increase with size, and the distortion also increases with size, there are two cutoffs of the pre-signal productivity z that determine the entry to market and whether a firm operates in the informal or the formal sector. Let’s define the entry cutoff as zE , and the informality cutoff as zI . Then, the labor demand in location i and the informal and formal sector are given by: −1 zI βwi I LiI = µiI (z )riI(z ) dF (z ) (D.27) F ( zI ) − F ( zE ) zE −1 ∞ βwiF LiF = µiF (z )riF (z ) dF (z ), (D.28) 1 − F ( zI ) zI where the variable ris (z ) represent the average revenue of a firm with presignal z . The labor supply function takes the same form as in the main text, and the equilibrium is determined by equalizing the labor demand and labor supply. Similarly, to solve for the commercial floor space equilibrium, the demand function is given by: zI ∞ ˜ D = q −1 (1 − β ) 1 1 Z i i µiI (z )riI(z ) dF (z ) + µiF (z )riF (z ) dF (z ) . (D.29) F (zI ) − F (zE ) zE 1 − F (zI ) zI On the other hand, the equilibrium equations for the residential floor space are the same as in the main text. Following the logic from Ulyssea (2018) and Dix Carneiro et al. (2018) this equilibrium exists. 30