Policy Research Working Paper 10802 The Welfare and Productivity Effects of Transit Improvements in Amman Tatjana Kleineberg Sally Murray, Yulu Tang Jon Kher Kaw Urban, Disaster Risk Management, Resilience and Land Global Practice & Development Research Group June 2024 Policy Research Working Paper 10802 Abstract This paper studies the long-run welfare and productivity they also lived farthest from the city center before the bus effects of transit improvements in the Greater Amman rapid transit was established. Welfare in all neighborhoods Municipality. The paper builds a rich quantitative spatial increases, with the largest increases at the outer ends of the model that includes many aspects of the economic geog- new bus rapid transit lines. Phase 1 generally promotes raphy of Amman. It studies the effects of new bus rapid densification and welfare in already dense locations, while transit lines that improve the connection of more peripheral phase 2 encourages additional densification to the south. areas to the city center, in two phases: phase 1 (approxi- The preliminary analysis of the interaction of zoning restric- mately) connecting the north-eastern and north-western tions with the bus rapid transit suggests that legal zoning regions, and phase 2 adding the southern and south-west- limits are binding in a few locations where excess demand erns regions. It finds that the bus rapid transit increases for real estate after the expansion of bus rapid transit is output by 4.4 to 5 percent in phase 1 and 7.2 to 7.6 percent expected to be large. in phase 2. Workers in manufacturing benefit the most, and This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Practice and 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 authors may be contacted at smurray2@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 The Welfare and Productivity Effects of Transit Improvements in Amman* Tatjana Kleineberg Sally Murray Yulu Tang World Bank World Bank Harvard University Jon Kher Kaw World Bank JEL Classification: R13; R32; R40; R52 Keywords: Urban Transport; Urban Land; Urban Productivity; Urban Jobs; Amman; Jor- dan; Spatial Computable General Equilibrium Model * We would like to thank Hogeun Park, Myriam Ababsa, and the World Bank transportation team for data and their expert reviews and comments. The findings, interpretations, and conclusions expressed in this pa- per 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 repre- sent.Contact: Development Research Group, World Bank, 1818 H Street NW, Washington, DC 20433. Email: smurray2@worldbank.org; tkleineberg@worldbank.org. 1 Introduction In developing countries, urbanization is a key driver of economic productivity growth. Through their scale, density, and connectivity, cities offer economies of scale and agglomeration, and facilitate knowledge spillovers, which together drive efficiency gains in production and ser- vices. To create thriving cities, urban planning, infrastructure, and service provision must enhance connectivity, through productive density and strong transportation, while offering high living standards to workers. Transport system must provide workers access to the most productive jobs and places within cities. This paper analyzes the welfare and productivity effects of transit infrastructure improve- ments in Amman, the capital of Jordan. We collect rich microdata at a very granular level to characterize the city structure of Am- man. We observe the neighborhoods in which people live and work and their sectors of employment. We further observe locations’ residential and commercial floor space and the legal zoning restrictions. Data on the transportation network and travel speeds allows us to calculate the travel times between all neighborhood pairs of the city. We use this data to calibrate a quantitative urban model in which workers make residential, commuting, and employment decisions (including where to live and work, what sectors to work in, and how to commute), and developers make decisions about housing supply, following Ahlfeldt, Redding, and Sturm (2016), Redding and Rossi-Hansberg (2017) and others. Structural parameters and elasticities are calibrated to values from the literature. Based on these parameters and the data described above, the model derives baseline amenities, productivities, wages, and rents for each neighborhood. After thereby describing the city at baseline, we then use the estimated model to simulate counterfactual policy scenarios. In particular, we analyze how investments in Bus Rapid Transit (BRT) systems affect aggregate and distributional outcomes. The spatial model sim- 2 ulates effects of the BRT investment on aggregate output and welfare, sectoral output, and on the distribution of population and economic activity across neighborhoods within Amman. The model further analyzes the extent to which Amman’s zoning restrictions can limit the development of new residential or commercial real estate which can lower potential produc- tivity gains. Despite the richness of the model, there are some caveats that should be kept in mind. Es- pecially, we cannot measure or distinguish informal employment and unemployment in the data, so we focus our analysis only on formal employment. The total employment and total population of Amman are fixed in our model; this means we cannot capture any impact that the BRT may have on drawing workers or residents into the city, or inactive residents into the labor force. It also does not capture exogenous population, income, and output growth over time. We keep amenities (i.e. neighborhood attractiveness) fixed in our policy coun- terfactuals, so we do not capture potential responses in the attractiveness of neighborhood characteristics due to changes in the population composition or density. It is, for example, possible that neighborhood become less attractive as they become denser or alternatively the might become more attractive through additional resources or investments as more educated and higher-income residents move in. The remainder of the paper is organized as follows. Section 2 describes the context of the Greater Amman Municipality and the BRT expansion. Sections 3 and 4 describe the model and data, respectively. Section 5 describes the calibration. Section 6 presents the results of our policy counterfactuals, and Section 7 concludes and suggests directions for future research. 3 2 Context and Transit Infrastructure in Amman Jordan is an upper middle-income economy, with GDP per capita of USD4,400 in 2021. Its capital and economic hub, Amman, is a primate city, hosting 42 percent of the national population (3.7 million people in the latest 2015 census). Amman has been continually inhabited since the 8th century BC, but its population has grown exponentially in the past century, from just 5,000 in 1906 to 1.8 million by 1999, and over 4 million today (Ababsa and Ahmad (2020)). This growth was driven in part by large successive waves of refugees from fragile regional neighbors, most recently from the war in the Syrian Arab Republic. Its role as a safe harbor has spurred Amman to become an important regional hub for bank- ing, business, and tourism, but has also put urban services and labor markets under intense strain, as populations have soared almost overnight during waves of refugees crises. Most recently, at the height of the Syrian refugee crisis, water demand rose by 40 percent and unemployment by 30 percent (100 Resilient Cities, 2017). A hilly topography, population pressures, and insufficient investment in public transport have exacerbated traffic congestion in Amman in recent years. The city is built around nine- teen hills, forcing a reliance on bridges or tunnels to connect many major city roads. The density of buildings, population, and cultural heritage sites has also limited the scope for road widening as the population has boomed. In response, the Greater Amman Municipality (GAM) began constructing a Bus Rapid Transit (BRT) system in 2015, to improve the effi- ciency of the road network. The first BRT routes were opened in 2021 and 2022, with a third (connecting Amman to Zarqa) expected to open in 2023. 4 3 Model Outline The spatial model that we use to study the effects of the BRT investments is based on recent work in the urban literature such as Ahlfeldt, Redding, and Sturm (2016); Redding, Sturm, and Heblich (2018); Tsivanidis (2019). The full model is specified in Appendix B. Here, we describe the model set up, workers’, residents’, firms’, and developers’ choices, market clearing and the equilibrium conditions of the model. Production. We consider three economic sectors: manufacturing, tradable services, and non-tradable services. Tradable services are defined as typically higher-end services that can be traded across neighborhoods within the city (such as finance, marketing, or accounting), while non-tradable services are those that require face-to-face interactions (such as grocery stores, coffee shops, and hair dressers). Sectoral prices for goods and services are assumed to be the same across the city, reflecting the assumption that all goods and services (but not workers) can be freely traded across Amman’s neighborhoods. Within each sector and each city neighborhood, perfectly competitive firms use labor (efficiency units) and commercial floor space as inputs, to produce a homogeneous output good. Due to perfect competition, firms make zero profit in each location and sector, so that their revenues equal total wages paid to workers plus total rent paid for commercial floor space. Geography. We model 152 neighborhoods in the Greater Amman Municipality. Each neighborhood is characterized by residential amenities (which make locations more or less desirable for residents), sectoral productivities, and residential and commercial housing sup- ply (floor space). Each neighborhood pair is separated by a bilateral commuting time, which depend on the city’s road and public transit infrastructure. The new BRT investments im- prove these commuting times. 5 Individuals’ Choices. Average welfare in the model depends on residential amenities (factors affecting the attractiveness of the neighborhood in which each person lives) and real incomes. Individuals choose a location in which to live and a location and sector in which to work, optimizing their utility which depends on residential amenity and real income. These decisions define workers’ commuting routes between their residence and work loca- tions. Each worker’s income depends on the efficiency units that she or he supplies to her or his job and on the wage rate per efficiency unit that is offered by her or his sector and work location. Workers’ efficiency in each job depends on their location-and-sector choice, and is reduced by commuting times, since workers can work longer times when spending less time in transit. Workers spend their income on consumption goods from all three sectors and on residential housing, which they rent at local unit-prices in their chosen residence location. Developers’ Choices. Residential and commercial floor space is characterized by a fixed and a variable component. The fixed component captures limits in land availability. The variable component allows floor space to increase in response to rents, capturing the fact that real estate developments can increase the housing supply if local housing demand and rents increase. The housing supply elasticity captures how strongly housing supply responds to changes in local rents. We include zoning restrictions in our model, which provide a legal upper limit to residential and commercial housing supply in each neighborhood. Market Clearing and Equilibrium. The model is characterized by a set of structural parameters and regional characteristics. The structural parameters describe individuals’ preferences, the responsiveness of individual choices to neighborhood or job characteristics (elasticities), the production technology, and the housing supply elasticity. The regional char- acteristics are residential amenities, sectoral productivity, the fixed component of residential and commercial floor space, and commuting costs. Conditional on this set of parameters and neighborhood characteristics, the equilibrium of 6 the model is defined by sectoral prices, local rental rates for commercial and residential housing, and location- and sector-specific wage rates, which ensure that: 1. Workers maximize their utility by optimally choosing their consumption, the location in which they want to live, and the location and sector in which they want to work. 2. Firms in each location and each sector maximize their profit by making optimal input choices over labor and commercial floor space, making zero profits in equilibrium due to perfect competition. 3. Good markets clear in each sector, equating firms’ production with workers’ consump- tion. 4. Labor markets clear in each location and each sector, equating workers’ labor supply with firms’ labor demand. 5. Commercial and residential floor space markets clear in each location, equating each respective housing supply with firms’ demand for commercial floor space and with consumers’ demand for residential floor space. 4 Data This section describes the data and patterns of the spatial distribution of residents and em- ployment in the Greater Amman Municipality. Our analysis is at the neighborhood level and covers 152 neighborhoods and 20 mantiqas. We analyze economic activity in three sec- tors: manufacturing, non-tradable services, and tradable services. We gather the following data at the neighborhood level: total employment and residents by sector, commuting times between all neighborhoods, total floor area for residential and commercial use, and zoning restrictions (i.e., maximum allowed floor space area). Appendix A provides more information on the data construction and data sources. 7 Population. We retrieve population data for each neighborhood from the Housing and Population Census (2015). Figure 1 describes the distribution of population across neigh- borhoods and the relationship between population density and neighborhoods’ distance from the center. The first panel shows a scatter plot relating population density and distance. The plot shows a negative relationship, suggesting that people concentrate in the center of the city. The second plot confirms this finding by grouping neighborhoods into deciles ac- cording to their distance to the center. The third plot is a map which shows the population density for each neighborhood. The map show that some of the neighborhoods in the out- skirts also have high population density, suggesting that Amman has some resemblance to a poly-centric city. Employment. We use employment data by sector at the matiqa level from the Establish- ment Census (2018). We impute this data to the neighborhood level based on the within- matiqa distribution of building volume. Figure 2 shows the distribution of employment across neighborhoods and the relationship between employment density and neighborhoods’ distance from the center with three plots that are analogous to Figure 1. We see that the highest employment density is in the center of the city, with a particular concentration in the center-west. Figure 3 shows the distribution of the employment density separately for each sector: we see that jobs in tradable services are most spatially concentrated, with a clear cluster in the center-west of the city. Non-tradable service jobs are somewhat more dispersed: they have high density in the same areas (center/center-west), but they additionally have higher density in the southern and eastern parts of the center. Manufacturing is less centralized, with distinct clusters in the east, south, and west. Housing Volume and Zoning Restriction. We collect data on total built floor space in each neighborhood from satellite data from the European Space Agency. In addition, we 8 obtain shapefiles with land use typologies from the municipal government, which allows us to approximately classify floor space in each location as either commercial or residential. We further obtain the Amman zoning restriction rules, which assign to each building type the percentage of the plot that can be used for construction and the maximum number of floors that can be constructed (among other constraints not modeled here). We merge the land use typology of each plot to the relevant building categories from the zoning rules to calculate the maximum floor space that can legally be developed in each location. We perform this step separately for commercial and residential building types. Travel Times with Road Travel. To compute average travel times between all locations, we use the road network for Amman and assign average travel speeds to different types of road segments based on estimates from the World Bank Transportation Unit. Specifically, we assign a speed of 35 km/hour for primary roads, 20 km/h for secondary roads, 10 km/h for tertiary roads, 8 km/h for other road types, and 5 km/h for walking. For metro lines we assign a speed of 35 km/h and for monorail 45 km/h. In the counterfactual, we compute commuting times with the new BRT routes to which we assign a traveling speed of 25 km/h. Figure 4 shows average commuting times for each residency neighborhood to all possible work locations where we weigh travel times of each work destination by the number of jobs available in this destination. Aligned with the previous results on employment density, we see that central neighborhoods are best connected to jobs while the outskirts have larger average commuting costs to available employment opportunities. Commuting times are par- ticularly large for the south and some of the most western and northern neighborhoods. 5 Calibration Results 9 5.1 Calibration of the Model Parameters and Regional Characteristics In our model calibration, we use structural parameters from the literature as shown in Appendix B. We then use these parameters, the data moments described in the previous sec- tion, and the equilibrium conditions from our model to infer regions’ characteristics, which include economic productivity (by sector) and residential welfare. The calibration procedure solves for locations’ sectoral wage rates, rental rates for commercial and residential floor space units, and prices for goods from each sector. These wages, rents, and prices are en- dogenously determined and they ensure that all markets clear in equilibrium. Appendix B provides more details on the calibration procedure. We now present the distribution of our regional characteristics. Productivity. Figure 5 shows the distribution of productivity across neighborhoods for each sector. Consistent with the employment density by sector, the productivity of manu- facturing is higher in the southeastern part of Amman (where most industries are located), while the non-tradable and the tradable sectors exhibit higher productivity in the north- western area. Wage Rates. Figure 6 shows that average wage rates are highest in the center-west, where wages are two or three times higher than the average wages. Amenities. Amenities are a residual capturing how attractive neighborhoods are to resi- dents due to unexplained factors. Welfare. Figure 7 shows that welfare is highest for residents in the western part of Am- man. (West Amman is an affluent part of the city, where international companies, banks, 10 private schools and main Embassies are located.) In this figure, we estimate workers’ wel- fare from their indirect utility, which includes neighborhoods’ residential amenities, average real income, and average commuting costs for residents.1 6 Policy Counterfactuals 6.1 BRT Investments Improve Transit Times and Job Accessibility We simulate two phases of BRT investments, which are shown in Figure 8. We first analyze the effects of Phase 1 of the BRT investment, which connects the eastern and western parts of the city to the center (yellow and orange lines on Figure 8). We refer to this policy as "BRT Phase 1". The yellow line was completed and opened in 2022, and the orange line is scheduled for completion in 2024. We then analyze the effects of the full planned BRT investment, which includes the former lines and adds the BRT lines to the south and south- west (pink lines on Figure 8). We refer to this full BRT investment as "BRT Phase 2". These lines are expected to be completed and opened between 2024 and 2027. Figure 9 shows how each BRT phase affects job accessibility from each neighborhood. In particular, we calculate the (new) average commuting time from each residence neighbor- hood to all possible work locations, weighing each travel time by the number of jobs in the work location. For this calculation, we estimate travel speed of the BRT line to be 25 km/h, accounting for speeds as well as stoppage time.2 Figure 9 shows substantial travel time re- ductions due to the BRT. BRT Phase 1 primarily improves the travel times from the eastern and western parts, with decreases in average travel times by up to 22 percent. BRT Phase w js d − 1ϵ 1 i j i js More specifically, we average the indirect utility across all residents in each neighborhood: Vi js = −α .) P α r1 i 2 This compares to estimates of road speeds that vary between 8km/h for unfinished roads and 35 km/h for primary roads. We maintain these road speed estimates in our counterfactual with the BRT investments, which ignores additional potential travel time improvements if the BRT’s would alleviate road congestion. 11 2 additionally improves travel times from the south into the center, decreasing travel times by up to 26 percent. 6.2 Aggregate Effects of the BRT on Productivity and Output Since the analysis holds total population and employment fixed, it does not capture potential gains from workers who respond to the BRT by entering formal employment in Amman from unemployment, from informal work, from outside the labor market (inactive), or from regions that are outside of our model area. We focus on two key channels through which better commuting times increase economic output and workers’ productivity: first, workers can work longer hours since they spend less time in transit which increases workers’ effective work input. Second, better connectivity between employment locations and residences makes it easier to match workers to jobs in which they are most productive even if those jobs are far away from workers’ residences. This allows locations to become more "specialized" using their comparative advantage as residence or employment locations or as employment hubs for specific sectors. Table 1 shows that the BRT investments have large effects on sectoral output. Phase 1 of the BRT increases output by 4.4-5 percent in all sectors, while Phase 2 increases output by 7.2-7.6 percent in all sectors. Overall, output gains are relatively similar in all sectors. Similarly, we find very little reallocation of workers across sectors. These findings imply that the BRT supports output growth homogeneously across sectors and does not prompt a significant reallocation of workers across sectors. 6.3 Distributional Effects of the BRT across Neighborhoods Distribution of Employment. We next discuss how each phase of the BRT roll-out affects the distribution of jobs and economic activity across neighborhoods. 12 Table 1: Large Aggregate Effects of BRT on Sectoral Output By sector Phase 1 of BRT Phase 2 of BRT Manufacturing 4.37 % 7.60 % Non-tradable services 4.35 % 7.24 % Tradable services 4.96 % 7.61 % Under this assumption, Figure 10 shows how employment density would change under BRT Phase 1 and Phase 2. We see large distributional effects with employment density decreasing in some neighborhoods by up to 11 percent and increasing in others by up to 31 percent. BRT Phase 1 increases employment density in the northeastern and northwestern parts of the center where density was already relatively high before the BRT investments. At the same time, employment density would decrease in the southern and northern periphery where density is relatively low before the BRT investments. This finding suggests that BRT Phase 1 would make Amman more mono-centric and spatial inequality in the city would increase. BRT Phase 2 would additionally increase employment density in the south and would decrease it in the eastern and northern periphery. Additionally, figure 11 and figure 12 shows how employment density and wage change by sector under BRT Phase 1 and Phase 2. We can see that workers in manufacturer sector benefits the most from the BRT, who also live the farthest away from the city center before the BRT investment. Distribution of Population, Rent, and Welfare. Figure 13 shows the effects of the BRT on population density, rents, and welfare. In Phase 1, welfare increases in all locations, with the largest gains in the north, northwest, and northeast, where welfare increases up to 16 percent. Welfare increases in locations that are directly connected to the BRT line due to shorter commuting times and productivity/wage increases; some locations that are not directly connected to the BRT line also benefit from reductions in long commuting distances and can further gain from spillovers such as lower rents. BRT Phase 1 increases population density and rents most in the north and center, while it decreases in the south and center- 13 west. Under BRT Phase 2, welfare again increases in all regions, with largest gains (up to 16 percent) in the south and northwest. BRT Phase 2 increases population density and rent most in the south and north-west, with increases up to 14 percent for population density and up to 18 percent for rent. 6.4 The Impact of Zoning Restrictions on Outcomes of the BRT Zoning restrictions can limit economic development and agglomeration forces in highly- productive locations. Our analysis in Figure 14 shows that (holding population and em- ployment constant) zoning restrictions may bind in several locations located mostly in the center west of Amman. Before the BRT investments, zoning restrictions bind in 9 neighbor- hoods for residential floor space and in one neighborhoods additionally for commercial floor space. With the opening of the BRT lines (Phases 1 and 2), our model predicts that the de- mand for commercial floor space would further increase, making zoning restrictions binding in a total of 13 neighborhoods among which commercial and residential restrictions would simultaneously bind in 10 or 11 neighborhoods (cf. Figure 14, (a) and (b) respectively). The neighborhoods in northwestern Amman in which zoning restrictions are binding (high- lighted in Figure 14), fall around a BRT interchange site in Sweileh. The neighborhoods are characterized by a mix of formal and informal mid-rise structures, with both main high- ways and more narrow winding streets, and medium levels of planning and modernization. The BRT terminal building itself, and the surrounding areas, are underused and under- leveraged. The analysis suggests this neighborhood will receive considerable new demand as an employment and residential location in response to the BRT, and has considerable potential to create new economic opportunities, with strong potential as a site for urban regeneration, densification, and land value capture through intentional transit-oriented de- 14 velopment. However, realizing this potential may require a revision of zoning rules that prevent the neighborhood from adjusting to its new opportunities. The neighborhoods with binding zoning restrictions to the South-East of Sweileh in Figure 14 are currently wealthy, planned, predominantly residential, neighborhoods near central Amman. These neighborhoods also host some substantial commercial centers, such as shop- ping malls, international hotels, the new Central Business District, and Sport City. The analysis suggests that many of these neighborhoods are mainly zoned for low-density res- idential development: Residential zones A and B, requiring lot sizes of at least 750-1,000 square meters, only four floors above street level, and maximum lot coverage of 39-45 per- cent. There is expected to be demand for densification, and for more productive and commer- cial activities, in these neighborhoods due to the BRT, above the levels allowed by current zoning regulations. Low-density residential zoning may prevent this productive evolution of the neighborhood, dampening the more transformative job- and wealth- creating bene- fits that the BRT may bring. The city may consider allowing for additional stories, lower setbacks, and more mixed-use zoning in these neighborhoods, to allow a more productive adaptation to these new opportunities. Development of these neighborhoods could create opportunities for lower-income residents in other areas of Amman, such as the East and South, if zoning restrictions are addressed and transit-oriented development encouraged. After Phase 1 of the BRT, the discussed center- west neighborhoods are at approximately 30 to 45 minutes commuting distance from lower- income communities of East Amman, and 45 to 60 minutes from informal settlements fur- ther South around Wahdat refugee camp. These latter southern informal settlements are expected to benefit from further commuting time reductions under Phase 2 of the BRT, to which they are proximate. Accessibility in eastern low-income neighborhoods could be sup- ported by ’last mile’ connectivity improvements: currently, walking to taxi or minibus stands, waiting for taxis, and taxi rides to reach the BRT station comprise 20 to 30 minutes of the 30 to 40 minute commute time estimated. 15 One of the highlighted neighborhoods in Figure 14 , Amir Al Hassan, located East of Am- man, largely consists of a Palestinian refugee camp with unique planning, zoning, and social characteristics analysis of which is beyond the scope of this paper. The above results on the bindingness of zoning could be subject to measurement errors due to the imperfect capture of housing volumes from the satellite data and due to the difficulty of matching all land-use plots and building types to the legal GAM Zoning Restrictions. In addition, restrictions would become more binding over time, to the extent the Amman population and economy continues to grow, raising aggregate demand for floor space. It is therefore useful to complement these results with a measure of how close the supply and demand for floor space is to the legal limit under the zoning restriction. We show this in Figure 14, which plots the ratio between current floor space demand and the limit under existing zoning restrictions, after BRT Phase II. A ratio larger than 1 implies that zoning restrictions are binding in the neighborhood, while a low number implies that the building stock is far from the legal limits. Figure 15 shows that floor space demand is much lower than the zoning restrictions in most neighborhoods with ratios below 50 percent in many locations. However, the demand for floor space in the most constrained locations in the center-west far exceeds the legal limits with demand being up to 40 percent higher than the legal limit. 7 Conclusions This paper studied the long-run welfare and productivity effects of recently completed and ongoing expansion of the Greater Amman Municipality’s BRT system. It constructed a rich, quantitative spatial model that captured many aspects of the economic geography of Am- man, including households’ residential locations, firms’ and workers employment decisions, building supply and regulation, and sectors’ spatial distribution and productivity. 16 We find that the BRT expansion should support large increases in output for all sectors. Compared to the baseline, it is expected to raise output in manufacturing, tradable services, and non-tradable services by 4.4 to 5 percent in "Phase 1", and 7.2 to 7.6 percent in "Phase 2". Workers in manufacturing benefit the most from the BRT’s expansion, reflecting that they lived farthest from the city center at baseline. Both phases of the BRT improve welfare in all of Amman’s 152 neighborhoods. Phase 1 tends to promote greater mono-centricity of the city, with the greatest welfare and density increases in already-dense neighborhoods in the north-west and north-east. Phase 2 encour- ages more poly-centricity, with additional densification of the (initially less dense) south. Anticipating these incremental changes in employment, population, and/or building density across the city can help city planners to plan appropriate complementary policies to the BRT. This may include, for example, spatially targeting new services (such as education and health facilities, or water and sanitation) to accommodate incoming populations or firms; reviewing and adjusting land use plans to accommodate densification in areas made more productive or attractive by the BRT; land value capture targeting locations benefiting from the BRT; and complementary transport services for targeted areas currently expected to benefit less from the BRT, among other policies. Among these considerations, we analyze the interaction of zoning restrictions with the BRT. This analysis suggests that current land use zoning limits would restrict the private sector’s response to the BRT, by restricting new construction in some locations with large excess demand for real estate resulting from the BRT. The desirability of amending zoning regula- tions in these areas will depend on further considerations, such as any risks associated with densification of those locations. 17 8 References Ababsa, Myriam and Abu Hussein Ahmad. 2020. “Metropolitan Amman. Comprehensive Climate Plan.” Ahlfeldt, Gabriel M, Stephen J Redding, and DM Sturm. 2016. “A Quantitative Framework for Evaluating the Impact of Urban Transport Improvements.” Tech. rep., Working Paper. Bird, Julia and Anthony J Venables. 2019. “Growing a developing city: A computable spatial general equilibrium model applied to Dhaka.” World Bank Policy Research Working Paper (8762). Monte, Ferdinando, Stephen J Redding, and Esteban Rossi-Hansberg. 2018. “Commuting, migration, and local employment elasticities.” American Economic Review 108 (12):3855– 90. Ngai, L Rachel and Christopher A Pissarides. 2008. “Trends in hours and economic growth.” Review of Economic dynamics 11 (2):239–256. Redding, Stephen J and Esteban Rossi-Hansberg. 2017. “Quantitative spatial economics.” Annual Review of Economics 9:21–58. Redding, Stephen J, Daniel Sturm, and Stephan Heblich. 2018. “The Making of the Modern Metropolis: Evidence from London.” . Tsivanidis, Nick. 2019. “Evaluating the impact of urban transit infrastructure: Evidence from bogotaâs transmilenio.” Zárate, Román D. 2022. “Spatial Misallocation, Informality, and Transit Improvements: Evidence from Mexico City.” 18 9 Figures Figure 1: Population density vs. distance to the center (a) Scatter plot (b) Bar plot (c) Spatial distribution Notes: This figure plots the relationship between population density and the distance to the center. Panel (a) plots a scatter plot relating population density to the center distance. Panel (b) groups neighborhoods into deciles according to their distance from the center and plots the average population density in each bin. Panel (c) is a heat map which shows the spatial distribution of population density. 19 Figure 2: Employment density vs. distance to the center (a) Scatter plot (b) Bar plot (c) Spatial distribution Notes: This figure plots the relationship between employment density and the distance to the center. Panel (a) plots a scatter plot relating population density to the center distance. Panel (b) groups neighborhoods into deciles according to their distance from the center and plots the average population density in each bin. Panel (c) is a heat map which shows the spatial distribution of population density. 20 Figure 3: Distribution of Employment Density by Sector (a) Manufacturing (b) Tradable Services (c) Non-tradable Services Notes: This figure plots the spatial distribution of the employment density by sector at the neighborhood level. Panel (a) plots employment for manufacturing, panel (b) for tradable services, and panel (c) for non-tradables. 21 Figure 4: Baseline Travel Time Notes: This figure plots the average travel times between all locations. We first calculate the travel time between each pair of neighborhoods based on the quickest transportation available, where the average travel speeds to different types of road segments are based on estimates from the World Bank Transportation Unit. We then calculate the average travel times from each origin-neighborhood to all possible destinations where we weigh times by the number of jobs that are available in each destination. 22 Figure 5: Baseline Productivity by Sector (a) Manufacture (b) Non-tradable (c) Tradable Notes: This figure plots the cailibrated productivity by sector across neighborhoods. 23 Figure 6: Baseline Average Wage Notes: This figure plots the calibrated average wage across neighborhoods. Figure 7: Baseline Welfare Notes: This figure plots the calibrated welfare across neighborhoods. 24 Figure 8: Construction of BRT Line - Two Phases Notes: This figure plots the BRT investments which are added to the existing transit system. In Phase 1 of the BRT (2022-2024), we only consider the yellow and orange lines. In Phase 2 (2024-2027)we additionally consider the pink lines, assuming the full BRT system is in operation. Figure 9: Travel Time Changes (a) Phase 1 (b) Phase 2 Notes: This figure plots how travel time changes after the BRT investments. 25 Figure 10: Effect of BRT on Employment Density (a) Phase 1 (b) Phase 2 Notes: This figure plots how employment density changes after the BRT investments. 26 Figure 11: Effect of BRT on Employment Density by Sector (a) Manufacture: Phase 1 (b) Manufacture: Phase 2 (c) Non-tradable: Phase 1 (d) Non-tradable: Phase 2 (e) Tradable: Phase 1 (f) Tradable: Phase 2 Notes: This figure plots how employment density in each sector changes after the BRT investments. 27 Figure 12: Effect of BRT on Wage by Sector (a) Manufacture: Phase 1 (b) Manufacture: Phase 2 (c) Non-tradable: Phase 1 (d) Non-tradable: Phase 2 (e) Tradable: Phase 1 (f) Tradable: Phase 2 Notes: This figure plots how wage in each sector changes after the BRT investments. 28 Figure 13: Effect of BRT on Population Density, Rent, Welfare (a) Change in Population Density: Phase 1 (b) Change in Population Density: Phase 2 (c) Change in Rent: Phase 1 (d) Change in Rent: Phase 2 (e) Change in Welfare: Phase 1 (f) Change in Welfare: Phase 2 Notes: This figure plots how population density, rent and welfare changes after the BRT investments. 29 Figure 14: Zoning Restrictions Binding after BRT (a) Phase 1 (b) Phase 2 Notes: This figure plots where zoning Restrictions are binding after the BRT investments. We distinguish between places where restrictions are binding only for commercial real es- tate, only for residential real estate, or for both types of real estate. Figure 15: Ratio between floor space demand and zoning restrictions (a) Phase 1 (b) Phase 2 Notes: This figure plots the ratio between demand for housing and the allowed housing from zoning law limits after the BRT investments. A ratio larger than 1 implies that zoning restrictions are binding and that floor space demand exceeds the allowed limit from zoning restrictions. 30 Appendix Appendix A: Data Construction In this Appendix Section, we explain the data construction and data sources. Employment by Neighborhood and by Sector We obtained information about the numbers of firms by sector and by size category from the establishment census for all Mantiqas. We allow the firm size distribution to vary across sectors to match total employment by sector for the total Amman area. We then use data on commercial floor area to impute employment numbers at the neighborhood level within Mantiqas. Due to data constraints, we limit our analysis to formal employment. Effects of counterfac- tuals could be larger if they provide incentives for informal workers to switch from informal low-productivity jobs to formal employment with higher value added. Residents by Neighborhood and by Sector To compute the number of formally employed residents by neighborhood and by sector, we gather data from several sources. First, we obtain data on residents by education status and by neighborhood from the census. To focus only on residents who are formally employed, we use employment numbers by education status for the total Amman area which we obtain from the establishment census. With this data, we compute formal employment rates for each education group and adjust the resident data in each neighborhood accordingly. Sim- ilarly, we use sectoral employment shares by education status for the total Amman area to 31 impute the sectoral distribution of residents according to each neighborhood’s educational composition. Residential and Commercial Floor Area and Zoning Restrictions by Neighborhood For our region of analysis, we obtain detailed information on the building types for many small parcels. We can link these building types to zoning restrictions, which specify the percentage of the parcel’s area that can be used for construction and the maximum numbers of floors that are permitted for the specific building type. We can further classify building types as residential or commercial (e.g., "office buildings" are commercial, etc.). With this data, we can then compute each neighborhood’s zoning restrictions, i.e. the maximum floor area that can be developed for residential or commercial use in each neighborhood. We combine this data with information on actual floor area that is estimated by neighbor- hood from satellite images. To assign the actual total floor area to either commercial or residential, we use the respective split from the zoning restrictions for each neighborhood (i.e., we assume that zoning restrictions are "equally binding" for residential and commercial buildings at the neighborhood level). Commuting Times between Neighborhoods To compute the commuting times between neighborhoods, we first divide the Greater Am- man area into a grids with cells that are each of size 550 m × 550 m. For these small cells, we then identify the highest-speed transportation mode available in each cell and assign the corresponding traveling speed to it. For example, if there is a BRT connection in a cell, then we assign a traveling speed of 35 kilometers per hour. If there are no roads, then people have to walk and we assign a traveling speed of 5 kilometers per hour to the cell. Next, 32 we identify the centroids of each neighborhood and link each pair of centroids through the shortest route of cells. The travel time for the shortest route is then weighted by each cell’s traveling speed. The commuting time between each pair of neighborhood centroids is then calculated by summing the time spent in each cell, which is equal to the length of the cell’s diagonal (778 m) divided by the cell’s assigned travel speed. 33 Appendix B: Model Appendix This section describes the spatial model that we use to study the effects of the policy coun- terfactuals. The model is based on recent work in the urban literature that has focused more on cities in advanced economies (Ahfeldt et al., 2016; Heblich et al., 2018; Tsivanidis, 2019). We now describe the model set up, workers’ choices, the production side, market clearing, and the equilibrium. Model Setup and Utility Geography. Locations i and j are separated by bilateral commuting costs d i j which de- pend on the city’s road and transit infrastructure. Each location is characterized by residen- tial amenities A i , which capture regional characteristics that make locations more or less desirable for residents and by sector-specific amenities B is , which can make locations more attractive for individuals who work in sector s. We model three sectors s which we refer to as manufacturing, low-skill services, and high-skill services. Locations further differ in residential and commercial housing supply which we denote respectively by H i and Z i . Worker Preferences. Workers choose to live in a location i and to work in a location j and in a sector s. In each work-location and sector, workers earn a wage rate of w js per efficiency unit. Workers have CES preferences over the three sectoral outputs, so they consume the CES aggregator: σ σ−1 σ−1 σ C= αs C s (1) s where αs denotes the consumption shares and σ denotes the elasticity of substitution be- tween the sectors. Workers have Cobb Douglas preferences over consumption and housing H i , so that the util- ity function of a worker who lives in location i and works in location js and in sector s is 34 given by: α 1−α C i js H i js 1 U i js = d− i j ϵ i js (2) α 1−α where α is the Cobb Douglas expenditure share on consumption, d i j captures the utility cost of commuting, and ϵ i js are idiosyncratic productivity shocks over a residence-and-work- location i j and a sector s. The indirect utility of consumption of a worker who lives in location i and works in location js and in sector s as: 1 i j ϵ i js w js d − Vi js = −α (3) P α r1 i where w js is the wage rate per efficiency unit in sector work-location j and sector s, P is the aggregate CES price index , and r i is the rental rate per unit of housing in location i . Workers Choices: Consumption, Residence Location, Work Location and Sector We assume that the idiosyncratic productivity shocks ϵ i js are extreme-value distributed with a nested Frechet distribution, where the upper nest presents the residence-location, the second nest the sector to work in, and the lower nest the work-location. Monte, Redding, and Rossi-Hansberg (2018) and Zárate (2022) show that you can then decompose the probability of living in i , working in js into three terms: η α 1−α A i Wi /P i ri κ B is Wis wθ d −θ js i j λ i js = λ i λ is| i λ i js| is = η κ θ −θ (4) A i′ Wi′ α 1−α /P i ′ r i′ s′ B is′ Wis′ j′ w j′ s d i j′ i′ where the first bracket is the probability of living in i which we denote by λ i . The second bracket is the probability of working in sector s conditional on living in i which we denote by 35 λ is| i . The third bracket is the probability working in j conditional on living in i and working in sector s which we denote by λ i js| is . η is the sensitivity of migration flows within cities to changes in real income, κ characterizes how substitutable workers are across sectors, and θ is the commuting elasticity which captures the extent to which commuting flows respond to wages and commuting costs. We further defined the wage index in location i and sector θ θ −θ s as Wis = j′ w j′ s d i j′ and the expected wage for workers who live in location i is given by Wiκ = κ s′ B is′ Wis′ . Worker’s average welfare (or expected utility before knowing idiosyncratic productivity shocks) is then given by: η A i′ Wi′ ¯= U (5) η α 1−α i′ Pi ′ r i′ Production and Housing Supply We assume free trade within the city. In each sector s and each location j , perfectly com- petitive firms produce a homogenous good with a Cobb Douglas technology that uses labor efficiency units L js and commercial floor space Z j as inputs. Firms profit maximization is therefore given by: β 1−βs π js = p s Q js L js s Z js − q j Z js − w js L js (6) where Q js is location- and sector-specific productivity, beta s is the labor expenditure share, q j is the local price for each unit of commercial floor space and w js is the location- and sector- specific wage rate for each labor efficiency unit. Firms’ first order condition then determines the optimal use of labor and commercial floor space: 1 1 w js L js = q j Z js = p s Y js . (7) βs 1 − βs 36 Construction Sector. Commercial floor space in each location i is equal to: ˜ i qγ Zi = Z (8) i ˜ i denote the exogenous component of commercial housing supply (for example due where Z to fixed land supply) and γ denotes the elasticity of housing supply to rent. If γ approaches zero, then floor space supply would be completely inelastic/fixed. Residential floor space is analoguously defined as: ˜ i rγ . Hi = H (9) i Market Clearing and Equilibrium The model has a set of structural parameters Ω = (αs , σ, α, η, κ, θ , βs , γ). These parameters define preferences (αs , σ, α), elasticities η, κ, θ and production parameters βs , γ). The model’s ˜ j , B is ) which are residential amenities A i , ˜ i, Z regional fundamentals are Γ = ( A i , d i j , Q js , H commuting costs d i j , sectoral productivity in each location Q js , the fixed component of resi- ˜ j , and average level of sectoral productivity B is . ˜ i, Z dential and commercial floor space H Conditional on this set of parameters and regional fundamentals, the equilibrium is defined by a set of sectoral prices p s , commercial and residential rental rates q i and r i , and location- and sector-specific wage rates per human capital unit w js , which ensure that: 1. Good markets clears in each sector so that: Y js = Ys = c s = ασ −σ σ−1 s ps P × I, (10) j where the last equality uses consumers’ CES expenditure shares for each sector s, P is the CES price index, and total income is given by I = i Wi λ i . 37 2. Firms in each location-sector make zero profits (due to perfect competition) so that: β s 1−βs 1 1 p s Y js = p s A js L js Z js = w js L js = q j Z js . (11) βs 1 − βs 3. Labor market clears in each location-sector so that workers commuting decisions and sector choices aggregate to firms’ total labor demand in each location and each sector L js . 4. Commercial and residential floor space markets clear in each location so that: 1 − βs qjZj = w js L js (12) s βs r i H i = (1 − α) I i (13) Derivation of Choice Probabilities with Nested Frechet Distribution Utility function: α 1−α C i js H i js 1 U i js = d− i j ϵ i js (14) α 1−α Workers receive an idiosyncratic productivity shock ϵ i js and they earn a wage rate w js per efficiency unit. Commuting costs reduce workers productivity/efficiency units due to less time spent at work. The income of a worker living in i and working in j and in sector s is 1 i j ϵ i js and the worker’s indirect utility function is equal to: therefore given by: w js d − 1 i j ϵ i js w js d − Vi js = α 1−α (15) Pi ri 38 Productivity shocks ϵ i js are drawn from the nested Frechet distribution H(*):   κ η  κ θs −θ H (ϵ) = exp − Ai  B is ϵ i jss , (16)    i s j where A i captures amenities of living in location i and B i s captures the average productivity of a worker who chooses to work in location j and sector s. Probability of living in i, working in s and j λ i js = λ i × λ is| i × λ i js| is : α 1−α η κ sθ −θs A i Wi /P i ri B is Wis w js di j λ i js = η κ θs −θs (17) α 1−α A i′ Wi′ /P i ′ r i′ s′ B is′ Wis′ j′ w j′ s d i j′ i′ We derive this expression backwards starting first with the conditional probability of choos- ing to work in j conditional on living in i and working in s. The distribution of this condi- tional indirect utility Vi js| is is given by: −θs −θs 1 Vi js| is ∼ exp ϵ i js = exp Vi js w− js d i j , (18) so that the probability of working in j conditional on living in i and working in s is equal to: s θ −θs w js di j λ i js| is = θ −θ (19) j′ w j′ss d i j′ s We now derive the probability of working in sector s conditional on living in i . This condi- tional indirect utility Vi js| i can be written as: κ κ −θs     θs θs Vi js θ −θs κ Vi js| i ∼ exp B is 1  = exp B is s w js di j Vi− js . (20) j w js d − ij j 39 so that the probability of working in s conditional on living in i is equal to: κ s θ −θs θs κ B is j w js d i j B is Wis |i λ is| i = κ = κ , (21) θs −θs θs s′ B is′ Wis′ | i s′ B is′ j w js′ d i j 1 s θ −θs θs where we define: Wis| i = j w js d i j . The Frechet property further implies that average income of all residents who live in i can be expressed as: 1 κ κ Wi = B is′ Wis ′|i . (22) s′ Last, we need to derive the probability of living in i . The indirect utility is given by:  κ η     −θs θs κ 1 −1    w js d − ij   Vi js ∼ exp  A i  B is  Vi js (23)       α 1−α      s j Pi r i  Ai κ κ η −η Ai η −η = exp α 1−α η (B is Wis ) Vi js = exp W V α 1−α η i i js , (24) (P i ri ) s (P i r i ) so that the probability of living in i is equal to: α 1−α η A i Wi /P i ri λi = η. (25) α 1−α i′ A i′ Wi′ /P i ′ r i′ The Frechet property further implies that average welfare (or expected utility) before know- ing idiosyncratic productivity shocks is given by: η A i′ Wi′ ¯= U (26) η α 1−α i′ Pi ′ r i′ For market clearing, we need to derive the total amount of efficiency units in each work 40 location j and each sector s. Let us first express total income in j and s as the sum of workers’ average income from all residence locations i weighted by the share of workers who commute between i and j , which is equal to: tot W js = Wi′ λ i′ js . (27) i′ It follows that the number of efficiency units L js in each location-sector pair js is equal to: tot L js = W js /w js (28) 41 Appendix C: Model Calibration Calibration of Model Parameters We calibrate a set of structural parameters from the literature as shown in the table below. Table 2: Parameters calibrated from the literature Parameter Description Reference θ=4 commuting elasticity Zárate (2022) sensitivity of migration flows within η = 1.5 Zárate (2022) cities to changes in real income substitution elasticity of workers are κ=2 Zárate (2022) across sectors δ = 0.65 housing supply function Zárate (2022) β= Bird and Ven- production share on labor (0.85, 0.93, 0.94) ables (2019) Zárate (2022), α = 0.7 Consumption expenditure share survey esti- mates Bird and Ven- ables (2019), σ = 0. 8 CES preference Ngai and Pis- sarides (2008) αs = Bird and Ven- consumer sectoral expenditure share (0.24, 0.12, 0.64) ables (2019) Inferring Regional Characteristics by Fitting the Data to Model Equations To infer the regional characteristics, we use the following data moments: 1. Travel time matrices d i j between all neighborhood pairs i j 2. Population by sector and residence neighborhood 42 3. Employment by sector and workplace neighborhood 4. Residential floor space by neighborhood 5. Commercial floor space by neighborhood 6. Zoning restrictions which specify maximum residential and commercial floor space that can be developed in each neighborhood The calibration proceeds in the following steps. 1. We first solve for wage rates per efficiency unit for each work location and sector w js . To do so, we use Equation 4, data on travel times d i j , data on population by neighborhood and by sector λ i and λ is| i , and data on the number of employees by work location j . Using this data, we guess wage rates w js and iterate until the number of workers in each work location and sector js matches the data. 2. We then solve for the average productivity shifters of workers who live in location i and work in sector s B is . To do so, we use Equation 4, the wage estimates w js from Step (1), and data on population by neighborhood and sector λ is| i . 3. We then solve for residential rents and the fixed component of residential floor space using the market clearing condition in Equation 9, our previous estimates, and data on the residential floor space in each neighborhood. We use our previous estimates to compute total income in each neighborhood i which implies the expenditure share on residential housing due to the assumed Cobb-Douglas preferences. Data on residential floor space and the labor market clearing condition 13 then allows us to solve for each neighborhood’s rental rate r i per housing unit. Using Equation 9, we can then infer the fixed component of residential housing supply. 4. We further solve for commercial rents and the fixed component of commercial floor space. We first use our previous estimates to solve for firms’ total expenditure on labor 43 in each location j . Given fixed expenditure shares under the Cobb Douglas production functions, this implies firms’ expenditure on commercial floor space. Using data on commercial floor space and market clearing conditions (Equation 12) therefore allows us to solve for commercial rental rates q j in each neighborhood. Using Equation 8, we can then infer the fixed component of commercial housing supply. 5. We can then solve for sectoral prices p s and productivities Q js in each location and sector js using the goods market clearing condition (Equation 10) and the zero-profit condition in each js pair (Equation 11). We normalize the CES price index to 1 without loss of generality. 6. Last, we can solve for amenities A i using Equation 4 and data on population in each neighborhood. 44