Working Papers

Threshold Externalities and Optimal Pricing: Evidence from Traffic Congestion (with Antonio Bento and Jonathan Hall) 

Traffic congestion is a major problem in large cities worldwide. This paper uses high-frequency data from the Los Angeles metropolitan area combined with an instrument that varies spatially and temporally to estimate the causal impact of an additional vehicle mile traveled on travel times. Specifically, we exploit the network structure of the Los Angeles highway system and use crashes on close alternative routes as exogenous shocks to traffic demand. To do so, we rely on Google Maps to determine the ideal route and alternatives for over 19,000 real-world commutes. We uncover strong nonlinearities in the speed-flow relationship: below a congestion threshold, an additional trip does not affect speeds, while above the threshold the effect rises to 3.3\%. Using these estimates in a congestion pricing model, we find that an optimal toll of 30 cents per mile during peak hours would more than double average highway speeds while reducing peak vehicles miles traveled by less than 10\%, generating annual welfare gains of $380 million. In contrast, setting an optimal toll without accounting for this nonlinearity leads to welfare losses of $18 million per year, highlighting the importance of incorporating nonlinear speed-flow effects in congestion policy design.

All Cops are Bayesian: How Selection Neglect Drives Statistical Discrimination in Policing (with Juan Gonzalez, Santiago Perez-Vincent and Ervyn Norza)  [Draft]

Policing is increasingly data-driven, but crime data is a selected sample that reflects where crime is observed, not where it happens. We run a lab-in-the-field experiment testing how senior police officers in Colombia infer crime from selected data. Officers take crime data at face value, turning differences in observability into biased predictions. This failure reflects officers’ mental models of crime data, not missing information. Because crime is more observable among minorities, selection neglect can sustain inaccurate statistical discrimination purely from bad inference on selected data.

The Impact of Crime Perceptions on Public Transport Demand: Evidence from Six Latin American Capitals (with Juan Gonzalez and Santiago Perez-Vincent) [Draft]

Public transit is a central tool for urban mobility in developing countries, but crime may undermine its use and the policies that support it. We study how crime affects transit demand, price responsiveness, and policy preferences through a pre-registered survey experiment with over 5,000 participants in six Latin American capitals. First, commuters are willing to pay a premium of more than half the current fare to avoid crime exposure. Second, crime reduces the choice of public over private transport and limits the efficacy of price instruments to induce potential users to switch from private to public modes. Third, higher perceived crime crowds out support from forward-looking investments, including service frequency and emissions reduction. Crime imposes a cost on frequent riders, deters marginal ones, and creates a second-best environment in which common transit policies become less politically supported. 

The Effect of Bikesharing on Subway Ridership (Master's Thesis) [Draft]

Accessibility to public transportation is a key determinant of transport mode choice. When the distance to or from a public transport hub is beyond walking range, commuters tend to dismiss trip alternatives that require walking to such a hub. This paper examines whether bikeshare systems can improve accessibility to subway stations, thereby expanding their catchment areas and increasing ridership. I study this problem in the context of the City of Buenos Aires, where more than 400 bike stations opened in staggered dates. Following a differences-in-differences design, I estimate that when a bikeshare station opens within 400m of a subway station, the daily volume of passengers of the latter increases by 14.5\%. Moreover, this effect is stronger in non-working days, where I register an increase of 20.7\%. Furthermore, I find that for each additional subway station receiving a nearby bikeshare station, the ridership of its subway line increases by 13.6\%, indicating that the results are not influenced by spillover effects. Finally, I propose a discrete-choice model that could be used to estimate the optimal locations for bikeshare stations, contingent on the availability of additional data. These results highlight the potential of bikeshare systems to boost subway ridership in urban areas, thereby reducing congestion and emissions through the promotion of sustainable transportation.

Work in Progress

The Geography of Opportunity: Effects of Public Transit Subsidies on Labor Markets, Housing and Welfare 

Why do Immigrants in Commute Shorter Distances? The role of information frictions, network and value of time (with Pierre Coster) 

Estimating Demand for Battery Swaps to Apply Efficiency Pricing: A Field Experiment with Electric Boda Bodas (with Mychaela Paetow and Juliana Pinillos)