The Fiscal Contract up Close: Experimental Evidence from Mexico City

2009-07-01

2011-12-31

-- We will focus on the intensive margin of compliance; that is, we will use the amount of yearly property taxes that is paid in 2012 as our main outcome.

-- In a secondary analysis, we will report results using the extensive margin; from recorded payments, we will construct an indicator vari- able that takes a value of one if any payment is made in a given bimester (including retroactively for yearly payments, which are an available option for taxpayers).
-- To examine the medium-term effects, we will also use the yearly property taxes paid in 2013.

We will reanalyze a large experiment conducted between 2009 and 2011 as part of the randomized evaluation of a major national infrastructural spending program (Ordóñez-Barba et al. 2013; McIntosh et al. 2018). The Hábitat program targeted low-income neighborhoods and invested in local amenities to improve the quality of life in benefited communities.

Following the registration of the pre-analysis plan, we will match the randomized evaluation data to the property tax administrative records to measure the impact of this urban infrastructure program on property tax compliance in Mexico City.

We rely on the prior randomization implemented by McIntosh et al. (2018).

The randomized evaluation of Hábitat consisted of the selection of eligible low-income urban “polygons,” roughly equivalent to neighborhoods, across 65 participating municipalities. The randomization proceeded in two stages. First, the probability of polygon assignment into treatment was randomly assigned, between 0.1 and 0.9, at the municipal level. This led to different polygon treatment assignment probabilities in each municipality. Second, using the realized assignment probability in each municipality, polygons were assigned into treatment or control within participating municipalities.

Yes

-- In Mexico City, the original evaluation included 20 polygons.
-- In a secondary analysis, we will also use additional polygons from the scale up phase of the program as additional control units to increase statistical power. Specifically, we will use polygons that received Hábitat during 2014 but were not included in the program during the program's evaluation (2009–2012) or during 2013.
-- Finally, if we are unable to reject the main null hypothesis, we will construct an even bigger control sample, using areas in Mexico City that were not covered by the program until at least 2014. We will design a data-driven algorithm to cut the spatial data into polygons that match the treated polygons on observable characteristics.

We anticipate the total number of properties included in the main analysis --- which will use the 20 polygons from the original evaluation --- to be as large as 11,000. We will know the exact number of observations when we match the randomized evaluation data to the property tax administrative records.

Theres is only one treatment arm in this study.
-- In the main analysis that uses the assignment from the original evaluation, 12 polygons were assigned to treatment and 8 to control.
-- In the secondary analysis with and augmented sample of control polygons, 12 polygons were assigned to treatment and 29 to control.

See Pre-Analysis Plan.