Combining cluster analysis with synthetic control for evaluating economic impacts of the dam breach in Mariana, Brazil

The Fundão dam breach, located in Mariana-Minas Gerais, Brazil, took place in November 2015 and caused social, economic and environmental impacts in the affected region. Here we use synthetic control to assess its economic impact on the municipality of Mariana. We analyzed real GDP per capita as the response variable and used socioeconomic indicators to characterize the municipalities affected by the dam breach and select the control group. We propose a hybrid method that uses cluster analysis as a sample preselection step before synthetic control. The initial donor sample of 818 municipalities was reduced using hierarchical clustering or time series clustering (dynamic time warping). Then, we applied the synthetic control. The results obtained in the synthetic control without the prior sample selection stage did not prove to be a good counterfactual for Mariana. In addition, as the sample size of the control group is large in this first situation, the computational time of the analysis was large, exceeding 5 h. When the approaches using clustering and time series clustering were used, the time spent was reduced by more than 99% in both situations and the counterfactual obtained had a satisfactory fit.