Local Incentives and Illegal Mining: The Effect of a Royalties' Law Reform

Student Researchers: Mauricio Romero and Santiago Saavedra
Location: Colombia

The project looks at the effect of  a "recentralization reform" in Colombia that reduced the share of royalties to mining municipalities from 55 to 10 percent. Theoretically, the reform could lead to an increase in illegal mining as local politicians have fewer incentives to monitor mines and ensure they comply with the law. The research team is using Landsat satellite images and machine learning algorithms in order to identify illegal mining activity.

To calculate the impact of the reform on illegal mining, the team will compare municipalities affected differentially by the reform across time, controlling for time and municipality fixed effects. This identification strategy relies on the assumption that, after controlling for time-fixed effects, municipality-fixed effects and time-varying controls (like price of minerals), the extent of illegal mining is only affected differentially by the impact of the reform in the municipality budget.

A necessary input to perform this analysis is a time series of illegal mining. Previous work has used static measures of illegal mining. Since we require a time-varying measure of illegal mining we rely on remote sensing.Information from the mining census of 2010-2011, which has the geolocation of over 14,000 mines (of which over 9,000 are illegal), will be used to train a machine learning model (such as a random forest) that detects illegal mining. Researchers will use this model with Landsat 7 imagery from other years to get a proxy of the mining activity in each municipality across time.

Results

The first version of the panel of illegal mining by municipality for the years 2004-2014 has been created. The current prediction model has a precision of 79%, that is, of 100 pixels we predict as mines 79 actually are according to our testing sample.

However, the true positive rate of the model is only 32%, that is, of 100 pixels we know are mines the model only finds 32. Therefore, some work remains to be done to improve the model, especially having stable predictions across different
years.

The research team will repeat the analysis for Peru, to provide a comparison country for Colombia.