Can Machine Learning Predict Quantity and Duration of Migration to the USA?

Abstract

The number of Mexican migrants in the USA has become tripled between 1990 and 2017. This surge in Mexican migrants has attracted a lot of attention not only from policymakers but also from economists. We use a set of pre-immigration variables for more than 25,000 individuals from Mexico to predict (i): whether individuals from Mexico migrate to the USA (ii) if they do so, how long they stay in the USA. We use 8 machine learning techniques and we conclude that we can predict correctly 72% of Mexicans who migrate to the USA and 93% of Mexicans who do not migrate to the USA. However, by using only pre-immigration variables our model does not perform well in predicting how long Mexican migrants will stay in the USA. We can only predict 35% of the variation in the number of months that Mexican migrants will stay in the USA.