Thesis title: Machine learning to assess relatedness in Economic Complexity
Relatedness is a measure of the affinity between an activity and a location. In the Economic Complexity framework it is used to quantify how much a country is close to the export of a product. Since knowing which products are within the reach of a country is a powerful tool to increase its diversification and with it its fitness, relatedness is a key tool for institutions and policy makers, and a driver for investments. In my PhD thesis I address the problem of measuring the relatedness focusing on the importance of empirical evidence to validate or falsify a model, that is something at the basis of every study in physics, but it is often neglected in economy. Our proposal is to interpret the relatedness between a country and a product as the probability that the country will export the product. Quantifying how good a relatedness measure is in forecasting the future exports we can determine which is the best model. I will show that a machine learning approach, in particular a decision tree based algorithm, provides the best assessment of the relatedness and I will discuss about its critical points like the explainability of the results and the computational effort.