I'm an Economist by training and Data Scientist by trade. I ran my first Python script in the summer of 2016. The code scraped airlines' website to get airfare prices. These prices would later be used to forcast inflation and other economic indexes. It wasn't much, but it managed to free some of my time while I worked as an intern at the brazilian Ministry of Economics. It didn't take long before I used that freedom to diverge into the brave new world of big data and machine learning, which, at the time, was less than embryonic in Brazil.

During all my graduation, I got tired of hearing how bad economists are at predicting, well… anything. The low R2 in the econometrics course depressed me. For a time, I thought I'd found the cure for that in Machine Learning. I couldn't wait to go home and watch dozens of videos on Deep Neural Networks and Boosted Trees.

Eventually, all that studying paid off and I landed a job at a rising fintech. Finally, I could dedicate all my time to squeezing the last drop of predictive performance out of my models. It took me about a year to realise something wasn't quite right. I had very performatic models and yet, they weren't as useful as I thought they would be. Sure, I could predict what would happen to a customer or what would be next month's marketing metrics. But that was all rather passive. I couldn't influence the results in the direction that I wanted.

Eventually, I would say by destiny, but probably by luck, my interest in Econometrics came back. A web series on Econometrics and Machine Learning by Susan Athey and Guido Imbens showed me that the thing I was after would take more than good predictions. I was finally on the trail paved by Joshua Angrist, Alberto Abadie, Hal Varian, Scott Cunningham, Victor Chernozhukov and many others. As I progressed, good predictions became less and less interesting. I guess I finally understood the discreet charm of Linear Regression.

Now, like those that taught me, I too "spend my days (at least, the good ones) happily pursuing regression output" and there is nothing I would rather be doing. I hope this blog explains why.