Machine learning seems to be everywhere in economics these days. I wondered – has these been a gradual trend or is this a sudden explosion? So I again turned to Academic Sequitur data. This time, I decided to stick to NBER working papers as my data source, largely because they lead journal publications by a few years. I looked for the following terms in the abstract or title: “machine learning”, “lasso”, “neural net”, “deep learning”, and “random forest”. The graph below shows the percent and number of NBER working papers that meet these criteria over time (on the left and right y-axis, respectively).
An explosion indeed! Virtually no paper abstract/titles mention anything machine-learning related in the abstract in 2000-2014. Then we have a respectable five papers in 2015, one paper in 2016, followed by 15 papers in 2017, 22 papers in 2018, and five papers so far this year. As a percentage of total papers, the machine learning papers are small, however, making up at most 1.5% of total papers. Whether the numbers stagnate or keep skyrocketing remains to be seen!
And in case you’re wondering, the prize for the first NBER working paper to utilize machine learning goes to…”Demand Estimation with Machine Learning and Model Combination” by Patrick Bajari, Denis Nekipelov, Stephen Ryan, and Miaoyu Yang, issued in February of 2015.
Update: here’s how the graph would look if you also counted “big data” as indicating machine learning. Prize for first NBER paper to mention “big data” goes to “The Data Revolution and Economic Analysis” by Liran Einav and Jonathan Levin, issued in May 2013.
Forthcoming (if this post is popular): published papers utilizing machine learning!