top of page
  • Writer's pictureFederico Nutarelli

Economic complexity with the eye of machine-learning

Updated: Jan 9

Since 2009, the concept of economic complexity has been widely recognized in the field of economics, thanks to the influential work of Professors Hidalgo and Hausmann, who first introduced this idea. To understand what economic complexity is and why it's so crucial, let's hear from Professor Hidalgo himself:

In their groundbreaking research, and in subsequent studies by other scholars, the concept of economic complexity was calculated using rather intricate algebraic methods which I won't cover here (check out the white paper of Hidalgo and Hausmann if you are curious). Although the idea itself is straightforward and impactful, the actual computations behind it can be quite improved via machine-learning. Not only this, also the interpretation of what a complex country is, remains quite simple although from a novel perspective.


A country is deemed as complex if it is harder for the machine-learning to predict its RCA entries.

Now all complexity measures base on this concept of Revealed Comparative Advantage (RCA). The RCA is a way to figure out what a country is really good at producing compared to other countries. Imagine every country is a student in a class, and each student is really good at one subject. RCA is like a report card that shows which subject each student excels in more than the others.

For a country, if the RCA in making a certain product is high, it means the country is better at producing that product compared to other products it makes, and also better than other countries making the same product. It's like saying this country is the "class topper" in making that particular thing. This helps countries understand what they are best at and can focus on for trade.

The idea is that the least our algorithm is capable to reconstruct the true RCA value of a country in a product category, the more that country is in a way "complex to understand".

Of course there are refinements that we made. For instance predictions' failures are not the same...

What if the algorithm says that a country should have an advantage in product P, but the data show that this is not the case? Isn't this different from the reverse?

For these cases we adjusted our complexity index, which by the way is called MONEY, through a weighting system.

But I do not want to spoil everything. If you are interested check out our published paper here. And stay tuned for the github package of MONEY!!!

Recent Posts

See All


bottom of page