The OECD asked Thibault Schrepel, Associate Professor of Market Law + Technology at the VU Amsterdam, to record a video on computational antitrust. Here it is, following with its transcript.
Transcript:
We often discuss how algorithms can be used to infringe antitrust law. We, less often, discuss how they can be used to also assist antitrust agencies. And yet, there are several reasons why we may want to. First, the detection of infringement to antitrust law remains unsatisfying. Second, it seems safe to assume that the detection rate will lower in the coming years without the use of the proper tools. That is mainly due to the fact that the amount of data we produce is increasing exponentially. And third, in an increasingly complex economy, when they want to take more elements into account when detecting infringements to competition law than just static parameters.
Well, I come with some great news. Computational tools, which I define as computer-based problem-solving methods, can help. To begin with, one can use these tools to detect more practices, for example, using markets creating tools or natural language processing, thanks to which one can analyze thousands of documents. And as a matter of fact, there is one place where all antitrust agencies can start: using their own case law. Should they label their past decisions properly, they could train machine learning systems to improve detection capacity.
Second of all, competitional tools could help in the context of merger control. For example, federated learning could help to detect potential killer acquisitions by studying the fitness of companies, which measures a firm’s ability to translate size into growth. And lastly, these tools could help with the design of policies and remedies. Agent-based modeling is already used in numerous fields such as biology, sociology, or economics. Antitrust law could benefit from it to simulate the effects of a merger or a new policy on the market.
Now, of course, there are challenges ahead of computational antitrust, for example, how to compute innovation, or… what to do with non-computable evidence versus computable results, or… how to think Black Swans into account. We will come to some answers. In the meantime, let’s remember that although weather forecast is not perfect, it remains quite useful. The same has to be true for competitional tools in antitrust.
