04/11/2025
🇯🇵 In Japan, it seems the topic is now firmly on the radar.
Rashid Baxter, “Self-learning singled out as most challenging algorithmic pricing risk,” GCR (Nov 3, 2025) https://globalcompetitionreview.com/article/self-learning-singled-out-most-challenging-algorithmic-pricing-risk
In this interview, Aoki (公正取引委員会) said that [self-learning algorithms may instead push businesses towards collusion as an optimal solution for profit maximisation] and then stressed that [agencies would have to find ways of linking liability to those who wrote the algorithm, or those that deploy it in their businesses, she said. “That would be a completely different frontier.”]
I’m not well-versed in technology, so this might be a naïve question, but—
if firms are driven by profit maximisation, and algorithms are designed purely based on instrumental rationality, then why do pricing algorithms in theory converge to colluding? Wouldn’t it actually be more profitable to collude first and then defect?
* However, I just realized that, according to Calvano et al. (American Economic Review 2020), "the algorithms consistently learn to charge supracompetitive prices, without communicating with one another. The high prices are sustained by collusive strategies with a finite phase of punishment followed by a gradual return to cooperation." Interesting... https://www.aeaweb.org/articles?id=10.1257%2Faer.20190623
My current intuition is that, if any algorithmic collusion happens, then the firms' claim that they had no involvement whatsoever will, at least in most realistic scenarios, be hard to accept. In such cases, the real difficulty seems to lie in competition authorities' investigative and evidence-gathering capacity (access to code, data, logs), rather than in any fundamental flaw in a choice-based competition law framework.
Alternatively, maybe, at least as far as purely self-learning, black-box collusion scenarios are concerned, much of the discussion about algorithmic collusion is simply an overblown concern about something that is unlikely to happen in reality.
See, other jurisdictions move: “EU Steps Up on Algorithmic Pricing Cartels, Joining the US and Other Jurisdictions” (Jul 16, 2025) https://riskandcompliance.freshfields.com/post/102kt76/eu-steps-up-on-algorithmic-pricing-cartels-joining-the-us-and-other-jurisdiction
Self-learning algorithmic pricing is a major concern because it can encourage collusion on key business decisions, but building successful antitrust cases to tackle those practices remains highly challenging, a senior Japanese enforcer has said.