Pricing algorithms have been proven to be especially challenging to regulate from a legislative competition law perspective. Algorithmic monitoring of competitors’ prices does not necessarily constitute an infringement if there is no evidence of an agreement to collude.
In a recent study ordered by the Swedish Competition Authority it is however concluded that price algorithms can be expected to reach collusive outcomes in the long run. It is therefore possible that this finding will have a future impact on legislation concerning digitalised manners of communicating and cooperating on innovation as it concerns central themes in competition law.
Many companies selling through e-commerce platforms as well as through individual websites use algorithms to control pricing. In 2017, 53% of the respondents in the European Commission’s e-commerce sector inquiry tracked the online prices of competitors, and a majority subsequently adjusted their own prices using software. This number has likely increased dramatically. Depending on how the algorithm is programmed to act, it can affect competition either positively or negatively. Previous studies have indicated that the use of autonomous algorithms can lead to higher prices and profits for the businesses utilising them. In the study Collusion in Algorithmic Pricing by Tuwe Löfström, Hilda Ralsmark and Professor Ulf Johansson, the authors point out that the same result can occur when businesses use different types of autonomous algorithms or algorithms whose learning occurs at different times. The results also show that the threat of new businesses entering the market can inhibit price development and lead to lower price levels.
Technological developments in recent years have enabled autonomous agents to use artificial intelligence to learn optimised pricing policies by interacting with the market. Both competition authorities and researchers have expressed concerns that autonomous price-optimising agents acting independently in the same market may learn each other's policies through the implicit interaction. Their concern is that in this way the agents reach a price outcome that resembles regular price cooperation.
According to the results of the study, price-optimizing agents can be expected to achieve price outcomes close to those prevailing in regular price cooperation, regardless of the strength of the agents used. Pricing algorithms can thus be expected to learn to reach collusive outcomes. When algorithmic agents are different in some aspect, e.g., their underlying algorithm or their updating frequency, they will no longer share the profit equally. Instead, one algorithm (the stronger algorithm) will dominate the other, and obtain a clear majority of the profit gain. Unfortunately, the prices are often higher than for markets created by two identical agents.
Furthermore, the empirical evidence implies that businesses can achieve competitive advantage by continuously improving their price-optimizing agents. The threat of new entrants inhibits price evolution and results in reduced price levels. This creates a clear incentive for businesses to continuously improve their technical solutions.
The report on Collusion in Algorithmic Pricing can be downloaded (in English) here.