In this era of digitisation, pricing strategies are amongst business approaches that have been revolutionised. Critical business decisions have shifted into the ‘hands’ of computers, via pricing algorithms (PA). Exponential growth in data use and collection, combined with the rising commonality of PA, has resulted in extraordinary levels of market transparency. Ultimately, this may benefit consumers, enabling informed purchase decisions whilst facilitating easier switching. However, these seemingly pro-competitive consequences represent only one side of the ‘double-edged sword’. This blog will examine the other ‘edge’, considering the overarching competition implications placing markets at an increased risk of distortion. These competition concerns stem from anti-competitive scenarios the implementation of PA facilitate, impacting competitive dynamics at the same level of the supply-chain as well as upstream and downstream, requiring consideration from the horizontal and vertical perspective. Whilst many concerns addressed throughout have been branded ‘futuristic’ with limited evidence grounded in few infamous cases, the author believes these merely represent the ‘tip of the iceberg’ for competition authorities. This blog post seeks to address whether this newly emerging ‘norm’ to digitally track and alter prices according to competitors’ is sufficiently addressed under existing provisions, or whether PA warrant an adaptation to the rules via a new competition toolkit.
PA in the Digital Sector
PA’s have extended far beyond their expected purpose in markets such as quantitative finance. The widely circulated statistic that Amazon were making 2.5 million price changes per day as early as 2012 suggests PA are utilised only by big-tech. However, Walmart’s acquisition of Jet.com which related to underlying motivations to develop Walmart’s algorithmic competency, signified the growing relevance of PA in more ‘traditional’ markets. Furthermore, the E-Commerce Sector Inquiry uncovered that a majority of retailers digitally track competitors, two thirds of which automatically adjust their prices accordingly. PA are no longer a ‘luxury’ tool refined to multinational conglomerates with advanced technical capabilities, they are now considered common in nearly every sector, as well as amongst smaller players.
The Horizontal Context
Implementing Pre-Existing Agreements
In considering whether an algorithmic decision might constitute an agreement between competitors, the “messenger” scenario outlined within the “Erzachi & Stucke taxonomy” is the most straightforward. Where PA are merely the facilitative tool to implement a pre-existing agreement, the prior human contact will satisfy the law’s prerequisite of a “concurrence of wills between two parties”. Therefore, the relevance of PA is simply a subsequent fact to the case at hand.
However, the overarching competition concern in this context is the almost assured stability of these agreements. As noted by the CMA, PA programmed to react almost instantaneously to market developments “make cheating on the cartel arrangement more difficult”. Consequently, the possibility of error and the requirement for ongoing communication between competitors is largely reduced, if not eliminated. Moreover, parties to the agreement need not be concerned with issues such as “agency slack” whereby company representatives may undercut the collusive price. As a result, factors the economy rely on to naturally break down cartels, are combatted by PA. Considering in tandem, these inherent detection difficulties and the renowned asymmetry between profitability from ‘successful’ cartel activity and fines issued by the Commission, the core concern is that PA may make collusive ventures more attractive.
Exacerbating the Legislative Gap
PA in this “predictable agent” scenario are designed and implemented individually. Whilst this often means little evidence to suggest anti-competitive intent amongst competitors, the core function of PA is to respond to market characteristics which unavoidably includes competitors’ market behaviour. Matching competitors’ discounts and presenting consumers with lower prices is theoretically pro-competitive. However, the reality is much less obliging. Realistically firms will be deterred from discounting entirely, as the reactionary PA will promptly retaliate to the market by aligning prices, subjugating any economic advantage obtainable from the initial price drop.
Considering that ‘price leaders’ can operate confidently assuming other players are likely to align prices in light of possible higher profit-margins, they are in a powerful position to fluctuate prices above the competitive equilibrium. Ultimately, the overarching concern is that consumers will be presented with persistently high and often aligned prices. On the off-chance any price-competition is to persist, this will occur so quickly, consumers will be unaware of it.
Some refute the materialisation of these concerns, claiming tacit collusion (TC) is unlikely in practice, as “even in simple static games” co-ordination is unachievable without communication. However, applying this “no collusion absent communications” theory to the German fuel market that attempted to promote competition by requiring firms to report price changes, prices should have decreased. Unsurprisingly, the enhanced market transparency resulted in higher prices. However, this price increase was not attributable to “communications”. Rather, firms’ acknowledgement of their interdependence, and realisation that by preserving higher prices, to the dismay of competition and consumers, higher profits will be aggregated.
Whilst this resembles, at least, mere ‘conscious parallelism’, this falls short of the legal stipulation that there must be a “meeting of the minds” to establish an agreement/concerted practice. Perhaps this so-called ‘gap’ in the law that pushes TC out-with competition authorities’ reach may have been unproblematic when few markets resembled oligopolistic features. However, this ‘digital norm’ of utilising PA which increases market transparency, exposes many markets to this “oligopoly problem”. Firms will likely claim they are merely “adapting intelligently to the existing and anticipated conduct of competitors”. However, this is realistically a convenient transcript to mutually benefit from an outcome akin to explicit collusion – namely, supra-competitive prices and high profit-margins. Whilst these anticompetitive outcomes once considered ‘incidental’ to this market structure may remain free from agreement in this environment, they are certainly not free from intent.
The Vertical Context
Algorithm Driven Hub-and-Spoke
The widespread use of PA has led to the incidental creation of a new market for PA suppliers which is vital for smaller players with limited data points, realistically incapable of developing comparable PA to large players. However, when various market players utilise the same or “similarly minded” PA and thus align at code level, a de facto hub-and-spoke arrangement may arise. However, in this algorithmic environment, PA as the “hub” does the talking and therefore there is no ‘explicit’ co-ordination between competitors to suffice Art.101 TFEU “by object” agreement. As a result, they must be considered within the scope of “by effect” agreements.
Conversely, the CMA understates the feasible concern that outsourcing PA could lead to an alignment of prices, claiming firms need to “resist the temptation” to gain short-term profits by altering their PA and undercutting competitors, this may ring true when considering platforms such as Airbnb where the PA only recommends a price. However, this argumentation is weak when referring to platforms like Uber whose PA are delegated ultimate pricing autonomy. Moreover, Uber carefully phrase their surge PA which essentially increases prices, as a mechanism for meeting market demand. The fact similar PA have been justified in the EU on efficiency terms, is inherently concerning from a competition perspective. By charging prices determined by Uber’s surge PA, with mutual awareness that all other Uber drivers have agreed to charge the same fares, these drivers (who are independent contractors capable of competing) TC to an agreement not to undercut one another on price. The overall result of this type of vertical restraint is akin to that of a horizontal arrangement between drivers which would amount to a “by object” price-fixing agreement. However, as this algorithmic environment removes the prerequisite communication amongst competitors, these competition concerns may slip through the grasp of regulators.
Despite using identical, or at least similar PA, learnings from Eturas indicate only those aware of other firms use of this PA could be considered to have TC to involvement in a hub-and-spoke arrangement. The Rotterdam petrol market where popular software provider a2i systems boasts about their ability to “lower the cost of price wars or better yet, to avoid them“ falls in line with Erzachi’s claim that “surely each retailer knows that the vendors PA will make use of its own and its rivals information in assessing prices”. If construed as ‘knowledge’, the traditional hub-and-spoke format starts to emerge, each firm contributes its data and delegates pricing autonomy to the PA, acknowledging competitors doing the same, with the overall aim of foreclosing competition. This example similarly demonstrates that only a quarter of the market need be utilising the same PA for the overarching concern of higher prices and reduced choice for consumers to materialise. However, the CMA refute that knowledge could be derived without being able to “reverse engineer a competitor’s algorithm”. Considering this technical capability would be limited to few firms, the CMA’s statement suggests that the current toolkit sets an unreasonably high standard for knowledge. Ultimately, this appears contrary to their claim that this hub-and-spoke conspiracy presents the most imminent concern.
The recent Asus decision digitalised an established competition concern. Demonstrating that manufacturers, by identifying only a few retailers not complying with resale price maintenance (RPM), and punishing this ‘deviation’ by forcing them to charge supra-competitive prices at the threat of cut-off supplies, can conduct a wide-reaching form of vertical restraint. Ultimately, this softens competition at intra-brand level as reactionary PA in the market will extend the effect of this clause to retailers far beyond its scope, with few evidential actions. Although the current toolkit seemingly addresses this by expressly prohibiting RPM, the lack of action now required to enforce these clauses may stimulate such practices. In light of this increased monitoring facilitated by PA, retailers’ may be hesitant to deviate from the recommended price. Therefore, recommended prices which do not generally trigger competition law may function and raise similar concerns to fixed resale prices.
PA may substantiate the Commission’s concern that MFNs “work as a fixed resale price”. Whilst the competition concerns raised therefore broadly resemble those outlined above, it seems perverse that MFNs are not expressly prohibited under EU law as they arguably go further than RPMs in undermining competition. Notably, the minimum price set may be manipulated by fluctuating commission fees. Therefore, increased commission is likely to be presented through higher prices and passed onto consumers.
PA and Big Data
Staples, a customer of Boomerang in the US, expressed that implementing PA is no longer a competitive pressure, but an obligatory pre-requisite to compete on the market. However, this exposes a new competitive dimension, where companies compete to develop the most successful algorithms. The Commission’s decision in Google Shopping which established access to data as significant for competition to “improve the relevance of its results”, can be applied to the PA scenario, whereby data is characterised as ‘essential’ for most algorithms. Therefore, the playing field is further tilted as this intrinsic link between PA and big data is exposed. Consequently, firms like Amazon who have “always known they were competing on how well they understood their data” are arguably at an initial competitive advantage. Refuting this form of algorithmic competition disregards that large players with almost unlimited access to differentiated data points are undoubtedly in a better position. Although the increased market transparency suggests smaller players PA could essentially ‘copy’ larger players in a follow-the-leader strategy, data-driven barriers to entry remain for the initial development of a well-functioning PA. Industry-wide implementation of PA may contribute to the omnipotence of big-tech, expand the gap between online and brick-and-mortar retailers, inevitably lessening competition between market players.
Personalised Pricing (PP)
In light of this emerging ‘data arms race’, the concern has been raised that wide-spread use of PA will increase the prevalence of PP. Prima facie, it appears this “first degree price discrimination” which allows companies to “make a margin on people who don’t care” will result in supra-competitive prices or be utilised by dominant players to undercut competitors to tip the market in their favour. However, viewing PP through the lens of Erzachi & Stucke’s model of TC and PP, paints an entirely different picture. In contrast, this suggests PP could reinforce competition in markets susceptible to TC, introducing product differentiation on the factor of price via ‘secret discounts’ which effectively undermine the collusive price, in the consumers favour. Similarly, this reduces entry barriers for emerging players who can utilise this form of secret deal as a means to enter the competitive playing field. However, the CMA signified that the likelihood of PA and PP occurring simultaneously is highly unlikely – even where explicit co-ordination is present.
To conclude, the competitive advantage which may be obtained from implementing dynamic pricing has led to the widespread use of PA. The infamous case which brought this form of dynamic pricing to the attention of competition authorities could be characterised as simply an error. However, PA developed since demonstrate they can be manipulated discretely to reduce the hazards of competition, often simultaneously falling through legislative gaps. Coinciding with the CMA’s view that hub-and-spoke arrangements are likely to present the most immediate risk raises a dilemma for regulators if this is to be addressed via a new competition toolkit. Removing the ability to outsource PA to third-party vendors establishes barriers to entry for smaller players, and may contribute to the omnipotence of big-tech. Whilst curtailing the creation of excessive market transparency may function as a ‘quick-fix’ for most of the concerns raised throughout, competition authorities must ensure they don’t overstep the mark and limit potential benefits to consumer welfare. Although PA remain a relatively new phenomenon, the limited case law in this sphere may realistically be indicative of the success of PA in concealing anti-competitive behaviour. The characterisation of the competition concerns raised throughout as ‘futuristic’ and something competition authorities need not be alerted by, is a dangerous narrative. Not long ago, platforms reached an unforeseeable level of market power, if competition authorities don’t react promptly to limit these competition concerns via a new competition tool, they may find themselves playing a similar game of ‘catch-up’ in the case of PA.
This essay was written by Kaela Murie within the University of Strathclyde Law School for the module “Competition and the Digital Economy” coordinated by Dr Oles Andriychuk.