Christopher Gardner and Juan Londoño
Automated pricing tools are becoming increasingly ubiquitous in the modern economy as many businesses automate their pricing strategies to ensure their prices reflect market realities. Media attention, regulatory proposals, and a timely congressional hearing have largely focused on “surveillance pricing” and what we discussed as individualized dynamic pricing (IDP) in a previous blog. Despite fears of price gouging, these pricing methods have real efficiency benefits and are particularly effective at reducing waste.
However, there are other forms of automated tools in our day-to-day lives that do not necessarily rely on individuals’ data, as IDP does. This piece will focus on algorithmic pricing, understood as the general practice of automating the price-setting process, and the allegations that it might enable collusion amongst competitors. In the context of general algorithm-powered pricing, this alleged “collusion” typically refers to the use of automated tools to analyze market data to coordinate competitors’ prices. With a few notable exceptions, this is typical competitive behavior. For firms to compete on price, they must be able to track their competitors. Algorithmic pricing reduces the costs associated with monitoring and responding to market conditions. Taken alone, it is just using new technology to make an old process more efficient.
The Economics of Algorithmic Pricing
Firms often face difficulties in finding the “optimal” price for their goods. Economists like Armen Alchian have written extensively about the struggles of finding that “goldilocks” profit-maximizing price. Firms will oscillate between overpricing and underpricing in an attempt to track the optimal market price—a process that has traditionally been tremendously expensive, wasteful, and time-consuming. There is also a time constraint, as market fluctuations can quickly alter what is considered the optimal price.
Pricing a good optimally is also important for the consumer. It ensures that goods are distributed efficiently to those who value them most and prevents potential shortages, so consumers can actually buy them when they need them. It also sends price signals for the long run, allowing firms to decide where to invest and what markets to enter. This translates into a market that is constantly seeking to anticipate demand and attract consumers with lower prices and better products.
Algorithmic pricing helps speed up and automate the price discovery process by enabling firms to learn from and adjust to vast amounts of market data. These micro-level price shifts help ensure efficient market outcomes in both the short and long term. In the short term, accurately pricing a good enables the allocation of goods to those who value them the most. In the long term, accurate prices enable firms to plan effectively for the future and enhance competition by attracting more suppliers to enter the market. At the macro level, accurate pricing guides investment and development in our economy to their most productive uses. Put simply, prices are the single most valuable piece of information in the market; Algorithmic pricing makes this information more responsive and accurate.
However, there are concerns among policymakers that algorithmic pricing could facilitate otherwise unlawful collusion among competitors. Most of these concerns stem from two different scenarios. The first scenario is a form of “hub-and-spoke” collusion. In this scenario, competing firms can indirectly coordinate their prices through a third party. Policymakers worry that algorithmic pricing services may choose to act as that third party.
The second scenario is that, over time, multiple pricing algorithms might tacitly collude by independently converging on a super-competitive price that exceeds what the market would typically allow. The subtle but significant differences between these two scenarios give rise to two distinct economic analyses that help explain why (or why not) collusion can materialize through algorithmic pricing. This blog will break down the subtle yet significant differences between these two scenarios and their optimal policy solutions.
Will Algorithms Coordinate Competitors?
Policymakers and advocates concerned about the advent of algorithmic pricing tools claim that these tools can serve as intermediaries for competing businesses to collude. Their concerns stem from two claims: 1) When competing companies all contract the same third-party pricing tool, the tool could end up suggesting the same price to all of its customers, which will discourage firms from competing on price and increase prices for consumers. 2) These tools can become a medium for competing firms to share proprietary information with the algorithm, which then will be used by the algorithm in its pricing suggestion. Given access to this competitively sensitive information, the suggested price would mirror the price firms would have arrived at through an information-sharing cartel. This type of coordination is usually referred to as a “hub-and-spoke” model, in which the algorithm acts as a hub, coordinating the pricing decisions of its customers, known as the spokes.
The heightened attention to potential algorithm-based collusion stems from the belief that it can be a particularly harmful form of collusion. One could argue that this sort of collusion will naturally degrade as firms have an incentive to “cheat” and undercut their colluding competitors. However, as algorithmic pricing tools reduce monitoring costs, it is easier for cartel members to detect when another member cheats. Thus, in supply-constrained markets or markets with high barriers to entry, these tools can incentivize more companies to engage in anticompetitive practices.
This reasoning has led to lawsuits against the Rainmaker and RealPage tools, two algorithmic pricing tools used in the hospitality and housing industries, respectively. In their lawsuits, antitrust enforcers claim that these tools were intermediaries for hub-and-spoke collusion amongst seemingly competing companies in their respective industries. However, the results of these lawsuits also highlight that existing antitrust legislation provides adequate guardrails against potentially anticompetitive conduct and can evolve to offer targeted, narrow protections that would make additional legislation on algorithmic pricing unnecessary and redundant.
The Rainmaker case demonstrated that the use of the same pricing tool by competing businesses does not necessarily constitute collusive practices. On the contrary, judges have consistently found that the hospitality industry has used the tool to become even more competitive. As put by a judge, “rather than eliminating competition, pricing one’s hotel rooms in a manner calculated to maximize profits is how one competes.” As this tool foregoes any collection of proprietary data and instead relies on public, broad-market data to generate its pricing suggestions, it resembles an in-house market analyst rather than a collusion tool. By automating and outsourcing this pricing process, firms can better compete with other firms and predict demand fluctuations. This benefits potential competitors and incumbents alike, as resource-constrained competitors can contract the service rather than hire in-house analysts and see better returns. This reduces barriers to entry, which leads to lower prices, higher profits, and greater consumer welfare.
On the other hand, the RealPage case demonstrates that existing antitrust legislation is equipped to address concerns about information sharing. To generate its pricing suggestions, RealPage collected and processed competitively sensitive data from its customers, such as vacancy rates, profit margins, and other performance indicators. While the RealPage lawsuit ended in a settlement, RealPage will now face restrictions on how often its algorithms may suggest price increases and on the use of nonpublic customer data. For businesses that want to update their prices in real time using current market data, RealPage may use only publicly available data or create a bespoke suggestion using the business’s internal data.
However, if a business wishes to use non-public data, RealPage can use only data at least a year old and aggregated at the national level, barring the use of models that rely on regional, state, or local data. These restrictions aim to slow down the flow of information that could potentially foster collusion without the need for additional legislation.
The importance of non-public data in antitrust cases is further emphasized through a different Rainmaker case that was recently reversed and remanded, highlighting the use of non-public data in Atlantic City. Settlement agreements, however, are an imperfect regulatory mechanism, as they are binding only on the two parties involved at a specific point in time and lack the stickiness of legal precedent or legislative action. A subsequent administration could still pursue action for behavior deemed acceptable at the time of settlement. Nonetheless, they can work as a guideline for the algorithmic pricing industry, as other firms now know what behavior enforcers find sanctionable and will be deterred from engaging in it.
However, aside from enforcing current antitrust legislation, the most effective barrier to potential collusion is maintaining healthy, competitive markets. Algorithmic pricing tools, as their name indicates, are merely tools. They can only enable collusive behavior when the markets in which they are used already foster anticompetitive behavior. Going back to the RealPage example, competition in the housing market is often hampered by overly restrictive zoning laws that prevent competing landlords from entering the market. Even with reduced monitoring costs, highly competitive markets still create an incentive for the cartel to degrade, as there are still too many competitors to coordinate.
Will Competitors Use Algorithms to Collude?
The second fear is the potential for algorithms to independently converge on a super-competitive price that would be higher than what consumers would typically experience. Policymakers’ concerns stem from multiple studies finding that firms can use algorithms to raise prices. These studies can be misinterpreted to suggest that allowing widespread adoption of algorithmic pricing practices will, over time, result in collusive prices.
However, these experimental findings hold only under specific conditions, namely, high market concentration and a particular experimental structure. In other words, these studies found that the problem is not algorithmic pricing per se. Rather, very specific experimental conditions, particularly high market concentration and no firm entry and exit, will induce collusive behavior among algorithms. As these conditions were relaxed, collusive behavior decreased. The largest disruption in collusive behavior occurred when the experiment allowed new competitors to enter at random. Another study took this further, finding that reconfiguring experimental conditions to resemble a modern market led algorithms to become more competitive, driving down prices. These are highly technical studies designed to draw out theoretical patterns in market behavior under specific conditions. So, while they can be useful for understanding potential risks, they cannot be legitimately used to indict algorithmic pricing.
Beyond experimental contexts, algorithms can play a significant role in reducing barriers to entry. Algorithmic pricing can maximize profit by coordinating a firm’s prices with present and future demand. This type of knowledge was previously available only to the most entrenched incumbents, making it risky to enter a new market. An algorithm trained on past market data and allowed to experiment with pricing can model market demand and respond to competitors much faster than has historically been thought possible. Algorithmic pricing services also help level the competitive playing field between large and small companies. Developing modern pricing engines is a costly process that many small companies cannot afford. This means that larger companies gain a competitive advantage in pricing and price responsiveness that small companies simply cannot match.
Tacit collusion is not a new concept. Humans have been incentivized to do it for centuries. The core lessons that we have used to protect consumers and markets from collusion still hold. The greater the barriers to entry in a given industry, the higher prices must rise above the competitive level before a firm is willing to make the required investment to enter the market. As a rule of thumb, it is much more efficient to address trust concerns by reducing the barriers to entry to a given market rather than investing thousands of man-hours subjecting American companies to dubious investigations.
Conclusion
There are legitimate concerns over how algorithmic pricing could lead to anticompetitive behavior amongst firms. Algorithmic pricing is a dual-use technology in the sense that it can be both pro- and anti-competitive. It all depends on how the tool is used. This underscores the importance of maintaining competitive markets alongside rigorous antitrust enforcement. Each has a role to play, and together they serve as an effective protection against abusive pricing practices. However, it would be a mistake to label the use of pricing tools as collusion by default.
