Yes, Retailers Are Colluding to Inflate Prices Online

From Fast Company:

Have you ever searched for a product online in the morning and gone back to look at it again in the evening only to find the price has changed? In which case you may have been subject to the retailer’s pricing algorithm.

Traditionally when deciding the price of a product, marketers consider its value to the buyer and how much similar products cost, and establish if potential buyers are sensitive to changes in price. But in today’s technologically driven marketplace, things have changed. Pricing algorithms are most often conducting these activities and setting the price of products within the digital environment. What’s more, these algorithms may effectively be colluding in a way that’s bad for consumers.

Originally, online shopping was hailed as a benefit to consumers because it allowed them to easily compare prices. The increase in competition this would cause (along with the growing number of retailers) would also force prices down. But what are known as revenue management pricing systems have allowed online retailers to use market data to predict demand and set prices accordingly to maximize profit.

These systems have been exceptionally popular within the hospitality and tourism industry, particularly because hotels have fixed costs, perishable inventory (food that needs to be eaten before it goes off), and fluctuating levels of demand. In most cases, revenue management systems allow hotels to quickly and accurately calculate ideal room rates using sophisticated algorithms, past performance data and current market data. Room rates can then be easily adjusted everywhere they’re advertised.

. . . .

These revenue management systems have led to the term “dynamic pricing.” This refers to online providers’ ability to instantly alter the price of goods or services in response to the slightest shifts in supply and demand, whether it’s an unpopular product in a full warehouse or an Uber ride during a late-night surge.

. . . .

However, new algorithmic pricing programs are becoming far more sophisticated than the original revenue management systems because of developments in artificial intelligence. Humans still played an important role in revenue management systems by analyzing the collected data and making the final decision about prices. But algorithmic pricing systems largely work by themselves.

. . . .

The algorithms study the activity of online shops to learn the economic dynamics of the marketplace (how products are priced, normal consumption patterns, levels of supply and demand). But they can also unintentionally “talk” to other pricing programs by constantly watching the price points of other sellers in order to learn what works in the marketplace.

These algorithms are not necessarily programmed to monitor other algorithms in this way. But they learn that it’s the best thing to do to reach their goal of maximizing profit. This results in an unintended collusion of pricing, where prices are set within a very close boundary of each other. If one firm raises prices, competitor systems will immediately respond by raising theirs, creating a colluded non-competitive market.

Monitoring the prices of competitors and reacting to price changes is normal and legal activity for businesses. But algorithmic pricing systems can take things a step further by setting prices above where they would otherwise be in a competitive market because they are all operating in the same way to maximize profits.

This might be good from the perspective of companies, but is a problem for consumers who have to pay the same everywhere they go, even if prices could be lower. Non-competitive markets also result in less innovation, lower productivity and, ultimately, less economic growth.

. . . .

The European Commission has warned that the widespread use of pricing algorithms in e-commerce could result in artificially high prices throughout the marketplace, and the software should be built in a way that doesn’t allow it to collude.

Link to the rest at Fast Company

In the US, price-fixing is illegal under U.S. antitrust laws.

From The Federal Trade Commission:

Price fixing is an agreement (written, verbal, or inferred from conduct) among competitors that raises, lowers, or stabilizes prices or competitive terms. Generally, the antitrust laws require that each company establish prices and other terms on its own, without agreeing with a competitor. When consumers make choices about what products and services to buy, they expect that the price has been determined freely on the basis of supply and demand, not by an agreement among competitors. When competitors agree to restrict competition, the result is often higher prices. Accordingly, price fixing is a major concern of government antitrust enforcement.

A plain agreement among competitors to fix prices is almost always illegal, whether prices are fixed at a minimum, maximum, or within some range. Illegal price fixing occurs whenever two or more competitors agree to take actions that have the effect of raising, lowering or stabilizing the price of any product or service without any legitimate justification. Price-fixing schemes are often worked out in secret and can be hard to uncover, but an agreement can be discovered from “circumstantial” evidence. For example, if direct competitors have a pattern of unexplained identical contract terms or price behavior together with other factors (such as the lack of legitimate business explanation), unlawful price fixing may be the reason. Invitations to coordinate prices also can raise concerns, as when one competitor announces publicly that it is willing to end a price war if its rival is willing to do the same, and the terms are so specific that competitors may view this as an offer to set prices jointly.

Not all price similarities, or price changes that occur at the same time, are the result of price fixing. On the contrary, they often result from normal market conditions. For example, prices of commodities such as wheat are often identical because the products are virtually identical, and the prices that farmers charge all rise and fall together without any agreement among them. If a drought causes the supply of wheat to decline, the price to all affected farmers will increase. An increase in consumer demand can also cause uniformly high prices for a product in limited supply.

. . . .

Antitrust scrutiny may occur when competitors discuss the following topics:

  • Present or future prices
  • Pricing policies
  • Promotions
  • Bids
  • Costs
  • Capacity
  • Terms or conditions of sale, including credit terms
  • Discounts
  • Identity of customers
  • Allocation of customers or sales areas
  • Production quotas
  • R&D plans

A defendant is allowed to argue that there was no agreement, but if the government or a private party proves a plain price-fixing agreement, there is no defense to it. Defendants may not justify their behavior by arguing that the prices were reasonable to consumers, were necessary to avoid cut-throat competition, or stimulated competition.

. . . .

Q: The gasoline stations in my area have increased their prices the same amount and at the same time. Is that price fixing?

A: A uniform, simultaneous price change could be the result of price fixing, but it could also be the result of independent business responses to the same market conditions. For example, if conditions in the international oil market cause an increase in the price of crude oil, this could lead to an increase in the wholesale price of gasoline. Local gasoline stations may respond to higher wholesale gasoline prices by increasing their prices to cover these higher costs. Other market forces, such as publicly posting current prices (as is common with most gasoline stations), encourages suppliers to adjust their own prices quickly in order not to lose sales. If there is evidence that the gasoline station operators talked to each other about increasing prices and agreed on a common pricing plan, however, that may be an antitrust violation.

Q: Our company monitors competitors’ ads, and we sometimes offer to match special discounts or sales incentives for consumers. Is this a problem?

A: No. Matching competitors’ pricing may be good business, and occurs often in highly competitive markets. Each company is free to set its own prices, and it may charge the same price as its competitors as long as the decision was not based on any agreement or coordination with a competitor.

Link to the rest at The Federal Trade Commission

Price fixing is illegal whether competitors set minimum or maximum prices or establish a range of prices within which they will price their goods.

One of the key elements of illegal price-fixing is an agreement (written, verbal, or inferred from conduct) among competitors. A third party that mediates, organizes or facilitates price-fixing among competitors is also guilty of price fixing. (See, for example, Apple and a group of major publishers agreeing to fix prices on ebooks and force Amazon to increase its ebook prices, in PG’s indescribably humble opinion, one of the more inept attempts at price fixing in the hundred-plus years that the practice has been outlawed in the U.S.).

The OP raises an interesting question about whether pricing systems executed by computers using artificial intelligence constitute illegal price fixing.

Under present law, it is clear that price-fixing agreements established among competitors through a third party are illegal and, per Apple and other cases, the third party is also chargeable with price-fixing. If each competitor appoints a third party and the third parties agree to fix prices or set up a system for establishing uniform prices, PG believes that’s also a slam-dunk price-fixing violation.

The issue of whether artificial intelligence systems that look at the same market data and set prices in a similar manner are engaged in price-fixing is very interesting.

Competitors who each look at market, pricing and available competitor data without using artificial intelligence and set the same prices are not guilty of price-fixing so long as there is no agreement between them to fix prices. Competitor A can look at the prices being charged by Competitor B and use that information to adjust its prices. As described in the OP, that’s how many gas stations typically set prices within a given geographic area.

In the gas station illustration, each station is sending pricing signals to the general public, including other gas stations.

If gas station A reduces its price, other gas stations may respond by matching the price cut, cutting prices below those of A as a competitive move, or leaving prices higher than A and banking on other competitive advantages – a more convenient location or better prices on Diet Coke, for example – to offset A’s pricing advantages.

Not matching a price cut represents a temporary strategy, however, because, based on its own decision factors, a competing station can adjust its prices at any time if it perceives its pricing strategy is less than optimum.

Going back to the OP, PG doesn’t see that AI systems watching the prices other AI systems are setting constitutes illegal collusion. If the AI systems somehow communicated with each other and simultaneously increased or dropped prices, the owners of those systems might be guilty of price-fixing.

However, in the absence of some sort of connection beyond closely watching the public pricing activities of competitors, PG doesn’t see any sort of illegal collusion or conspiracy to fix prices. Setting prices to maximize profits is not, by itself, a violation of any law of which PG is aware. It’s a fundamental principle of capitalist economies.

Back to the gas station example – If two gas stations are located across the street from each other and each station assigns an employee to watch the posted prices of the other station and immediately change prices whenever the station across the street changes its prices, that’s not an illegal price-fixing agreement between the two stations.



11 thoughts on “Yes, Retailers Are Colluding to Inflate Prices Online”

  1. I must be missing something…

    . If one firm raises prices, competitor systems will immediately respond by raising theirs, creating a colluded non-competitive market.

    except that if a price is raised to match that of a competitor, the merchant forgoes a competitive advantage. I thought the invisible hand of the market deals with such things.

    • This can happen if everyone raises their price to the same level, but only when demand is “inelastic” – i.e., whatever the item or service costs, people will buy the same amount of it (or the consumption doesn’t change very much). Then the higher price equates to a higher profit.

      The so-called “classic” example of this is actually gasoline. Although heating fuel is a better example – consumption is far more inelastic there.

      But the fact is that even those are not truly inelastic, they simply have a much longer lag between price increase and consumption decrease. It takes more time for people to buy a more fuel-efficient car, convince their government to add bus routes, switch from oil to natural gas, add insulation to their home, etc., etc., etc.

  2. PG, What if competitors use copies of the same AI? If the behavior of competitors’ software is indistinguishable from collusion, then does that constitute illegal action?

    • Better yet, what if the AIs are different but reach the same endpoints?

      Colluding computers? Good luck with that argument.

      In the baseball world there is a slowburn firestorm building because over the past few years teams have stopped giving long term deals to “older” (over-30) players and in many cases only offering up one year deals. (Unlike other sports, baseball contracts are typically guaranteed, not conditional). Union advocates find this gradual change “suspicious”.

      On the other hand, baseball analysts and statisticians have been pointing out for a couple decades that 90+ percent of extra long contracts and 9-figure contracts end up in heartburn for the team with non-performing players eating up roster and payroll, limiting team competitiveness. There is extensive public data that confirms that all players perform along an age-based curve so that given a few points of data early in the player’s career future performance can be calculated.

      For years, teams ignored the accumulated data.
      Then a few teams didn’t. And they prospered.
      Now it seems all 30 teams are working off the same data and similar curves. Players in their early 30’s are reaching free agency dreaming of seven year deals and getting offers for 3 or 4, expecting 3 and getting 1.

      Teams are still following different roster building strategies but the contracts look suspiciously similar for similar players.

      Collusion of just rational behavior based on common (public) data?

      Odds are labor strife and a strike is coming once the current collective agreement expires, even though the fraction of game revenues is remaining constant and higher than other pro sports.

      And that is a simple case.
      Now try to argue collusion between thousands of retailers using dozens of computer systems and private data. And watch the other side utter two magic words:

      “emergent behavior”

      • Felix, You chose an excellent example. Kudos.

        By now, all the MLB owners have seen Moneyball and checked out the book behind the movie and the math behind the book. I have heard that every farm system has been using the new stats for years. That must percolate up to the majors. It may not matter that the clubs use different software. (I doubt that their ‘software’ is more sophisticated than Excel.) What matters are the algorithms they use. Likely those are identical. So they yield identical results.

        Is this theory colorable? Is it good enough to get a case into court and survive a MSJ?

        Do you know a judge who would understand ’emergent behavior’? Would you like to explain it to a jury?

        • Actually, the teams have very sophisticated analysis systems that go well beyond the public ones. By now every team has an analytics system working 24×7 to process the masses of data coming through THE TRACKMAN system, which is a radar and high-speed video array of sensors in every MLB park.

          It is sensitive enough that it can measure in realtime the spin rate of a baseball. It can measure the initial acceleration of running players as well as peak speed; it can measure a batted-ball launch angle and exit velocity.

          They are slowly expanding the system to minor league parks and (surprise of surprises!) the eight ballparks of the independent Atlantic League. Last night they announced a deal to test “robot” umpiring in the Atlantic league for three years using the system to call balls and strikes.

          The Union has access to the raw data but apparently they haven’t been doing much with it.

          As for explaining emergent behavior to a court, it shouldn’t be hard. (Worst case scenario, the judge will appoint a special master.) The obvious example is flash crowds.

      • Any two traders can use the same trading system, yet each system can perform very differently. This is because each trader trains the system on a set of trading history, and trains at different intervals. That means the experience of the systems immediately diverge.

        After installation and initial training, no two systems have developed the same network. As a simple example, one instance of the system may have learned to optimize using a simple 10 day moving average. Another may have learned to optimize using an exponential moving average. (It’s more complex than that.)

        A gross example is identical twins separated at birth. They learn different languages, and have different life experiences.

        But, if I have a winning system, won’t other people buy it? Sure, but if I have that winning system, why on earth would I sell it to the rest of those guys? Better to just sit back and take their money.

    • Antares – If competitors agree to use the same AI, there might be a beginning of a case, but if they just acquire the AI because it’s the best one available, I don’t think there is an agreement.

      I think it’s analogous to competitors watching each other’s public pricing and promotional behavior and copying where they decide it’s a good idea.

      Good comments, as usual, by antares and Felix.

  3. This describes markets moving to an open auction system. Each seller sees what every other seller does, and adjusts his own behavior in response. It’s the result of better information.

    There is no reason to think consumers are being hurt. Markets facilitate trade by discovering price. That’s what’s happening here.

    Consider the days of the open outcry commodities markets. (Pause for a moment of silence.) Every trader in the pit saw what every other trader was bidding and offering. All adjusted their own bids and offers. The guy in the pulpit adjacent to the pit wore a headset and told the guy in the backroom what price to send out.

    The prices hit the wires in about five seconds, and the whole world adjusted their own prices. Nobody could fix prices because someone would always jump on a quarter cent profit and screw it up.

    God Bless the free market, for the pits are silent, but not forgotten.

    • The commodities pits at the Chicago Mercantile Exchange were a great show, T., a bunch of sweaty guys (it was all guys until the last few years, sweating and jammed together, yelling, bumping, using obscure hand signals) creating a flow of billions of dollars in cattle, hogs, sheep, corn, wheat, dry whey, cheese, lumber, etc., etc.

      Just a few blocks away, at the Chicago Board of Trade, traders used similar systems to trade options and futures contracts.

  4. > only to find the price has changed?

    Amazon has been doing that for at least five years, to my personal knowledge.

    I searched for a book, decided to delay the purchase for a while, and noticed the price creeping up steadily over the next two weeks as I cycled through tabs in my browser. It was somehting like 5x times the original price when I decided I didn’t need the book after all and closed the tab.

Comments are closed.

Secured By miniOrange