Can advertising on Amazon create harms?*
The economics of information on Amazon
Ilan Strauss
Director of AI Commercialization Risks (Social Science Research Council, Brooklyn)�Honorary Research Fellow, UCL IIPP
31 May 2024
“Amazon’s Algorithmic Rents: The Economics of information on Amazon” (UC Law Science and Technology Journal, 2024)
with Tim O’Reilly and Mariana Mazzucato
Presentation to BECCLE Competition Law and Economics Conference
Contents
Motivation: Where is the harm?
Motivation: Why does Chicago ignore algorithms?��“Competition is Just a Click Away”
“Reduced information costs [online] also make it easier for businesses to provide and customers to seek out alternatives [...] Searching and switching are both easier and broader in online markets than on conventional markets. Customers can travel from one site to another with a mouse click. As a result, depending on a consumer’s location, the variety of sellers that are available online can be much greater than the variety that the brick-and-mortar world realistically permits. Price and product comparison can often be accomplished at little cost and almost instantly. [...] Monopoly is not realistically possible if buyers can costly [lessly] and quickly substitute to a different product. Switching costs are specific to the product [...] the fact that someone purchased dish detergent last month from a large online seller very likely has little or no bearing on where he will purchase it today.”
– Herbert Hovenkamp
Amazon’s Marketplace (third-party)�In economics, when does information (ads) leads to harms?�
From Structuralism to Institutionalism?
Structuralism: New Brandeis emphasizes market structure in shaping conduct and in turn the performance of a market (S-C-P)
But:
Interrogate market institutions online – algorithms?
From Structuralism to Institutionalism(?)
New Institutionalism (“NIE”): Institutions minimize transaction costs. More information solves this. Define the “new center” in U.S. antitrust.
(Old) Institutionalism: Institutions allocate resources and power, underpin behavioural patterns. Institutions process information.
Institutionalism online:
Does “consumer search”, as assumed in most economics models, exist?
Can information (ads) harm? History of information in economics
Traditional: Assume imperfect information. Advertising injects new information into the market. Reduces search costs.
Optimization: Consumers have the ability to overcome any informational issues in the market, while actual market institutions and mechanisms have little bearing on decision-making. “Competition is just a click away”.
�
Can information (ads) harm? History of information in economics
No imperfect information can’t harm consumers, they are rational
Chicago (Stigler, 1961):
Salop & Stiglitz
Yes – information can harm, even if consumers rational optimizer, are search costs:
“if information is costly [for consumers], each small firm obtains market power, and the equilibrium (if one exists) is characterized by prices above competitive levels and sometimes price dispersion as well. The relevant market structure with imperfect information is not perfect competition but rather monopolistic competition [and monopolistically competitive prices].”
Steve Salop, Information and Monopolistic competition, 66(2) The Am. Econ. Rev. 240 (1976).
Much ado about something?
Relies on high search & information costs. Assumes perfect rational processing of information.
“Informational power” 🡪 Eastman Kodak v. Image Technical Services, 1992
Digital Context
Are search costs high today or is information overwhelming?
Algorithm is perfect relative ranking. Is ads new information?
Can information (ads) harm?
Algorithmic Context�+ Information abundance�+ Satisficing humans�= Reliance on institutions
Information explosion which only algorithms can manage for us
Machines for information abundance
“In an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it….Filtering by intelligent programs is the main part of the answer.”
Herbert Simon, 1970
Information abundance (amidst imperfect humans) shapes competitive dynamics online
Competition along “informational” lines follow from attention scarcity <–> information abundance (+ plus time and cognitively constrained humans).
Platforms compete for user attention by curating abundant information, reducing an overwhelming choice (process) into one requiring little thinking.
Third-party firms compete within the platform, once retained, for user attention.
The “organic” algorithm and the positional mechanism
�The “constant against which competition occurs” is the organic algorithm.
The most relevant choices are presented to the user in a positional order.
The user saves time and cognition by using the positional heuristic presented by the algorithm 🡪 positional-driven click (or impression) behaviour.
Heuristics shape (evolving) behaviour
Content above the fold receives by far the highest share of the viewing time (57%).
What do Algorithms do?
What do Amazon’s Marketplace algorithms do?
1) Algorithms allocate value (attention between competing sellers).
Algorithms (search, feed, and recommendation*) allocate value through allocating attention. What is profit max on platform (does it set price or quantity)?
2) Algorithms process information (products + user data).
Online “aggregator” markets are informational. They provide “discovery”. Information is integral to the service. Harms are informational (quality or time).
What do Amazon’s algorithms do? [Continued]
3) Algorithms benchmark competition (conditions to gain user attention above the fold).
4) Algorithms shape consumer behaviour: a powerful heuristic device (“information processing shortcut”). This makes advertising very powerful and effective online.
Algorithms as Institutions
Algorithms create markets and make allocations
Herbert Simon on Institutions (Administrative Behaviour, 1947)
Findings and Policy Implications
Key Findings
The level of information and the level of competition are closely linked online (Hovenkamp & Areeda, ¶2023 : “output consists of everything in the product package, including the information that a competitive market would ordinarily provide and that is necessary for a consumer to determine willingness to pay”.)
Advertising results displacing prominent organic results serves as the primary algorithmic mechanism degrading information quality on Amazon’s Marketplace today. ��Advertising drives the relative prominence of Amazon’s search results and represents a significant additional platform cost for users, manifesting in higher search costs, reduced product quality and choice, less relevant results, and higher prices. ��Advertising is how Amazon extracts higher fees from its third-party firms, compelling them to pay for prominence, unable to earn it competitively.
Implications 1: Competition Concepts
�Dominance: A platform can make attention allocations independently of consumer preferences, information relevance internally, and information on competing platforms.
Market power: degrade quality of information profitability.
Price or profit benchmark vs organic benchmark?
Implications 2: Competitive Information Provision on Amazon’s Marketplace. �
A reasonable level and type of information which a competitive market should provide.
But is this related to market power?
Implications 3: A consumer approach�
End
This research is part of UCL IIPP’s, Omidyar Network funded, wider investigation into value creation and value extraction by large technology companies.
31 May, 2024
Why this Matters in the Age of AI
1) Existing social media, search, and recommendation platforms is where the money is (eyeballs).
1.1 AI is being monetized on these existing platforms.
1.2 Expanding productivity likely to lead to dramatic increase in leisure activities.
2) AI as large language models (ChatGPT) is not profitable yet. To make profits they will have to degrade quality.
3) A competition approach likely to face similar difficulties (lack of competitors) and similar recommendations (break down walled gardens).