💻 computer science

A Generative Model of Conspicuous Consumption and Status Signaling

This paper proposes and validates a computational theory demonstrating that status symbols and conspicuous consumption emerge endogenously through social observation and predictive pattern completion in LLM-based agent simulations, effectively bridging micro-level cognition with macro-level economic phenomena like Veblen effects and subculture formation.

Original authors: Logan Cross, Jordi Grau-Moya, William A. Cunningham, Alexander Sasha Vezhnevets, Joel Z. Leibo

Published 2026-03-16

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Logan Cross, Jordi Grau-Moya, William A. Cunningham, Alexander Sasha Vezhnevets, Joel Z. Leibo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: Why Do We Care About "Stuff"?

Imagine you are at a party. You see someone wearing a very expensive, obscure jacket. You don't know if the jacket is actually warm or comfortable. But you think, "Wow, they must be cool, rich, or part of a special club."

This is Status Signaling. Humans have always done this to show off wealth, fitness, or group membership. But for a long time, scientists couldn't explain how we decide what is cool.

Old Theory: "It's cool because it's expensive and hard to fake." (Like a peacock's tail: only a strong bird can grow a big, heavy tail).
The Problem: This doesn't explain why cheap things (like a specific slang word or a weird meme) can become status symbols, or why expensive things suddenly become "uncool" overnight.

This paper proposes a new theory: Status isn't about the price tag; it's about social copying. We decide what is "appropriate" by watching what others do and then doing it too. It's a feedback loop of observation and imitation.

The Experiment: The "Digital Town"

To test this, the researchers didn't use real humans (which is hard to control). Instead, they built a Digital Town using 50 AI agents (robots powered by Large Language Models, like the ones that write this text).

Think of these AI agents as digital actors living in a simulation. They have:

Money: Some are rich, some are poor.
A Marketplace: They can buy food, gadgets, and clothes.
A Social Life: They go on "first dates" where they chat and see what the other person is wearing.

The researchers ran two versions of this town:

Town A (The Lonely Town): Agents buy things and go home. They never talk to anyone.
Town B (The Party Town): Agents buy things, go on dates, see what others are wearing, and chat about it.

What Happened? (The Magic Results)

1. The "Labubu" Phenomenon

In the real world, a toy called "Labubu" became a massive global trend recently. The AI agents had never heard of it before (it was new after their training data ended).

In the Lonely Town: Nobody cared about Labubu. It was just a plastic doll.
In the Party Town: One agent bought a Labubu. Their date saw it, said "Nice!", and then everyone started wanting one.
The Result: The price of the Labubu skyrocketed. Even though it was just a toy, the agents treated it like a luxury gold bar. This is called a Veblen Good: a product that becomes more desirable the more expensive it gets.

2. The "Price vs. Hype" Mystery

Usually, economists think: "People buy it because it's expensive (signaling wealth)."
The researchers tested this by freezing the prices. They told the sellers: "You cannot raise the price, no matter what."

The Surprise: Even with fixed, low prices, the agents in the Party Town bought even more status goods!
The Lesson: It wasn't the high price that made them want it. It was the social hype. They wanted it because they saw their friends having it and talking about it. The high price in the real world is just a side effect of everyone wanting it, not the cause.

3. The "Influencer" Effect

The researchers added "Influencer" agents who told everyone, "Chanel bags are so last year! Wear this weird vintage camera instead!"

The Result: The agents immediately stopped caring about the expensive bags and went crazy for the vintage cameras.
The Lesson: Status symbols are fragile. If the "cool kids" change their minds, the whole market shifts instantly. This explains why fashion trends change so fast.

4. It's Not Just About Money

They also tested non-money things, like:

Political Outrage: In the Lonely Town, agents ignored political posts. In the Party Town, they started reposting angry political videos to show their friends they were "good people."
Coffee Choices: Agents in the "LA" simulation started drinking Oat Milk because their friends did. Agents in the "Kerala, India" simulation stuck to Cow Milk because of their cultural background, even when social pressure tried to change them.
The Lesson: We copy behaviors to fit in, whether it's buying a bag or posting a tweet.

The Core Mechanism: "Predictive Pattern Completion"

How do the AI agents know what to do? They use a process the authors call Predictive Pattern Completion.

The Analogy: The "Vibe Check"
Imagine you walk into a room. You don't have a rulebook that says "Wear a suit." Instead, you look around. You see everyone wearing suits. Your brain says, "Okay, the pattern here is 'suits.' If I want to fit in, I should probably wear a suit too."

The AI agents do this constantly:

Observe: "My date is wearing a Rolex."
Predict: "People like us wear Rolexes. It's appropriate."
Act: "I should buy a Rolex."
Repeat: Now the next person sees the Rolex and copies it.

This creates a feedback loop. The "rules of the game" aren't written down; they are written by the players themselves as they interact.

Why Does This Matter?

This paper is a big deal because it bridges the gap between micro (individual thoughts) and macro (big economic trends).

Old View: Humans are rational robots who calculate costs and benefits.
New View: Humans are cultural learners. We are like a giant, living generative model. We watch each other, update our "vibe," and collectively decide what is valuable.

The Takeaway:
Status isn't about the object itself. A $500 bag isn't "cool" because of the leather; it's cool because we all agreed, through a million tiny conversations and observations, that it is cool. If the conversation stops, the value disappears.

The researchers have built a "time machine" for culture. They can now simulate how viral trends start, how subcultures form, and why we suddenly stop liking things we used to love—all by watching digital agents go on digital dates.

1. Problem Statement

The paper addresses a fundamental gap in social science and economics: the generative mechanism of status signaling.

The Puzzle: While status signaling drives human behavior (resource allocation, mating, group cohesion), existing theories struggle to explain how specific goods or behaviors acquire prestige, why meanings shift dynamically over time, and why "tipping points" occur.
Limitations of Classical Frameworks:
- Rational Choice Theory: Fails to explain upward-sloping demand curves (Veblen effects) where demand increases as price rises.
- Costly Signaling Theory (CST): Posits that high cost ensures signal honesty. However, CST treats preferences as fixed and exogenous. It cannot explain why costless signals (e.g., slang, accents) work, why expensive signals lose value when adopted by the "wrong" groups, or how populations select which specific arbitrary symbol to converge upon (equilibrium selection).
The Core Question: How do cultural conventions emerge endogenously from micro-level interactions without pre-defined utility functions?

2. Theoretical Framework: Predictive Pattern Completion

The authors propose a new computational theory grounded in the Theory of Appropriateness and Predictive Pattern Completion.

Core Hypothesis: Status symbols are not fixed utility maximizers but emerge through a feedback loop of social observation and imitation. Agents act as "culturally trained generative models."
Mechanism:
- Agents possess a Habitus (an intuitive sense of "rules of the game") derived from cultural priors and in-context memories.
- Behavior is generated via predictive pattern completion: Agents sample actions from a probability distribution conditioned on their social context and the "weight of precedent" (memories of others' past actions).
- Conjecture 1: Social exposure increases the probability of status signaling ( $p(a_s|c, m_s) > p(a_s|c)$ ) because observing others updates the agent's priors regarding what is "appropriate" for their group.
Shift from CST: Instead of cost being the primary driver, semiotic appropriateness (how observers interpret a signal given context, role, and precedent) is the binding constraint.

3. Methodology

The study utilizes Concordia, a multi-agent simulation framework powered by Large Language Models (LLMs).

Environment:
- Agents: 50 unique personas (Los Angeles-based) with heterogeneous wealth distributions and formative memories.
- LLM Backbone: Gemma-3-27B.
- Architecture: Agents operate on a "logic of appropriateness" using three prompts: "What kind of situation is this?", "What kind of person am I?", and "What does a person like me do?"
Experimental Design:
- Duration: 5 simulated days.
- Phases:
  1. Marketplace: Agents buy/sell goods (Food, Gadgets, Clothing, Accessories) across Low/Mid/High tiers.
  2. Daily Life: Private events (e.g., work, hunger).
  3. Social Interaction (Key Manipulation): Agents engage in "first dates" (80 turns of dialogue) where they observe their partner's outfit/accessories.
Conditions:
1. Social Life: Marketplace + Daily Life + First Dates (Social visibility enabled).
2. No Social Life (Control): Marketplace + Daily Life only (No social observation).
3. Fixed Price Control: Social Life enabled, but prices are pinned to baseline to test if demand drives price or vice versa.
4. Synthetic Goods: Using procedurally generated brands (e.g., "Serrurier Juliette") to rule out pre-training bias.
5. Non-Monetary Signaling: Scenarios involving political virtue signaling, conspicuous altruism, and arbitrary conventions (e.g., coffee milk choice, banner colors).
6. Influencer Intervention: Confederates introduced to sanction mainstream goods and promote subcultures (Hipster vs. Streetwear).

4. Key Results

A. Emergence of Conspicuous Consumption

Demand Shift: In the Social Life condition, 35% of purchases were for high-tier status goods, compared to 21% in the No Social Life condition ( $p < 0.001$ ).
Price Run-Ups: Social interaction caused aggressive upward shifts in bid-ask bands for luxury items (Chanel, Rolex) and trending collectibles (Labubu).
Veblen Effects: The Pop Mart Labubu doll exhibited positive price elasticity (+1.45) in the Social condition. As prices rose, demand increased, a phenomenon absent in the control group.
Synthetic Validation: The effect replicated with synthetic brands, proving the behavior is emergent and not a retrieval of pre-trained brand prestige.

B. Causality: Hype Drives Price

Fixed Price Control: When prices were pinned, demand for status goods increased (45% of bids) compared to the dynamic price condition (31%).
Conclusion: This resolves the identification problem. Social diffusion (Hype) $\rightarrow$ Demand $\rightarrow$ Price. The price increase is an epiphenomenon of socially constructed demand, not the causal driver of the desire.

C. Mechanisms of Diffusion

Conversation Analysis: Agents discussed fashion in 85.6% of dates. However, the dialogue was not superficial; it centered on identity, authenticity, and critique of consumerism.
Subculture Formation: "Influencer" agents successfully drove the formation of distinct subcultures (e.g., Hipster vs. Streetwear) through targeted sanctioning. Interestingly, mainstream luxury goods (Chanel) retained value despite negative sanctions, indicating path dependence.

D. Generalization to Non-Monetary Signaling

Virtue Signaling: Social exposure increased the sharing of "moral outrage" content (42% vs. 29% in control).
Conspicuous Altruism: Public donations increased from 26% to 35% with social visibility.
Path Dependence & Cultural Priors:
- Los Angeles Agents: Converged on "Green" banners and "Oat/Black" coffee.
- Kerala Agents: Converged on "Green" banners and "Cow Milk" coffee.
- Finding: Social influence amplifies latent cultural priors. The specific equilibrium reached is path-dependent on initial cultural conditions.

5. Key Contributions

New Computational Theory: Proposes a unified model of status signaling based on predictive pattern completion and the theory of appropriateness, moving beyond static utility functions and Costly Signaling Theory.
Methodological Innovation: Demonstrates the use of LLM-based generative agents to simulate endogenous preference formation. This allows for the isolation of social visibility as a causal variable in a controlled, ecologically valid environment.
Empirical Validation: Successfully reproduces complex economic anomalies (Veblen effects, positive price elasticity) and sociological phenomena (subculture formation, path-dependent conventions) that purely rational models cannot explain.
Generative Bridge: Connects micro-level cognition (individual memory and imitation) to macro-level economic and sociological outcomes, offering a tool for forecasting cultural trends.

6. Significance

Theoretical Impact: Challenges the dominance of rational-actor models in economics and sociology by showing that "irrational" behaviors (like Veblen goods) are actually rational responses to a socially constructed environment where "appropriateness" is the utility function.
Forecasting Capability: Provides a framework to move from descriptive sociology to generative, predictive science. It allows researchers to simulate how viral trends, consumer tastes, and cultural codes emerge from interaction.
Limitations & Future Work: The authors acknowledge that LLMs have training biases and that the simulation timescale (5 days) is accelerated compared to real-world cultural evolution (months/years). However, the ability to generate novel trends (like the Labubu hype post-training cutoff) suggests strong generalization potential.

In summary, the paper argues that status is not inherent in objects but is generated through a feedback loop of social observation and predictive imitation. By simulating this loop with LLM agents, the authors provide a robust computational explanation for how cultural conventions emerge, shift, and drive economic behavior.