Applying SAPM to algorithmic pricing coordination

Algorithmic pricing converts 40.0B in private gain into 215.2B in system cost

The paper applies SAPM to U.S. algorithmic pricing across rental housing, hotels, and airlines. The resulting system welfare beta is β_W = 5.38 [90% CI: 3.9-7.5].

• Π = $40.0B/yr in excess producer surplus from algorithmic coordination
• W = $215.2B/yr in annualized welfare cost across five calibrated channels
• Π^SA = -$175.2B/yr after subtracting system cost from private payoff

SAPM prices coordinated pricing by welfare damage per dollar of industry revenue

The CAPM analogy is direct: market beta becomes welfare beta, and private payoff is measured against the competitive-market baseline.

• W₀ = $684B/yr in affected sectoral revenue at the competitive baseline
• β_W = 5.38 means 5.38 in system cost for each 1.00 of private gain
• μ* = 4.38 per 1 of private payoff, the break-even social value needed to offset damage

The $40.0B payoff comes from rent software, hotel RMS platforms, and airline yield systems

The calibration starts with direct excess revenue from algorithmic coordination, then adds system costs that standard antitrust damages do not count.

• Rental housing: 8.7B/yr, including 3.8B tied to RealPage excess rent cited by the White House CEA
• Hotels: 19.7B/yr from 4-7% RevPAR uplift in a 263B market using systems such as IDeaS and Duetto
• Airlines: $11.1B/yr from 4-5% yield management uplift through systems such as PROS, Sabre, and Amadeus

RealPage is the paper's clearest hub-and-spoke coordination case

The mechanism is not a vague algorithmic effect. YieldStar ingests non-public competitor data, returns rent recommendations, and participating landlords largely follow them.

• 12.4 million U.S. rental households receive rent figures generated or influenced by third-party revenue management software
• DOJ filings report 85% subscriber compliance within a 5% margin
• The software advises users to accept higher vacancy when higher rents maximize revenue yield

The β_W estimate adds five costs that sector damages leave out

SAPM treats algorithmic pricing as a system architecture with direct, behavioral, distributional, technical, and governance effects.

• Primary externality: coordinated price elevation transfers consumer surplus to producers
• Secondary cost: suppressed output and vacancy create deadweight loss
• Systemic cost: rent increases concentrate on 10.8 million cost-burdened renter households
• Indirect harm: reinforcement-learning feedback loops test higher prices and lock in new floors
• Governance cost: enforcement, settlements, lobbying, and regulatory-capture spending are counted inside W

The feedback channel uses evidence from Q-learning and live gasoline markets

The paper separates hub-and-spoke data sharing from algorithmic convergence. Both can produce supra-competitive prices without explicit human price fixing.

• Calvano et al. (2020): Q-learning agents discover collusive equilibria autonomously
• Assad et al. (2024): German gasoline margins rise 9% with single algorithm adoption and 28% with dual adoption
• Fish et al. (2025): LLM pricing agents reach supra-competitive equilibria through price-war avoidance

100,000 draws do not find a welfare-positive version of the current architecture

The robustness check varies calibration parameters but keeps the current algorithmic pricing architecture in place.

• Median β_W = 5.38 with 90% CI [3.9, 7.5]
• P(β_W < 1) = 0.0000%
• κ = 0.814, so 81.4% of total system cost exceeds the private gain

The harm depends on adoption density and data architecture, not only transaction volume

The paper argues that standard externality tools misprice algorithmic pricing because the damage is nonlinear and the affected parties cannot bargain over the coordination mechanism.

• Pigouvian taxes miss the adoption-density effect shown by 9% versus 28% margin increases in gasoline markets
• Coasean bargaining fails when 44 million renter households face centralized software used by competing landlords
• Rights assignment is endogenous because lobbying frames competitor-data pooling as pricing innovation

Settlements and lobbying are part of the welfare cost, not background politics

The paper counts resource spending used to preserve or police the architecture as an endogenous cost of algorithmic pricing coordination.

• $141.8M in RealPage MDL class-action settlements from 27 landlords
• RealPage lobbying rose from 4.8M in 2020 to 9M in 2024
• NMHC spent 6.8M on lobbying and 3.4M in campaign donations in 2022
• Thoma Bravo retained Ballard Partners for $200,000 in 2025 to lobby DOJ directly

The paper classifies algorithmic pricing harm as reversible by design reform

Unlike production externalities embedded in physics, this welfare cost depends on the pricing system's institutional architecture.

• The DOJ's November 2025 RealPage consent decree targets the real-time competitor-data feed
• Symmetric price governors are proposed to stop asymmetric price ratcheting
• California AB 325 and New York's algorithmic rent ban are treated as reform existence proofs

The remedies target competitor data, price ratchets, and retained gains

Because the coordination mechanism is architectural, the paper's policy proposals focus on changing what the software can observe, recommend, and profit from.

• Runtime data bans sever non-public competitor price and occupancy feeds
• Symmetric governors limit one-way upward exploration by pricing systems
• Algorithmic disgorgement recovers gains from coordinated surplus extraction
• Per se prohibition is reserved for architectures that reproduce hub-and-spoke coordination