Which rules soften the harm without destroying the gains?
Ten interventions, each defined as an exact change to what the seller may observe, how it may price, or how consumers respond. Select the seller they apply to.
How does each intervention move consumer and producer surplus?
Change relative to no intervention for the selected seller; 95% Monte Carlo intervals.
Which interventions actually close the group price gap?
Absolute price gap between the demographic-proxy group and the rest, and the Gini of consumer surplus.
Privacy vs welfare
Data-exposure index (share of sensitive signals used × share of consumers exposed) against total and consumer surplus.
The policy frontier: welfare vs inequality as the dial turns
Parameter sweeps of price caps, personalization bands, parity tolerances, audit probabilities and opt-out take-up against an individualized boosting seller. Point labels are the parameter value.
Definitions
- Disclosure
- Consumers are told prices are personalized. Behavioural response only: 15% of non-strategic consumers become strategic hold-outs and 10% become privacy-aware (mask signals at a small utility cost).
- Explanation
- Disclosure plus the seller must explain which signals raise prices. Above-median-valuation consumers manipulate browsing and urgency signals to the 10th percentile with probability 0.5. Third-party profile and segment signals cannot be manipulated.
- Opt-out
- Zero-cost masking of behavioural and profile signals. Privacy-aware consumers always opt out; above-median-valuation consumers opt out with probability 0.5. The seller receives no opt-out flag; masked signals take population means.
- Data minimization
- The seller may use only first-party transactional features: purchase history, loyalty, recent purchase rate, last-period decline.
- Feature restriction
- The seller may not use the demographic-proxy group or the income segment. Behavioural signals remain.
- Personalization cap (internal reference)
- Every price within ±20% of the seller's own pooled-demand optimum: a cap on dispersion, not level. Pre-registered; turned out to be gamed by the endogeneity of the seller's own demand estimate (see Policy results).
- Personalization cap (list-price reference)
- Same band, but around the list price a regulator observes before personalization (the analytic uniform-monopoly price). Added post hoc, logged in the pre-registration changelog.
- Price cap
- No price above 3 × marginal cost (the analytic uniform monopoly price is 2.55 × cost).
- Anti-discrimination
- Hard constraint: the mean price of the proxy group may not differ from the rest by more than 2% of the mean price; enforced each period by scaling the higher-priced group down (never below cost).
- Random audits
- Each period with probability 0.2 a regulator checks the parity condition; a violating seller pays 50% of that period's profit. The seller complies voluntarily when its own model says the expected fine exceeds the profit gain from violating; otherwise it gambles.