Field Notes / AdTech
№ 013 · 2026
Essay · Auction Theory · First-Price Mechanism Design

Toward Information Symmetry.

A bilateral disclosure framework for first-price ad auctions — modeled on Reg NMS, not good intentions.

975 words4 minApril 21, 2026Gad Benram, Vasco Reid

The transition of programmatic advertising to unified first-price auctions fundamentally altered the mechanism design of the ecosystem. In this environment, the primary defense for the supply side — SSPs and publishers — to protect inventory value and prevent aggressive buy-side bid shading is Floor Price Optimization (FPO).

However, the current application of FPO operates in a regulatory vacuum. Without standardized disclosure, FPO devolves into a game of adversarial black boxes. To move toward true market efficiency — defined here as allocative efficiency, where an impression reliably clears to the highest-value bidder without deadweight loss — the industry must implement a bilateral transparency framework modeled on mature financial markets.

The Mechanism Design Problem: The Dynamic Ratchet Effect

The sharpest point of friction between supply and demand is the SSP's use of intra-day, real-time bid data to calculate dynamic floor prices.

Buy-side platforms (DSPs) have historically pushed back aggressively against intra-day FPO, often mandating restrictive clauses in their Master Service Agreements. This is not paranoia; it is a rational response to a severe adverse-selection problem.

In a standard first-price auction, a bidder's utility is their valuation minus their bid (u = v − b). While first-price auctions inherently lack incentive compatibility for truthful bidding, the introduction of bid-feedback FPO transforms the interaction into a repeated game with learning. As explored in work like Amin, Rostamizadeh and Syed on repeated contextual auctions with strategic buyers, a continuous feedback loop dynamically shifts the reserve price based on the buyer's historical behavior.

This creates a ratchet effect. The dynamic game fails to have a stationary equilibrium where bidders report their optimal shaded bid. If a DSP bids aggressively to secure a high-value user, they inadvertently train the SSP's algorithm to extract their maximum willingness to pay on the next impression. The DSP is effectively penalized for revealing demand.

01 / The Ratchet Loop Each bid trains the floor. DSP BIDS HIGH reveals valuation SSP DETECTS algorithm learns FLOOR RAISED next round DSP SHAVES defensive bid RATCHET no equilibrium utility u = v − b shrinks each round demand is disclosed. no stationary equilibrium Amin · Rostamizadeh · Syed
Fig. 01 The bid-feedback loop. Each aggressive bid trains the SSP's algorithm, which raises the floor on the next round — punishing the DSP for revealing demand.

The Quantitative Reality of Dynamic Reserves

Conversely, forbidding the use of intra-day data entirely forces the supply side to operate inefficiently.

In the academic sphere, methods papers like Reserve Price Optimization for First Price Auctions in Display Advertising (Feng, Lahaie, Schneider & Ye, ICML 2021) demonstrate the complex gradient-based algorithms required to optimize reserve prices using experimental shocks. While Feng et al. focus on solving variance in gradient estimation rather than establishing industry benchmarks, commercial SSPs and vendor marketing routinely cite 10% to 15% revenue lift over static floors from deploying these machine-learning-driven dynamic models in production.

Setting floors too high severely depresses clearance rates, creating deadweight loss. But if a publisher experiences an exogenous shock — an article going viral, say — restricting floors to 30-day historical lookbacks prevents them from capturing real-time market value. Intra-day data is necessary for market efficiency; the problem is the opacity of its application.

The Symmetry Problem: Bid Shading vs. FPO

Any policy framework must acknowledge that DSPs are not passive victims. They deploy bid shading algorithms, the demand-side mirror to FPO. Both are algorithmic, opaque, and utilize historical and intra-day feedback loops to predict clearing prices.

Arguing solely for SSP transparency reads as supply-side limitation masquerading as neutral policy. A sustainable market requires bilateral disclosure. If an SSP is expected to disclose how it sets a floor, a DSP must be expected to disclose the parameters of its bid shading — acknowledging that real-world DSPs are usually optimizing for win-rate subject to simultaneous, strict CPA/ROAS constraints.

A Financial Market Analog: Reg NMS for AdTech

Mature financial markets do not tolerate this level of structural opacity. A precise analog is the SEC's Regulation NMS, specifically Rules 605 and 606, which mandate standardized execution-quality and order-routing disclosures by market centers. Brokers cannot route order flow to preferred dark pools without periodic, mandatory transparency.

AdTech requires an equivalent disclosure regime. But because the IAB Tech Lab is a standards body and not a statutory regulator like the SEC, the framework cannot rely on top-down fines. It must be commercially self-enforcing: codified in OpenRTB standards, embedded in bilateral MSAs, and enforced by the buy-side dynamically routing spend away from non-compliant inventory.

Fig. 02 — Bilateral Continuous Disclosure Supply Side SSP · floor_logic 1 · hard floor (publisher) 2 · static FPO (historical) 3 · dynamic FPO (intra-day) 4 · bid-feedback FPO Demand Side DSP · shading_strategy — win-rate target — ROAS hard cap — CPA boundary — intra-day feedback OpenRTB Extension · per-impression { "ext": { "floor_logic": 3, "shading_class": "winrate" } } Mandatory Quarterly Disclosure — SSP: aggregate floor_logic mix per seat — DSP: shading objective & constraint weights analog · SEC Reg NMS Rule 605 / 606 commercially self-enforcing via MSAs
Fig. 02 The proposed framework: standardized OpenRTB telemetry classifies how each side prices the auction, and quarterly MSA-mandated disclosures publish the aggregate strategies — bilaterally.

The Proposed Bilateral Transparency Framework

Instead of an easily-gamed "audit-on-suspicion" model, we propose a framework of continuous algorithmic disclosure, executed via OpenRTB standards and governed by MSA reporting.

1. OpenRTB Signaling (per-impression telemetry). SSPs pass an OpenRTB extension object (e.g. floor_logic) within the bid request, categorizing the floor into standard classifications:

  • 1 — Hard floor, set directly by publisher.
  • 2 — Static FPO (trailing historical data, no intra-day signals).
  • 3 — Dynamic FPO (intra-day algorithmic pricing on market-wide signals).
  • 4 — Bid-feedback FPO (explicitly uses the specific buyer's recent bid history).

2. Mandatory periodic disclosure — the AdTech Rule 606. Modeled on Reg NMS Rule 606, MSAs mandate quarterly disclosure reports:

  • SSPs disclose the aggregate percentage of impressions cleared under each floor_logic category, and whether they maintain seat-specific valuation modifiers.
  • DSPs disclose the overarching optimization constraints of their shading algorithms — for instance, the statistical weighting between win-rate maximization versus hard ROAS boundaries.

Aligning the Four Parties

Why would the broader ecosystem adopt this?

  • Advertisers benefit: predictable, allocatively efficient auctions yield more stable ROAS, freeing spend from the hidden tax of SSP–DSP algorithmic warfare.
  • Publishers keep the right to use dynamic pricing for intra-day value spikes while gaining long-term buyer trust. When DSPs know they aren't subject to bid-feedback ratcheting, they reduce aggressive defensive shading — raising baseline clearance rates.
  • SSPs and DSPs benefit by reducing the computational and commercial overhead of an adversarial arms race, competing instead on genuine infrastructure value and audience curation.

By shifting the focus from undefined "fairness" to structural market efficiency, a Reg NMS-style disclosure framework prevents intermediaries from using information asymmetry to extract undue rent — letting true market-clearing prices dictate the programmatic auction.

End.   Set in Fraunces & Newsreader.
№ 013 · 2026 · AdTech Lab