Bidders' Responses to Auction Format Change in Internet Display Advertising Auctions

Gabriel Y. Weintraub; Ralph Mastromonaco; Sam Seljan; Shumpei Goke

arxiv: 2110.13814 · v3 · submitted 2021-10-20 · 💰 econ.GN · cs.GT· q-fin.EC

Bidders' Responses to Auction Format Change in Internet Display Advertising Auctions

Shumpei Goke , Gabriel Y. Weintraub , Ralph Mastromonaco , Sam Seljan This is my paper

Pith reviewed 2026-05-24 13:24 UTC · model grok-4.3

classification 💰 econ.GN cs.GTq-fin.EC

keywords first-price auctionsecond-price auctiondisplay advertisingbid shadingdifference-in-differencesrevenue equivalenceauction format changeinternet advertising

0 comments

The pith

Publishers switching display ad auctions to first-price format see prices per impression rise 25-75 percent at first.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tracks what happens to prices when publishers of internet display ads change from second-price auctions to first-price auctions on a staggered schedule. It compares price movements for publishers that switch against those that do not, using both standard difference-in-differences and a synthetic version that accounts for pre-existing trends. Right after the change, revenue per sold impression jumps sharply for the treated publishers. In later switches the extra revenue shrinks over time, pointing to bidders gradually increasing how much they shade their bids rather than jumping straight to a new equilibrium. The work supplies field evidence on bidder adjustment that complements theoretical predictions about auction formats.

Core claim

When publishers adopt first-price auctions in place of second-price auctions, revenue per sold impression increases by 25 to 75 percent relative to publishers that continue using second-price auctions. For later adoptions this price gap narrows over subsequent periods, consistent with bidders learning to shade bids more aggressively rather than immediately reaching the Bayesian Nash equilibrium of the new format.

What carries the argument

Staggered publisher adoption of first-price auctions, measured with difference-in-differences regressions and a synthetic difference-in-differences estimator that constructs a counterfactual from control units.

If this is right

Bidders initially shade too little after the format change, producing a transitory revenue increase for sellers.
The price premium under first-price auctions tends to erode as bidders gain experience and adjust shading.
Later format changes show faster or more complete reversion than early ones in several specifications.
Theoretical revenue equivalence between the two formats appears to hold more closely in the longer run than immediately after the switch.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Auction platforms considering format changes should expect an initial revenue bump followed by partial reversion if bidders can learn.
Markets with frequent bidder turnover might see slower adjustment than markets with stable participants.
Publishers could test temporary format switches to capture short-term gains while monitoring long-term bid behavior.
The pattern suggests similar learning dynamics may appear in other repeated auction settings such as procurement or ad exchanges.

Load-bearing premise

Treated and control publishers would have followed the same price trends in the absence of the auction format change.

What would settle it

A difference-in-differences estimate showing no statistically significant price increase after the switch, or clear divergence in pre-treatment price trends between treated and control publishers.

Figures

Figures reproduced from arXiv: 2110.13814 by Gabriel Y. Weintraub, Ralph Mastromonaco, Sam Seljan, Shumpei Goke.

**Figure 2.** Figure 2: Types of publishers and dates of format changes. The four columns on the right indicate treatment– [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: Weekly time series of the average price by each publisher group as defined in Table [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

**Figure 4.** Figure 4: Estimates of βk (effects of format change on average price). The solid line indicates point estimates, and the band indicates 95% confidence intervals [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

**Figure 5.** Figure 5: Cumulative percentage of Global Company September Publishers’ revenue from advertising cam [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

**Figure 6.** Figure 6: Histogram of growth rates of bidders’ spending on Global Company September Publishers from [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗

**Figure 7.** Figure 7: Effects of auction format changes on the spending of the AppNexus/Xandr bidder and non [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗

read the original abstract

We study actual bidding behavior when a new auction format gets introduced into the marketplace. More specifically, we investigate this question using a novel dataset on internet display advertising auctions that exploits a staggered adoption by different publishers (sellers) of first-price auctions (FPAs), instead of the traditional second-price auctions (SPAs). We analyze the auction format change using difference-in-differences regressions and a synthetic difference-in-differences estimator, which better handles pre-trends. The results show that revenue per sold impression (price) jumps considerably for treated publishers relative to control publishers, with increases ranging from 25% to 75% of the pre-treatment price level of the treated group. Moreover, for later auction format changes, the increase in price levels under FPAs relative to those under SPAs tends to dissipate over time, reminiscent of the revenue equivalence theorem, although the extent of this reversion depends on the specification. We view these results as suggestive of initially insufficient bid shading following the format change, as opposed to an immediate transition to a new Bayesian Nash equilibrium, with prices tending to decline in several specifications in a manner consistent with gradual adjustment in bidding behavior as bidders learn to shade their bids. Our work constitutes one of the first field studies on bidders'responses to auction format changes, providing an important complement to theoretical model predictions. As such, it provides valuable information to auction designers when considering the implementation of different formats.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives field evidence from staggered FPA adoptions showing an initial 25-75% price jump that partially reverts in some specs, but the causal claim hinges on parallel trends holding amid publisher timing choices.

read the letter

The core finding is that publishers switching from second-price to first-price auctions saw revenue per impression rise sharply at first, then show some decline over time in later switches. This lines up with bidders initially under-shading before adjusting toward equilibrium behavior. The work stands out for using real marketplace data from a natural experiment rather than lab or theory alone, and the staggered timing lets them apply both standard DiD and synthetic DiD to address pre-trends. That combination is a reasonable way to handle observational auction data. The abstract is upfront that the reversion result is specification-dependent, which is honest. The main soft spot is identification: even with synthetic DiD, the design assumes no unobserved demand shifts or anticipation that line up with when individual publishers adopt the new format. If adoption correlates with other market changes, the initial jump and later dissipation could be overstated. Data construction details and full robustness checks would need close scrutiny in a review. This paper is useful for anyone working on auction format changes in digital markets or empirical auction design more broadly. It supplies a concrete empirical benchmark that theory papers can reference. I would send it to peer review; the question is policy-relevant and the approach is transparent enough to be worth referee time even if revisions are needed on the identification checks.

Referee Report

2 major / 1 minor

Summary. The manuscript investigates bidders' responses to the staggered introduction of first-price auctions (FPAs) replacing second-price auctions (SPAs) in internet display advertising auctions. Leveraging a novel dataset and employing difference-in-differences (DiD) regressions along with a synthetic DiD estimator, the authors report that revenue per sold impression increases by 25% to 75% for treated publishers compared to controls, with the effect tending to dissipate over time in later adoptions, interpreted as evidence of gradual bid shading adjustment rather than immediate equilibrium.

Significance. If the results hold under the identification assumptions, this study offers one of the first field-based analyses of auction format changes, providing empirical complement to theoretical predictions such as revenue equivalence. The application of synthetic DiD to better handle pre-trends strengthens the analysis, and the findings are relevant for auction designers in digital advertising platforms.

major comments (2)

[Empirical Strategy] The identification of the causal effect of the format change relies on the parallel trends assumption for DiD and the validity of the synthetic counterfactual. However, details on pre-trend diagnostics, the donor pool for synthetic DiD, and how weights are chosen for each adoption timing are not sufficiently elaborated, which is load-bearing given the staggered nature of the treatment and potential for time-varying confounders.
[Results and Discussion] The abstract notes that the extent of reversion (dissipation) depends on the specification. A more detailed presentation of the range of specifications and their implications for the gradual adjustment hypothesis would strengthen the central claim, as the interpretation hinges on whether prices decline in a manner consistent with learning.

minor comments (1)

[Abstract] Typo in abstract: 'bidders'responses' should read 'bidders' responses'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important areas for improving the clarity of our identification strategy and the robustness of our interpretation. We address each major comment below and will incorporate revisions accordingly.

read point-by-point responses

Referee: [Empirical Strategy] The identification of the causal effect of the format change relies on the parallel trends assumption for DiD and the validity of the synthetic counterfactual. However, details on pre-trend diagnostics, the donor pool for synthetic DiD, and how weights are chosen for each adoption timing are not sufficiently elaborated, which is load-bearing given the staggered nature of the treatment and potential for time-varying confounders.

Authors: We agree that greater elaboration on these aspects will strengthen the paper. In the revision, we will expand Section 3 to include additional pre-trend diagnostics (e.g., formal tests for parallel trends and placebo exercises), a detailed description of the donor pool construction, and the procedure for selecting and reporting weights across adoption cohorts in the synthetic DiD. These details, along with discussion of time-varying confounders, will also appear in a new appendix. revision: yes
Referee: [Results and Discussion] The abstract notes that the extent of reversion (dissipation) depends on the specification. A more detailed presentation of the range of specifications and their implications for the gradual adjustment hypothesis would strengthen the central claim, as the interpretation hinges on whether prices decline in a manner consistent with learning.

Authors: We concur that a fuller presentation of specifications will bolster the gradual adjustment interpretation. The revised manuscript will add a table (and associated discussion) reporting results across an expanded set of specifications, including alternative controls, fixed effects, and estimators. We will explicitly map how dissipation patterns vary and evaluate their consistency with bidder learning, while noting where the pattern is robust versus specification-dependent. revision: yes

Circularity Check

0 steps flagged

No circularity: standard empirical DiD on observational data with no self-referential derivations.

full rationale

The paper applies difference-in-differences and synthetic DiD estimators to real-world staggered adoption data on auction format changes. No equations, predictions, or central claims reduce by construction to fitted inputs, self-citations, or ansatzes imported from the authors' prior work. Identification rests on external econometric assumptions (parallel trends, valid counterfactuals) rather than any loop internal to the paper's outputs. Results are framed as suggestive field evidence complementing theory, without renaming known patterns or smuggling assumptions via citation chains.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the validity of DiD identifying assumptions and the quality of the proprietary advertising auction dataset; no new entities postulated.

free parameters (1)

DiD treatment effect coefficients
Estimated parameters capturing the price jump and subsequent reversion; fitted to the observed transaction data.

axioms (2)

domain assumption Parallel trends assumption between treated and control publishers absent the format change
Invoked to identify causal effects in the staggered adoption design.
domain assumption No anticipation or confounding events coinciding with adoption timing
Required for the difference-in-differences and synthetic DiD estimators to isolate the format change effect.

pith-pipeline@v0.9.0 · 5802 in / 1315 out tokens · 26582 ms · 2026-05-24T13:24:33.053943+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

[1]

Just Starting Out: Learning and Equilibrium in a New Market,

Accessed: 2021-08-11. Doraszelski, Ulrich, Gregory Lewis, and Ariel Pakes, “Just Starting Out: Learning and Equilibrium in a New Market,” American Economic Review, 2018, 108 (3), 565–615. Edelman, Benjamin, Michael Ostrovsky, and Michael Schwarz, “Internet Advertising and the General- ized Second-Price Auction: Selling Billions of Dollars Worth of Keyword...

work page doi:10.1145/3442381.3449841 2021
[2]

We weight the observations by the number of impressions

We ﬁrst regress, for each publisher p’s time series{ypt}, ypt = γp,dow(t) + γp,dom(t) + γp,month(t) + γp,eoq(t) + δpt, using the data before the format change. We weight the observations by the number of impressions

work page
[3]

We compute the ﬁtted values of the previous regression ˆyt and subtract its mean ˆy, which is obtained by regressing ˆyt on a constant

work page
[4]

The demeaned ﬁtted value ˆyt− ˆy shows the seasonal component, and so subtracting this seasonal component from yt gives the deseasonalized time series ˜ypt

work page
[5]

two-step

Regress ˜ypt as follows: ˜ypt = αp + ∑ k≤k≤k,k̸=−1 βkDp· 1(Kt = k) +γt + ˜εpt . Figure A.4 estimates the main regression (1) but without any seasonal ﬁxed effects. 60 0 60 Number of days from 9/21/2017 0.4 0.2 0.0 0.2 0.4 0.6 0.8 Difference in USD/1000 Global Company: CPM level, 2-step dummy 60 0 60 Number of days from 9/17/2019 4 3 2 1 0 1 2 3 Difference...

work page 2017

[1] [1]

Just Starting Out: Learning and Equilibrium in a New Market,

Accessed: 2021-08-11. Doraszelski, Ulrich, Gregory Lewis, and Ariel Pakes, “Just Starting Out: Learning and Equilibrium in a New Market,” American Economic Review, 2018, 108 (3), 565–615. Edelman, Benjamin, Michael Ostrovsky, and Michael Schwarz, “Internet Advertising and the General- ized Second-Price Auction: Selling Billions of Dollars Worth of Keyword...

work page doi:10.1145/3442381.3449841 2021

[2] [2]

We weight the observations by the number of impressions

We ﬁrst regress, for each publisher p’s time series{ypt}, ypt = γp,dow(t) + γp,dom(t) + γp,month(t) + γp,eoq(t) + δpt, using the data before the format change. We weight the observations by the number of impressions

work page

[3] [3]

We compute the ﬁtted values of the previous regression ˆyt and subtract its mean ˆy, which is obtained by regressing ˆyt on a constant

work page

[4] [4]

The demeaned ﬁtted value ˆyt− ˆy shows the seasonal component, and so subtracting this seasonal component from yt gives the deseasonalized time series ˜ypt

work page

[5] [5]

two-step

Regress ˜ypt as follows: ˜ypt = αp + ∑ k≤k≤k,k̸=−1 βkDp· 1(Kt = k) +γt + ˜εpt . Figure A.4 estimates the main regression (1) but without any seasonal ﬁxed effects. 60 0 60 Number of days from 9/21/2017 0.4 0.2 0.0 0.2 0.4 0.6 0.8 Difference in USD/1000 Global Company: CPM level, 2-step dummy 60 0 60 Number of days from 9/17/2019 4 3 2 1 0 1 2 3 Difference...

work page 2017