Research
Auctioning Time to Mitigate Latency Races: Theory and Evidence from Blockchains
Traders who compete for time-sensitive profit opportunities (arbitrage, liquidations) spend real resources simply to be first, and much of that spending is socially wasteful.

- arXiv
- 2512.10094
- Published
- 2025-12-10
- Authors
- Agostino Capponi, Brian Zhu
AI summary
Overview
Research area: Market design and mechanism design applied to blockchain transaction ordering — specifically the game theory of latency races (high-frequency trading arms races) and empirically testing a time-priority auction on a Layer-2 blockchain.
Technical level: Intermediate, with an Advanced game-theoretic core. The institutional background and intuition are accessible, but the formal results rely on Tullock contest theory, subgame-perfect Nash equilibrium, and exponential-order-statistic derivations.
Scope (one sentence): The paper builds a contest-theoretic model of costly latency competition, specializes it to blockchain duplicate transaction submission with linear (gas-fee) costs, and uses Arbitrum's Timeboost auction as a natural experiment to test whether auctioning artificial time priority reduces wasteful racing and raises platform revenue.
What This Paper Is About
Traders who compete for time-sensitive profit opportunities (arbitrage, liquidations) spend real resources simply to be first, and much of that spending is socially wasteful. In traditional markets these speed investments are private and unobservable, so the waste is hard to measure; on blockchains, traders submit many identical copies of the same transaction and the losers revert on-chain at a cost, leaving a public record.
The paper asks what happens when a platform stops awarding priority to whoever arrives first and instead auctions that priority explicitly. It studies Arbitrum's Timeboost — a sealed-bid second-price auction for 200 milliseconds of transaction time priority — as a natural experiment, predicting and then testing whether such an auction redirects resources from speed races into platform revenue.
Key Contributions
-
A general theory of auctioned time priority. The authors model a common-value opportunity contested by N ≥ 3 traders who invest in latency, with win probability in Tullock contest form (investment relative to total investment) and convex, log-convex-derivative cost functions. They prove a unique symmetric pure-strategy Nash equilibrium exists (Proposition 3.1) and that adding a sealed-bid second-price auction for a time advantage yields a unique symmetric subgame-perfect equilibrium with bid b* = u_w* − u_l* (Propositions 3.2 and 3.3).
-
A proof that auctioning priority reduces aggregate racing effort. Theorem 3.4 establishes K_A* < K*, i.e. total latency investment under the auction is strictly lower than without it, because losers scale back more than the winner scales up.
-
A microfoundation of blockchain latency races with closed-form expressions. The general model is specialized to exponential network delays and linear per-copy gas costs, giving closed-form equilibrium quantities: k* = ((N−1)/N²)·(V/C) copies per trader, K* = ((N−1)/N)·(V/C) total transactions, per-trader utility u_i* = V/N², user surplus W* = V/N, and platform revenue R* = ((N−1)/N)·V. Proposition 4.1 shows the auction lowers total transactions and unambiguously raises sequencer revenue.
-
An extension to multi-winner auctions plus causal empirical evidence. The model is generalized to n_w winners under a sealed-bid generalized second-price auction, analyzing both the intensive margin (size T of the time advantage) and the extensive margin (number of winners), and the predictions are tested with a difference-in-differences event study around Timeboost's deployment on Arbitrum against control Layer-2 networks.
Main Findings
-
Priority auctions cut total speed spending. Without an auction, all traders must invest to stay competitive and the surplus is largely dissipated; with a single-winner auction, only the winner has a strong incentive to keep investing while the remaining traders scale back, so aggregate effort falls (Theorem 3.4).
-
The winner invests more, the losers invest less. Proposition 3.2 establishes 0 < k_l* < k_w*: conditional on winning, the time advantage amplifies the marginal return to effort; conditional on losing, it reverses the slope of marginal returns. Both types retain positive continuation value, with 0 < u_l* < u_w*.
-
The auction extracts exactly the surplus gap. The unique symmetric equilibrium bid is b* = u_w* − u_l*, so the mechanism captures rents from the latency race while preserving participation incentives (Proposition 3.3).
-
Platform revenue rises even though revert fee income falls. Proposition 4.1 shows both K_tb* < K* and R_tb* > R*: auction payments more than compensate for the gas fees lost from fewer reverted duplicate transactions. This addresses the concern that exchanges resist efficiency-improving design because it cannibalizes speed-technology revenue (citing Budish et al. (2020)).
-
In the blockchain microfoundation, effort is close to linear. Because duplicate submissions cost gas fees proportional to the number of copies, the blockchain setting is the special case of approximately linear effort costs; the user's revert costs accrue to the sequencer node or associated DAO rather than to third-party service providers.
-
Multi-winner design can help along both margins. Along the intensive margin, total effort broadly decreases and platform revenue increases in the amount of time priority T awarded to winners. Along the extensive margin, when the total number of users is large, increasing the number of winners may further reduce total effort and improve platform revenue.
-
Empirical results match the theory. Following Timeboost's adoption, redundant (duplicate) transaction submissions decline and platform revenue increases relative to benchmark Layer-2 networks, and non-priority traders experience higher failure rates in contested opportunities. The specific numeric magnitudes of these estimated effects are not reported in the available text.
-
Distinctive L2 MEV structure supports the mechanism. The paper notes that sandwich attacks are essentially non-existent on Layer-2s (citing Gogol et al. (2026)); instead L2s exhibit "optimistic MEV" and backruns, where traders repeatedly submit conditional transactions that often revert — exactly the duplicate-submission behavior the model captures.
Methodology in Plain English
The paper works in two stages.
First, it builds a stylized contest model. A single valuable opportunity exists and only one trader can capture it. Each trader chooses how much "effort" to spend on being fast, and the chance of winning equals that trader's effort divided by everyone's combined effort — a Tullock contest — with convex costs. The authors solve for the symmetric equilibrium and then re-solve the game with a preliminary sealed-bid, second-price auction that grants the winner an artificial time advantage of size T before the race begins. Comparing total effort with and without the auction gives the central theoretical result.
Second, it grounds that abstract model in how blockchains actually work. Network delays are modeled as exponential random draws; a trader submitting k identical copies effectively draws k delays and only the fastest matters, so submitting more copies mechanically raises the win probability. Costs are linear because each reverted copy burns gas. With exponential delays, the win probability collapses to the same ratio form as the Tullock contest, which is what justifies treating latency races and blockchain spam as one phenomenon. This yields closed-form formulas for the equilibrium number of copies, total on-chain spam, user surplus, and sequencer revenue.
Third, the empirical test treats Timeboost's launch on Arbitrum as an exogenous change. Because duplicate submissions are publicly visible on-chain, the authors can proxy "latency effort" with bursts of identical transactions and run a difference-in-differences event study: outcomes on Arbitrum around adoption versus outcomes on comparable Layer-2 networks that did not adopt the mechanism.
Why This Matters
Impact on research. The paper links two literatures that normally run in parallel — traditional market microstructure work on high-frequency trading and latency races, and DeFi work on maximal extractable value — under a single model of costly time-priority acquisition. It also supplies something the HFT literature has lacked: a setting where the resource cost of losing a speed race is publicly recorded. Methodologically, it shows that Layer-2 sequencers can be analyzed as matching engines whose ordering rules are design choices rather than infrastructure details.
Real-world applications:
- Layer-2 sequencer design. Timeboost is live on Arbitrum, and this paper offers a parameterization framework for how much time priority to sell (T) and how many winners (n_w) to allow.
- MEV mitigation and blockspace congestion. Reducing redundant submissions frees blockspace that would otherwise be consumed by failed race attempts, an externality borne by all users.
- Traditional exchange design. The results bear on frequent batch auctions (Budish et al. (2015)), speed bumps such as the IEX 350-millisecond delay (Woodward (2018)), and Pigouvian taxes on speed investment (Biais et al. (2015)) — suggesting explicit priority pricing as a parallel option.
- Protocol treasury economics. Since revert fees and auction payments both flow to the sequencer or DAO, the paper shows a revenue-neutral-to-positive path for adoption.
Industry relevance. The finding that platform revenue rises under the auction directly challenges the assumption that venues resist efficiency-improving market design because it destroys co-location and speed-technology revenue. For exchanges that own or tightly control their co-location facilities, the same logic suggests the rent can be recaptured rather than lost. For L2 operators choosing between first-come-first-served and priority gas auction sequencing, the paper quantifies the cost of the status quo.
Future Directions
- Empirically testing the multi-winner design. The theory shows that more winners can reduce effort and raise revenue when the user base is large, but Timeboost as deployed is single-winner; the extensive-margin prediction remains untested.
- Calibrating T and the reserve price. Timeboost allocates 200 milliseconds per 60-second round with a 0.001 ETH reserve price. The model characterizes how outcomes vary in T, but the optimal size of the time advantage and the optimal reserve are left as open design questions, as is the effect of the 60-second auction cycle and the 15-second on-chain propagation window.
- Centralization and resale of priority access. Winners can resell Express Lane access to third parties for on- or off-chain side payments, and prior descriptive work (Messias and Ferreira Torres (2025)) documents concentration of express-lane usage and high revert rates. Whether resale undermines the efficiency gains, and how allocative efficiency (not just total effort) changes, is unresolved.
- Interaction with other sequencing reforms. The paper cites "revert protection" (Zhu et al. (2026)), where sequencers simulate and exclude failing transactions, as a mechanism that can make priority fee auctions more competitive. How revert protection interacts with, substitutes for, or complements a time-priority auction is an open question, as is extending the analysis beyond approximately linear costs to settings with convex effort.
Target Audience
Market-design and microstructure economists interested in latency races, batch auctions, and the economics of exchange competition; blockchain and MEV researchers studying ordering, extraction, and Layer-2 infrastructure; protocol designers and sequencer operators choosing among FCFS, priority gas auctions, and explicit priority auctions; and regulators or policy analysts working on high-frequency trading and market structure, who will find the blockchain setting a rare source of observable data on costs that are invisible in traditional markets. Readers without game-theory background can follow the institutional sections and empirical design; the propositions require comfort with contest theory and equilibrium analysis.
Authors’ abstract
High-frequency trading, in both traditional and decentralized markets, induces latency races and redundant order flow as traders spend resources to win time-sensitive opportunities. We show that auctioning artificial time priority can redirect resources away from wasteful speed races toward auction payments. While such waste is difficult to measure in traditional markets, blockchain transactions provide transparent records of these competitive costs through observable…
The full abstract on arXiv