TL;DR
This paper provides the first formal mathematical framework to quantify "value leakage" in blockchain-based Dutch auctions caused by discrete block times. The authors derive closed-form expressions for Loss-Versus-Fair (LVF)—the expected discount a seller gives up to arbitrageurs—and Time-to-Fill, allowing DeFi practitioners to finally optimize the trade-off between execution speed and price quality.
What Problem Does This Paper Address?
In a perfect, continuous-time market, a Dutch auction (where the price starts high and drops until someone buys) should execute exactly at the "fair" market price. However, blockchains operate in discrete intervals (blocks).
If the fair market price of an asset is \$100, but the next block doesn't arrive for 12 seconds, the auction's price might decay to \$99.80 before anyone can technically submit a bid. An arbitrageur will "snipe" the auction the moment the block is mined, pocketing the \$0.20 difference. This is a form of Maximal Extractable Value (MEV) where the seller loses out due to the "coarseness" of the blockchain.
While previous research (Milionis et al., 2023) quantified this for Automated Market Makers (AMMs) via "Loss-Versus-Rebalancing" (LVR), this paper fills the gap for Dutch auctions and their "gradual" variants (GDAs) used in liquidations, NFT drops, and protocol fee conversions.
Methods
The authors model the system using two stochastic processes:
- Asset Price: Follows Geometric Brownian Motion (GBM), the standard model for risky asset prices.
- Block Arrivals: Modeled as a Poisson process, which captures the random nature of block discovery (particularly in Proof-of-Work, though used here as a tractable approximation for Proof-of-Stake).
The "novelty" lies in defining the log mispricing process ($z_t$)—the gap between the auction price and the fair market price—as a Markov jump diffusion process. By analyzing the infinitesimal generator of this process, the authors solve for the stationary distribution of mispricing and derive exact formulas for the expected loss and time-to-fill.
Datasets
Not discussed in the paper. This is a theoretical and methodological paper. The authors use representative numerical examples (e.g., assuming 5% daily volatility and 12-second block times) to illustrate their formulas rather than training on historical datasets.
Key Insights and Ideas
- The "Memoryless" Efficiency: If an auction starts "out-of-the-money" (the starting price is higher than the market price), the expected loss (LVF) is independent of the starting price. This is because the price must first "drift" down to the market price; once it hits that boundary, the expected loss is determined solely by the decay rate, volatility, and block time.
- The Unavoidable Tax: There is a fundamental lower bound on loss. Even with a perfect decay rate, the seller faces an unavoidable loss of approximately $\sigma \sqrt{\Delta t / 2}$ (where $\sigma$ is volatility and $\Delta t$ is mean block time). This represents the "jitter" of the market price between blocks.
- GDA Deconstruction: The paper proves that a Gradual Dutch Auction (GDA)—which sells tokens continuously over time—is mathematically equivalent to a continuum of infinitesimal regular Dutch auctions. This allows the same LVF formula to apply to both.
- Pareto Frontier: The authors demonstrate that practitioners shouldn't just "guess" a decay rate. There is an "efficient frontier" where you can minimize a weighted sum of loss and time-to-fill.
Key Results
The paper provides a "plug-and-play" formula for Loss-Versus-Fair ($LVF_+$): $$LVF_+ = \frac{1}{1 + \frac{\delta}{\sigma^2} \left( \sqrt{1 + \frac{2\sigma^2}{\delta^2 \Delta t}} - 1 \right)}$$
Concrete Examples: * The Seller's Cost: For an asset with 5% daily volatility on a chain with 12-second blocks (like Ethereum) and a 0.01%/sec decay rate, the expected loss is 0.13%. For every \$100 sold, \$0.13 goes to arbitrageurs. * The Speed Cost: With the same parameters and a starting price 0.1% above market, the expected time-to-fill is 23.3 seconds. * L1 Design Rule of Thumb: If a blockchain wants to support Dutch auctions with less than 2 basis points (0.02%) of leakage for 5% volatility assets, it must have block times faster than 2.75 seconds.
Limitations and Future Work
- Deterministic Slots: Most modern chains (like Ethereum post-Merge or Solana) use deterministic slots rather than a Poisson process. The authors acknowledge this but use Poisson for mathematical tractability.
- Gas Fees: The model assumes arbitrageurs have zero transaction costs. In reality, high gas fees might protect the seller by preventing snipes until the mispricing exceeds the gas cost.
- Multi-block MEV: The model assumes block proposers are independent. If a single entity controls multiple blocks in a row, they could intentionally delay an auction to increase their arbitrage profit.
- Price Jumps: GBM assumes continuous price movements. Real-world markets often have "jumps" or "fat tails" during crashes, which would likely increase LVF significantly.
Summary updated