100ms Batch Auction Pitch
Logan Jastremski argued that the HFT speed race is a design flaw in continuous limit order books and pushed frequent batch auctions (every ~100ms) to eliminate mechanical arbitrages and shift competition to price. He backed the case with millisecond-level data analysis and framed auctions as a way to improve liquidity while removing pure latency arms races. (x.com)
Budish, Cramton and Shim originally framed frequent batch auctions as uniform-price sealed-bid auctions at discrete intervals (they evaluated intervals like 1 second and suggested options down to ~100 ms) and used millisecond direct-feed data and examples like Spread Networks’ ~$300 million NYC–Chicago fiber build to motivate the design change. (conference.nber.org) An “implementation details” note by the same authors lays out concrete engineering rules for FBAs — no pre-auction transparency, uniform-price clearing, random tie-breaking, and modifications for fragmentation and Reg NMS — and discusses choosing a batch interval based on processing and dissemination constraints. (aeaweb.org) Empirical and simulation work finds that 1-second FBAs produce nearly the same trades and prices as continuous limit order books while narrowing spreads and increasing displayed depth, and that increasing the auction interval (e.g., to one minute) materially changes market dynamics and crash severity. (nber.org) Recent engineering proposals and deployments have pushed the idea toward sub-second windows: Jump Crypto’s “Dual Flow Batch Auction” targets ~100 ms dual auctions, and academic/market studies record a measurable uptick in sub-second periodic-auction use following European regulatory changes. (jumpcrypto.com) Implementing sub-100 ms auctions imposes deterministic-processing and time-synchronization requirements — exchanges and vendors cite hardware timestamping and PTP/IEEE‑1588 with FPGA-based timestamping as standard techniques to achieve sub-microsecond synchrony and nanosecond-level ordering guarantees. (link.springer.com) Policy and practitioner analyses give mixed trade-offs: a BIS working paper finds randomized uncross/periodic auctions reduce dark‑pool latency arbitrage, while industry pieces from quantitative trading firms argue sub-millisecond signals still produce economically significant price discovery and therefore will drive continued latency optimization. (bis.org)
