2.7.6 Advanced Extensions

Basic Prerequisites: 02 · The Latency Race and economics fundamentals (auctions, Nash equilibrium).

---

1. Intuition & Practical Objective

If the arms race is a consequence of the market design (a continuous, first-past-the-post book), then the cheapest fix is to change the design. Budish, Cramton & Shim’s proposal is frequent batch auctions - uniform-price sealed-bid double auctions run at discrete intervals (e.g. every 1 second) instead of continuous matching. This page is the launchpad into that debate: the mechanism, the mathematics of why it kills the speed race, and the open questions.

The mechanism kills the race two ways (BCS §1, §6):

1. It collapses the value of a tiny speed edge. Under a batch interval $\tau$, a $\delta$ speed advantage is worth only $\frac{\delta }{\tau }$ of what it is under continuous matching - and a fast trader is exposed to sniping during only a $\frac{\delta }{2\tau }$ fraction of each interval. For $\delta =100\mu$s, $\tau =1$ s: that is $10^{-4}$ and $5\times 10^{-5}$, respectively.
2. It changes the competition from speed to price. In a uniform-price batch, all orders in the same interval clear at the same price with equal time standing, so a “stale quote” cannot be selectively picked off and several fast traders with identical information must compete on price, not arrival order.

The one-sentence essence. “Frequent batch auctions convert the arms race into a price race: because a $\delta$ speed edge is only worth $\frac{\delta }{\tau }$, renting a rack stops being the winning move - spreads narrow, books deepen, and the rent that the continuous book handed to speed is returned to fundamental investors.”

---

2. Mathematical Ground Truth & Derivations

Value compression (BCS §5). In a continuous book, a trader with a $\delta$ advantage over a rival sees every jump of the signal $y$ first, so the value of the advantage approaches the whole rent $V$. Under batching with interval $\tau$:

$$\text{value of a }\delta \text{ edge under batching}=\frac{\delta }{\tau }V,\text{sniping exposure}=\frac{\delta }{2\tau }.$$

For $\tau =1$ s and $\delta =1$ ms, the edge is worth $\frac{1}{1000}$ of its continuous value - exactly BCS’s headline “1 ms advantage is 1/1000th as valuable” example. With $V=$ $75 M/yr (ES–SPY), a 1 ms edge collapses from ~75M to ~$75k/yr.

Why sniping exposure halves. A fast trader is snip-able only if a jump occurs in the vulnerable window at the end of the interval - of length $\sim \delta$ - and it needs the trader’s relative latency jitter ($\le \delta$) to be unlucky; in BCS’s minimal model this fraction is $\delta /2\tau$.

Uniform-price clearing. A batch auction aggregates buy (qty, max-price) and sell (qty, min-price) orders and clears all crossing volume at a single uniform price - the price maximizing matched quantity (a “volume-maximizing” or midpoint-of-crossed ranges rule). Every order in the batch that clears does so at the same price regardless of the submission instant, so within a batch time-priority has zero value.

Database stability (BCS §7, the “free bonus”). Because the exchange processes orders in discrete, bounded batches, it also removes order-backlog and incorrect-timestamp failure modes - salient in the Facebook-IPO and 2010 Flash Crash incident reports (order backlog, bad timestamps, exchange-engine overload from ultra-fast cancel storms).

---

3. Computational Implementation - the batch-auction arithmetic

Stdlib only. It reproduces the $\delta /\tau$ compression table and a uniform-price clearing, showing the value that batching removes from the speed tier.

Read the table across: a 1 ms edge is worth $75Mundercontinuousmatchingbutonly\ast \ast$75k** under a 1-second batch - three orders of magnitude less (the BCS 1/1000 rule). The value column discriminates cleanly: 100 µs is worth 10× less than 1 ms, and the sniping exposure is linear in $\delta$. That single compression is what turns the speed race into a price race.

---

4. Failure Modes & First-Principles Breakdowns

1. Batch = latency for the valuable order. A 1-second batch forces everyone (including an urgent fundamental order) to wait up to a full interval to clear. For orders whose value is precisely “get it done now,” batching is a new cost - the core privacy/timing tension debated in the response literature.
2. Determining $\tau$ is market design, not physics. Shorter intervals restore some speed value ($\delta /\tau$ grows); longer intervals add clearing latency. There is an optimal-interval trade-off (too-short recreates racing, too-long hurts timeliness) that BCS acknowledge is an open empirical question.
3. Uniform price ≠ price improvement for everyone. A single clearing price is better than paying the spread on a stale quote, but it can execute at a price worse than your limit for marginal orders - the “all-or-nothing style” of the batch must be verified against real LOB depth (bridge to Queue Position & Fill Probability).
4. Arms-race rent doesn’t vanish; it relocates. BCS show spreads narrow and books deepen, but market makers in a batch still race to parse information and bid within the interval - the race becomes cheaper and more price-based, not absent. First-principles check: only the speed advantaged; the information advantage remains.
5. Institutional speed still wins within a batch. All orders in a batch clear together, so a slower party is protected - but the party that can process the signal and submit first in the interval still gains. Batching damps the transmission race more than the algorithm race.

---

5. References

• Budish, Cramton & Shim (2015), QJE 130(4)
• Cohen, Kalman; Schwartz, Robert (1989) and Economides, Nicholas; Schwartz, Robert (1995)
• O’Hara, Maureen (2015)
• Menkveld (2013)

---

6. Connected Graph Bridges

• Back: 02 · The Latency Race · 05 · Failure Modes · Index Hub
• Mechanism design: Market Microstructure & Order Types (continuous vs call/auction microstructure) · Queue Position & Fill Probability (what FIFO/pro-rata means in a batch)
• Sibling latency physics: 03 · Colocation & Networks · 04 · Clock Sync
• Liquidity & design welfare: Market Impact & Depth · Avellaneda–Stoikov & Optimal Quoting