2.1.6 Advanced Extensions

Basic Prerequisites: 05 - Failure Modes & Practice.

---

1. Intuition & Practical Objective

The order types and mechanics of the previous pages are the alphabet; this page is the grammar of what is written with it - the taxonomy of high-frequency strategies, the economics of the latency race, and the market-design and regulatory responses. The three questions:

1. What are the HFT strategies? Nearly all reduce to a small set: market making (earn the spread and rebates), arbitrage (latency, cross-venue, index/ETF, and statistical), and order-anticipation / liquidity detection (infer and front-run a predictable flow).
2. Why is there a latency arms race? Because in a continuous market, the first firm to react to a stale quote captures a riskless edge; the race is a rent-dissipation contest whose cost is pure social waste once it exceeds the value of the edge (Biais–Foucault–Moinas).
3. What are the design and regulatory responses? Speed bumps, batch auctions, order-protection and best-execution rules, and trading-obligation/market-access regimes - each an attempt to keep the race from consuming the market’s value.

The one-sentence essence. “HFT is a taxonomy of ways to be first - make, arbitrage, anticipate - and market design is the debate over whether being first should pay at all (hence speed bumps and batch auctions).“

---

2. Mathematical Ground Truth & Derivations

The latency race as a winner-take-most contest. Let the faster firm’s edge be $\Delta$ (its latency advantage) and let reaction times be exponentially distributed with scale $\tau$. The probability the faster firm wins the stale-quote race is

$$P(\text{win})=\mathrm{Pr}[T_{\text{slow}}-T_{\text{fast}}>\Delta ]\approx e^{-\Delta /\tau }.$$

With value captured per win $V$ and $N$ race events, the faster firm’s daily rent is

$$R(\Delta )=NV{e}^{-\Delta /\tau }.$$

Key implications: (i) the slower firm’s profit is $NV(1-e^{-\Delta /\tau })$ and it is negative once adverse selection dominates - speed is defensive as much as offensive; (ii) $R$ is concave in $\Delta$, so the race’s marginal returns vanish yet the spending continues - the classic arms-race signature; (iii) in equilibrium the social return to speed can be near zero (the losers just stopped losing), which is the Biais–Foucault–Moinas inefficiency.

Frequent batch auctions remove the race (Budish–Cramton–Shim). If all orders arriving in a window $[t,t+\delta ]$ are cleared together at a single uniform price with ties broken by size, then reaction speed within the window is worthless:

$$\mathbb{E}[\text{speed rent}]=0\text{for any }\Delta <\delta ,$$

because whoever arrives marginally earlier in the window loses ties to the larger order and crosses at the same price. This is the formal market-design counterpoint to continuous matching - the modern, high-frequency version of the call auction from 04 · Auctions & Continuous Trading.

Market-maker net economics. A passive market maker’s per-share profit decomposes as

$$\text{net}=\underset{\text{rebate}}{\underbrace{r_m}}+\underset{\text{realized spread}}{\underbrace{S_r}}-\underset{\text{adverse selection}}{\underbrace{\text{AS}}},$$

so a higher rebate is not “free money” - it is compensation that must cover the adverse-selection term (Hasbrouck Ch 5; the Glosten–Milgrom spread).

The regulatory frame. Regulation acts directly on the mechanics of this folder: order-protection / trade-through rules (Reg NMS Rule 611 in the US, shaping SOR and the NBBO), best-execution and algorithmic-trading rules (MiFID II, notably RTS 6 obligations on testing, kill-switches, and market-making schemes), market-access controls (SEC Rule 15c3-5), execution-quality disclosure (Rule 605/606), and dark-pool reporting. None of these changes the mechanics; all of them change what an engine is allowed to do with them.

---

3. Computational Implementation - latency rent and the batch-auction counterfactual

Quantify the sniping profit as a function of latency advantage, show the batch auction zeros it, and decompose the market-maker PnL. Standard library only; re-executed and reproduced.

Read the numbers. The winning probability decays exponentially ($e^{-\Delta /50}$), so a 10 µs edge keeps 82% of the $16M/day pot while a 200 µs edge keeps only 1.8% - and crucially, the slower firm’s 18% share at 10 µs is its losses, not its profits. The concavity is exactly the arms-race signature: the first 10 µs of speed buy ~$3M/day of rent; the next 190 µs buy a further ~$12.8M more but at vastly higher engineering cost. Batch clearing zeros the entire $\Delta$-dependence. Finally, the market maker’s 5.5 cents/share net is conditional on the adverse-selection term being small - raise adverse selection above rebate $+$ spread and the strategy is a guaranteed loser.

---

4. Failure Modes & First-Principles Breakdowns

1. Confusing the rebate with the edge. A maker-taker rebate is compensation for quoting, not profit; it is paid out of the adverse-selection and inventory losses the maker takes. A strategy that nets positive only because of rebates is short volatility in disguise.
2. Assuming the latency race is winnable. $R(\Delta )\propto e^{-\Delta /\tau }$ means a firm one order of magnitude slower than its peers is not “a bit behind” - it is the liquidity the fast firms trade against. The viable choices are to compete at the frontier or to not post defensible quotes; a middling latency is the worst of both.
3. Regulatory constraints are part of the strategy space. Order-protection, best-execution, kill-switch, and market-access rules change which strategies are legal, not just which are fast. A latency-arbitrage design that ignores Rule 611 / MiFID II obligations can be both unprofitable and non-compliant.
4. Batch auctions are a design trade, not a free lunch. They remove the speed race but impose latency, price uncertainty within the window, and their own tie-breaking incentives. No mechanism dominates; the choice is a policy question, which is why it remains contested.

---

5. References

• Budish, Cramton & Shim - “The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response,” QJE 130(4), 2015. The batch-auction counterfactual and the latency-race critique.
• Biais, Foucault & Moinas - “Equilibrium fast trading,” JFE 116(2), 2015. When private speed investment is socially wasteful.
• Hasbrouck & Saar - “Low-latency trading,” JFM 16(4), 2013. Order lifetimes and cancel rates that define the latency-sensitive trader.
• Menkveld - “High-frequency trading and the new market makers,” JFM 16(4), 2013. HFT as the new cross-venue market makers.
• Cartea, Jaimungal & Penalva - Algorithmic and High-Frequency Trading (2015). The mathematical spine: market making and latency under stochastic control.
• MacKenzie, Donald - Trading at the Speed of Light (2021). The socio-technical account of colocation, microwave links, and the race to the matching engine.
• Regulatory primary sources: SEC Reg NMS (Rule 611 order protection; Rules 605/606 disclosure; Rule 15c3-5 market access); MiFID II / RTS 6 (algorithmic-trading obligations); exchange colocation and timestamp specifications. Treat these as primary documentation, not secondary commentary.

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes & Practice · Index Hub
• Systems tier: Low-Latency Systems Architecture · Hardware Acceleration & FPGA
• Queue & fill: Queue Position & Fill Probability
• Market-maker economics: Market-Maker Economics & Rebates · Avellaneda–Stoikov & Optimal Quoting
• Design counterpoint: 04 · Auctions & Continuous Trading