Imported from Floorey/stock_screener (.claude/skills/high-frequency-trading/SKILL.md). Install upstream with npx skills add Floorey/stock_screener --skill high-frequency-trading. Copyright stays with the author.

High-Frequency Trading (Gomber et al. 2011)

What this skill is

A structured distillation of a 2011 study on HFT and algorithmic trading (AT) in European and U.S. equity markets. It exists because HFT discussion is unusually prone to sloppy terminology and folklore: people use "HFT" to mean a strategy, a firm type, a speed, or a moral category, and then argue past each other. This skill supplies precise definitions, a strategy taxonomy, what the empirical literature actually found, and how U.S. and European market structure differ — so answers rest on something citable.

Source. Peter Gomber, Björn Arndt, Marco Lutat, Tim Uhle, High-Frequency Trading, Chair of Business Administration esp. e-Finance, E-Finance Lab, Goethe University Frankfurt, March 2011. Cite it as "Gomber et al. (2011)".

Two caveats to carry into every answer:

1. It is a 2011 document. The regulatory sections describe proposals that were then open questions. Several were subsequently enacted in forms the paper argued against (MiFID II order-to-trade ratios, algo notification, market-making obligations). Never present the paper's regulatory picture as current law. references/since-2011.md tracks what changed — read it before saying anything about today's rules.
2. It was commissioned work. The title page carries a "Commissioned by" credit, and the study is generally cited as sponsored by an exchange operator. Its descriptive and literature-review content is solid and widely cited; its policy recommendations are informed but not disinterested, and they line up with lit-venue operator interests. Attribute conclusions to the authors rather than presenting them as settled fact.

The organizing idea

HFT is not a trading strategy. It is a set of technologies — low-latency market access, direct data feeds, co-location, automated order management — used to implement trading strategies that mostly predate it.

Almost every useful move in this domain follows from taking that seriously. When someone asks "is HFT good or bad", "should HFT be regulated", or "is HFT front-running", the question is underdetermined as posed. Redirect to the underlying strategy: market making, arbitrage, liquidity detection and momentum trading have different economics, different market-quality effects, and different abuse potential. Spread capturing and cross-venue arbitrage are legitimate activities that happen to be executed fast; layering and quote stuffing are market abuse whether executed fast or slow. Regulating "speed" hits both.

The same move applies institutionally: assess by function, not institution. HFT techniques are used by investment bank prop desks, specialist proprietary boutiques, registered market makers and quant hedge funds alike. Any rule aimed at "HFT firms" as a category both misses hybrid users and creates an unlevel playing field.

Classifying what someone describes

When a user describes a strategy, system, or firm, place it before evaluating it. The paper's definitions are deliberately family-resemblance ones: something is AT or HFT if it shows most but not necessarily all of these, not if it satisfies a checklist.

Shared by AT and HFT: pre-designed trading decisions · used by professional traders · real-time market data observation · automated order submission · automated order management · no human intervention in the loop · direct market access.

Specific to AT but not HFT — the "classical" execution-algorithm side: agent trading (working client orders) · minimizing market impact of large orders · targeting a benchmark · holding periods of days to months · working one parent order through time and across venues.

Specific to HFT: very high order counts · rapid order cancellation · proprietary trading (own capital) · profit from buying and selling as a middleman · flat at end of day · very short holding periods · very low margin per trade · low-latency requirement · co-location / proximity and individual data feeds · focus on highly liquid instruments.

HFT is a subset of AT. The tell that separates them is whose money and what horizon: agent execution against a benchmark over hours-to-months is AT; proprietary round-trips flat by the close is HFT.

Three neighbours get confused with HFT constantly; keep them distinct:

Full definition tables, the academic and regulatory definitions the paper compares, and the drivers behind AT/HFT adoption (DMA and sponsored access, maker-taker fees, latency, MiFID-driven fragmentation): references/definitions.md.

The strategy map

The last row matters: these are abuse categories, not HFT categories. Fast technology can make them easier and more profitable, which is a supervisory problem — but they are wrong when a human does them slowly too, and non-abusive HFT should not inherit their odium.

Mechanics of each strategy, worked examples, revenue sources for liquidity provision, and the latency-arbitrage dispute in full: references/strategies.md.

What the evidence showed (as of 2011)

Six of eight HFT-focused papers found no evidence of harm to market quality; most found improvements in liquidity and reductions in short-term volatility. AT studies (Hendershott et al. on NYSE) found algorithmic trading causally narrowed spreads and improved quote informativeness. The dissenting findings are real and worth stating: Jovanovic & Menkveld found HFT middlemen are better informed than average investors and can exacerbate adverse selection under some conditions, and Kirilenko et al. found HFT amplified volatility during the flash crash.

Two honest limits on all of it: the empirical work covers lit markets only (no data exists for automated trading in OTC/internalization space), and no dataset lets researchers identify HFT order-by-order, so identification rests on venue member categorizations or statistical proxies. Market-share estimates from that era vary enormously — 13% to 40% for Europe, 40% to 70% for the U.S. — which itself tells you how soft the measurement is.

Paper-by-paper summaries with methods and limitations, plus the market-sizing tables: references/evidence.md.

Why the U.S. and Europe are different

This distinction does most of the work in any regulatory discussion, and getting it wrong is the most common failure mode when people import U.S. HFT commentary into a European context.

• U.S.: Reg NMS codifies the NBBO, and the Rule 611 trade-through / order-protection rule forces venues to route away or cancel rather than execute worse than the NBBO. Best execution is effectively outsourced to venues. This inter-linkage is what makes flash orders, latency arbitrage against a stale consolidated tape, and cascading cross-venue effects possible.
• Europe: MiFID imposes a principles-based best-execution obligation on investment firms (price, cost, speed, likelihood of execution and settlement, size, nature), with no pan-European NBBO and no re-routing obligation. Share-by-share volatility interruptions have existed for decades.

The paper's central policy claim follows: many "HFT problems" are artifacts of U.S. market structure, and importing U.S. remedies into Europe risks fixing a problem that isn't there. Treat that as the authors' argued position rather than a fact — but do keep the structural distinction, which is simply true.

Its concrete proposals: prefer coordinated volatility safeguards (a second, security-specific circuit-breaker band over a ~5-minute horizon, triggering call auctions, coordinated across venues via the most liquid market) over market-making obligations (which firms will breach and pay fines for rather than "catch a falling knife") and over minimum order lifetimes / order-to-trade ratio caps (which trap orders as free options, invite gaming, and impede risk management). Full reasoning, the U.S. and EU initiative histories, and the systemic-risk requirements the paper places on trading firms, venues and regulators: references/regulation.md.

Recurring confusions worth correcting

• Flash orders ≠ the flash crash. Similar names, unrelated phenomena. Flash orders are an order type exploiting a trade-through exception; the flash crash was a May 6 2010 market event.
• "HFT is front-running." Front-running means trading ahead of a client order you were entrusted with. HFT liquidity detection infers other participants' intentions from public order-book patterns. Order anticipation may deserve scrutiny on its own terms; conflating it with front-running imports an agency-breach that isn't present in proprietary trading.
• Latency arbitrage is contested, not established. Critics describe trading against stale NBBO quotes; Tradeworx's rebuttal is that order-book priority is unaffected by NBBO latency, so there is nothing left to trade against — the real mechanism runs through intermarket sweep orders. Present both sides, and note the debate is NBBO-specific and therefore largely inapplicable to Europe.
• "HFT withdrew liquidity in the crash, so obligate them to quote." During the flash crash registered market makers with obligations also stopped quoting, or invoked technical difficulties. Obligations did not hold where it mattered — which is the paper's main argument for safeguards over obligations.
• Adverse selection cuts the other way from the usual story. HFT market makers on lit venues quote without knowing their counterparties and bear adverse-selection cost. Internalizers and dark venues do know counterparty identity and can select uninformed flow. The paper's point: regulatory attention concentrated on lit-market HFT while a ~40% OTC market share went largely undiscussed.
• High cancellation rates are not per se manipulation. Continuous requoting against moving reference prices is how voluntary liquidity provision works. Quote stuffing — flooding a venue with orders to degrade rivals' processing — is the abusive case, and it is distinguished by intent and effect, not by the raw order-to-trade ratio.

Answering well

Say which claims come from the paper and which are yours. Attach numbers to their source and date ("Brogaard's Nasdaq sample, 2010: HFT on one side of 68% of dollar volume") rather than floating them as current facts — HFT market shares, venue fee schedules and rebate totals have all moved since. When a question turns on present-day rules, check references/since-2011.md and say plainly which parts of the 2011 analysis have been overtaken.

The most valuable thing this skill offers is usually not a fact but a reframing: separate the technology from the strategy, the function from the institution, and the market structure from the behaviour. Lead with that when the question is muddled.

Reference files