High-frequency trading (HFT) is a subset of automated, proprietary trading defined by extreme speed and extreme order turnover: firms use co-located servers and custom hardware to submit, cancel, and execute enormous numbers of orders over horizons measured in microseconds, almost always ending the day flat. HFT is not a single strategy but a technological regime — a way of competing — layered on top of strategies (market making, arbitrage) that are themselves centuries old. Its core tension is that the same speed that lets HFT firms tighten spreads and link fragmented markets also fuels a zero-sum "arms race" for raw latency that, by the best academic measures, taxes the very investors it claims to serve.

What it is (defining characteristics)

The SEC's working definition lists five hallmarks: (1) extraordinarily high-speed, sophisticated order-generation systems; (2) use of co-location (renting rack space inside the exchange's data center) and direct proprietary data feeds to minimize latency; (3) very short holding periods; (4) submission of numerous orders cancelled shortly after entry; and (5) ending the day in as flat a position as possible (SEC, Concept Release on Equity Market Structure, 2010; CRS Report R43608). Crucially, HFT is a form of algorithmic trading, not a synonym — a pension fund's hours-long VWAP execution algorithm is algorithmic but not HFT. Speed is the differentiator: latency is fought down to the microsecond and nanosecond via co-location, kernel-bypass networking, FPGAs/ASICs, and even microwave/laser links that beat fiber over land routes.

The main strategy families

• Electronic market making. Post resting bid and ask quotes, earn the spread and exchange rebates, manage inventory by re-quoting constantly. This is the largest and most economically defensible HFT activity — it supplies the displayed liquidity most investors trade against.
• Statistical / index arbitrage. Exploit transient mispricings between correlated instruments (a stock vs. its ETF, futures vs. cash index, dual-listed names).
• Latency arbitrage. A distinct, more contested category: exploit the symmetrically public signal that a price has just moved on one venue to pick off stale quotes on another before they update. Budish, Cramton & Shim (2015) stress this is rents from public information, separate from the private-information adverse selection of classic microstructure theory.
• Order anticipation / structural strategies. Detecting large institutional orders and trading ahead. The legitimacy here shades into the prohibited (spoofing, layering, quote stuffing).

How it's used in practice

HFT is the dominant executor of modern equity volume rather than a discretionary "view." Commonly cited estimates put HFT at roughly 50–60% of U.S. equity volume, rising toward ~75% on stressed days (CRS R43608; Wikipedia, citing TABB/industry estimates) — figures that are estimates, not measured, because no regulator publishes a clean HFT volume tally. Economics are razor-thin per trade — commonly cited at fractions of a cent per share — so the model is high-volume, high-Sharpe, near-flat overnight risk. The barrier to entry is capital-intensive infrastructure (co-location, feeds, FPGA engineering, network engineering) rather than clever signals, which is why participation is highly concentrated: Aquilina, Budish & O'Neill (2022) find the top 6 firms win/lose over 80% of latency-arbitrage races.

For a discretionary or swing trader, HFT is not a tool to deploy — it is part of the environment you trade in. Its practical relevance is understanding why displayed liquidity can be illusory (quotes that vanish on a marketable order), why stop orders can be "run," and why microstructure noise dominates sub-second price action.

Adoption, debate & evidence

HFT is universally adopted in developed electronic markets and genuinely contested in its welfare effects. The honest split:

• Where evidence is broadly favorable: A large literature (summarized in the SEC's 2014 Equity Market Structure Literature Review Part II) finds HFT market making is associated with narrower bid-ask spreads, lower transaction costs, and better price efficiency in normal conditions. Most studies do not support the strongest "HFT is pure predation" claims.
• Where evidence is unfavorable: The arms-race critique is well-quantified. Aquilina, Budish & O'Neill (2022, QJE) measure latency-arbitrage "races" at ~1 per minute per symbol, lasting a modal 5–10 millionths of a second, accounting for ~20% of volume, imposing roughly a 0.5 basis-point tax on trading and ~one-third of the effective spread; eliminating it would cut investors' cost of liquidity by ~17%, with global sums on the order of $5 billion/year. Budish, Cramton & Shim (2015) argue this is a market-design flaw of the continuous limit order book and propose frequent batch auctions (discrete, e.g. every 100ms) to convert speed competition into price competition.
• Profitability is decaying. A widely cited Purdue estimate puts aggregate HFT profits falling from ~$5B (2009) to ~$1.25B (2012) as strategies commoditized and the speed race raised costs without enlarging the prize — "competition has not affected the size of the arbitrage opportunities, only how fast you must be to capture them" (Budish et al., 2015).

The 2010 Flash Crash sharpened the systemic-fragility concern: the joint SEC/CFTC report found HFT did not cause the crash but magnified it as automated liquidity withdrew. Enforcement (e.g. spoofing cases, quote-stuffing fines) targets specific manipulative tactics, not HFT as such.

Strengths & limitations

Strengths: tighter spreads and deeper top-of-book liquidity in calm markets; faster incorporation of information into prices; tighter cross-venue and cross-asset linkages. Limitations: liquidity is fair-weather and can evaporate in stress; the latency race is socially wasteful (real resources spent on speed that is zero-sum); a measurable tax on resting liquidity; and concentration creates systemic and competitive concerns. The #1 misuse / misconception is conflating HFT with manipulation — most HFT is legal market making; spoofing and quote stuffing are illegal subsets, not the definition. A second common error is assuming retail or swing traders can "compete on speed" — the infrastructure gap makes that hopeless and irrelevant to their edge.

Sources

• SEC, Equity Market Structure Literature Review Part II: High Frequency Trading (March 2014) — sec.gov/marketstructure/research/hft_lit_review_march_2014.pdf
• Congressional Research Service, High-Frequency Trading: Background, Concerns, and Regulatory Developments, R43608 — everycrsreport.com/reports/R43608.html
• Budish, Cramton & Shim, "The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response," QJE 130(4), 2015 — academic.oup.com/qje/article/130/4/1547/1916146
• Aquilina, Budish & O'Neill, "Quantifying the High-Frequency Trading 'Arms Race'," QJE 137(1), 2022 (also BIS WP 955, NBER w29011) — academic.oup.com/qje/article/137/1/493/6368348
• Wikipedia, "High-frequency trading" (for share-of-volume, Purdue profit estimate, Flash Crash, enforcement) — en.wikipedia.org/wiki/High-frequency_trading

Disputes flagged: (a) Net welfare effect of HFT is genuinely contested — spread/efficiency benefits (SEC review) vs. the quantified latency-arbitrage tax (Aquilina/Budish/O'Neill); both can be true simultaneously. (b) Share-of-volume figures (50–75%) are industry estimates, not regulator-measured. (c) HFT's role in the Flash Crash is "amplified, did not cause" per the official report, not consensus that HFT is the trigger.