Skip to content
Menu
Finabase Learn

How Wall Street really works Lesson 9 of 14

Hedge funds and AI · Quants and high-frequency trading

High-frequency trading explained: strategies, scale and rules

What high-frequency trading is by the SEC’s own description, its four strategy families, how much of the market it is, how it rose after decimalization, and its limits.

On this page
  1. What counts as high-frequency trading
  2. The four strategy families
  3. How big it is
  4. How it grew: decimals, electronic markets and speed
  5. What the research says about its effects
  6. Where high-frequency trading fails
  7. How Finabase does it
  8. Frequently asked questions
  9. Summary
  10. Check your understanding
  11. Exercise

High-frequency trading (HFT) is automated trading by firms that use their own capital, very fast computers and the quickest available connections to exchanges to buy and sell large numbers of shares, usually holding each position for seconds or less and ending the day with almost nothing on the books. It is not one strategy but a set of them, from quoting both sides of the market all day to racing to trade against prices that are a few millionths of a second out of date.

High-frequency firms are a large share of the US stock market's trading and much of the liquidity that ordinary investors trade against. This pillar lesson explains what HFT is in the SEC's own description, the four families of strategy, how large it is, how it grew after prices went decimal and trading went electronic, and the concerns about it. The lessons after it go deeper into speed and into whether it costs individual investors money.

What counts as high-frequency trading

There is no legal definition of HFT in US securities law. When the SEC opened a broad review of market structure in January 2010, its Concept Release on Equity Market Structure said the term was new and not yet clearly defined. It described the firms involved as professional traders acting for their own account, whether a stand-alone proprietary trading firm, the trading desk of a broker-dealer or a hedge fund, whose strategies generate a large number of trades each day. It then listed five characteristics often attributed to them:

  1. extraordinarily fast and sophisticated programs for generating, routing and executing orders;
  2. colocation services and individual data feeds from exchanges, used to cut delays (latency);
  3. very short time frames for opening and closing positions;
  4. many orders that are cancelled shortly after they are sent; and
  5. ending the trading day as close to flat as possible, without significant unhedged positions overnight.

The SEC staff's 2014 review of the research added a warning: a firm need not show all five traits to be a high-frequency trader, and HFT is not one thing. Studies with data that identified HFT accounts on Nasdaq found that more than half of HFT activity came from aggressive orders that took liquidity, not from resting quotes that provided it.

HFT is also a subset of something much larger. Algorithmic trading includes the execution algorithms that pension funds and mutual funds use to split big orders, and the smart order routers every broker runs. Those are computer-driven too, but they serve investors holding positions for months, not traders holding for seconds.

The four strategy families

The Concept Release grouped HFT strategies into four broad types. They overlap, and one firm may run all of them.

Four families of high-frequency strategy (SEC Concept Release, 2010) Passive market making Rests bids and offers in the book. Earns: the spread, plus exchange rebates. Risk: being picked off by better- informed or faster traders. Orders: many, most cancelled quickly. Arbitrage Trades price gaps between related things, e.g. an ETF and its stocks. Earns: the gap, mostly by taking liquidity; positions are hedged. Risk: the gap closes before it fills. Structural Exploits how the market is wired, e.g. a venue still quoting a stale price after the market has moved. Earns: the stale price’s error. Debate: who bears that cost. Directional Bets on a short-term price move. Some aid price discovery; the SEC flagged two for comment: order anticipation and momentum ignition. Risk: an ordinary trading loss. Summary in Finabase’s words of the strategy types in SEC Release 34-61358, section IV.B.1.
The four families of HFT strategy in the SEC’s 2010 Concept Release, summarized in Finabase’s words.

Passive market making. The firm posts bids and offers that wait in the order book for others to trade against. It earns the spread when it buys at its bid and sells at its offer, plus any rebate the exchange pays for resting liquidity (the maker-taker model, which has its own lesson). Because prices keep changing, the firm constantly replaces its quotes. The Concept Release noted that layering the book this way can produce cancellation rates of 90% or more, with many orders living for a second or less. This is the electronic successor to the human market maker described in Market makers: who quotes both sides and how they get paid, with one important difference: most high-frequency market makers have no obligation to keep quoting when markets turn rough.

Arbitrage. The firm trades price differences between related things: an exchange-traded fund and the basket of stocks it holds, a stock index future and the index's stocks, or the same stock on two venues. Arbitrage mostly takes liquidity, because the gap must be captured before it closes, and the positions are hedged, though they can last for days. This trading is what keeps an ETF's price close to the value of its holdings.

Structural. These strategies exploit how the market is wired. The example the SEC gave: a firm with the fastest market data could profit by trading with participants still offering executions at stale prices, because their view of the market is a few microseconds behind. The SEC staff's 2020 report on algorithmic trading, citing interviews by Donald MacKenzie, notes that reported response times for this kind of strategy fell from about 5 microseconds in 2011 to about 300 nanoseconds in 2018. This is the "latency arbitrage" measured in Does high-frequency trading hurt retail investors? The evidence.

Directional. Some HFT strategies simply bet that a price will move up or down in the next seconds or minutes, and many of these help prices adjust to new information. The Concept Release asked for comment on two kinds it said may present serious problems: order anticipation, which tries to detect a large buyer or seller and trade ahead of it, and momentum ignition, which tries to start a quick price move in order to profit from it. Entering orders one never intends to execute, to move a price, is spoofing, which is illegal.

How big it is

The Concept Release said estimates of HFT's share of US stock volume "vary widely, though they typically are 50% of total volume or higher", and called HFT a dominant component of market structure. A 2020 paper from the UK's Financial Conduct Authority repeated that by some estimates HFT firms account for over 50% of trading volume. No US regulator publishes an official, current share, partly because, as the SEC staff's 2014 review explained, public trade data does not say who traded, so researchers can only measure HFT precisely with special datasets from exchanges or regulators.

Concentration is easier to see than size. In the FCA's study of the London Stock Exchange over nine weeks of 2015, the top three firms won about 55% of the latency races and the top six about 82%.

How it grew: decimals, electronic markets and speed

HFT is a product of the market structure built in the 2000s. Three changes mattered most.

Decimal prices. Until 2000, US stocks were quoted in fractions of a dollar; for most stocks the minimum increment was a sixteenth or a thirty-second of a dollar. In June 2000 the SEC ordered the exchanges and Nasdaq to phase in decimal pricing, and the change was complete by April 9, 2001. Quoting in pennies narrowed spreads: the SEC staff's 2012 report to Congress summarized a study finding that effective spreads in large NYSE stocks fell from 8 cents to 4 cents around the switch (Does high-frequency trading hurt retail investors? The evidence has the full comparison). A narrower spread meant less profit per share for whoever provided liquidity, which rewarded firms that could trade many more shares at lower cost: automated ones.

Electronic, fragmented markets. Regulation NMS, adopted by the SEC in 2005, protected the best displayed quotes across all exchanges and rewarded venues that could respond automatically; its rules are covered in Order routing and best execution: NBBO, Rule 611, PFOF and 606. Trading spread across many electronic venues and the old floor-based model shrank.

Trading in NYSE-listed stocks, 2005 against 2009 (SEC, 2010) Measure 2005 2009 Change Consolidated trades a day 2.9 million 22.1 million +662% Consolidated shares a day 2.1 billion 5.9 billion +181% Average trade size 724 shares 268 shares −63% NYSE’s own share of volume 79.1% 25.1% −68% NYSE speed, small marketable orders 10.1 seconds 0.7 seconds −93%
From the SEC’s Concept Release on Equity Market Structure (January 2010), figures 1–5. Trades, shares and trade size compare 2005 with January–October 2009; market share and speed compare January 2005 with October 2009.

The figures the SEC put at the front of its Concept Release show how fast the change came. In four years the number of trades in NYSE-listed stocks rose more than sevenfold while the average trade shrank by almost two-thirds, the signature of orders being sliced small and traded by machines. NYSE's own share of trading in its listed stocks fell from 79.1% to 25.1% as volume moved to other electronic venues, and its average execution time for small marketable orders fell from 10.1 seconds to 0.7.

Speed. As exchanges automated, the time to accept and execute an order fell below a millisecond; the Concept Release cited published averages of 294 microseconds at Nasdaq and 320 at BATS in 2009. Once everyone was fast, being faster than the others became the advantage, which is where colocation and private data lines come in.

What the research says about its effects

The SEC staff's 2020 report summarized a large literature: studies have generally concluded that HFT may have improved standard measures of market quality during normal periods. Competition among electronic market makers narrows spreads, and liquidity-taking strategies can help prices absorb information quickly. The same report notes the other side: some HFT strategies may raise costs for other traders, and the fastest traders can trade against resting orders before those orders are updated.

The 2014 staff review put the distinction plainly. Passive HFT strategies generally appear to help market quality, for example by narrowing spreads; aggressive strategies raise more questions, helping price discovery over very short periods while imposing adverse-selection costs on slower traders who leave orders in the book. The evidence on retail and institutional costs is weighed in Does high-frequency trading hurt retail investors? The evidence.

Where high-frequency trading fails

  • Liquidity is not guaranteed. High-frequency market makers are usually the main suppliers of quotes, but the SEC staff noted in 2020 that they usually have no obligation to provide liquidity and may pull back when conditions are uncertain. In the The 2010 Flash Crash: what happened on May 6 and what changed, research summarized by the SEC staff concluded that HFTs did not trigger the crash but that their responses to the unusually large selling made the volatility worse.
  • Costs fall on slower traders. When a fast firm trades against a quote that has not yet been updated, the firm that posted the quote loses. Market makers price that risk into wider spreads, so everyone pays a little.
  • An arms race. Firms spend heavily on microwave networks, hardware and colocation to be a few microseconds faster than one another. The FCA study estimated the sums at stake in such races at about $5 billion a year across global equity markets.
  • Technology risk. Automated systems can do damage quickly when they fail. On August 1, 2012, a software error at one market-making firm, Knight Capital, cost it more than $460 million in about 45 minutes (Knight Capital 2012: the 45-minute glitch that lost $460 million).

How Finabase does it

Finabase does not trade, route orders or run any high-frequency strategy, and it does not show the order book or a quote-by-quote feed. Symbol pages show the last price, the time of the last trade, the day's range and volume, and the one-day chart on each symbol page plots every minute of trading, where you can see the spikes in activity at the open and close that much of this trading creates. For depth and the tape, see Level 2 and time and sales: reading depth and the tape, which explains what those screens show and where their data comes from.

Frequently asked questions

Yes. Trading quickly with one's own money, posting and cancelling quotes, and using colocation and exchange data feeds are all legal in the US. Particular practices are not: entering orders with no intent to execute them in order to move a price is spoofing, a form of manipulation.

How much of the stock market is high-frequency trading?

No one publishes an official figure. The SEC's 2010 Concept Release said estimates were typically 50% of volume or higher, and later studies have used similar estimates, but measuring it requires data that identifies who traded.

Is high-frequency trading the same as algorithmic trading?

No. Algorithmic trading is any trading in which a computer decides how, when or where orders are sent, including the execution algorithms of long-term investors. HFT is the subset that trades for its own account at very short horizons and depends on speed.

Do high-frequency traders front-run my orders?

Trading ahead of a customer's order using knowledge of that order is front-running, and a broker or market maker that does it breaks the rules. What fast firms can do is react within microseconds to orders and trades that are already public, which is legal but is one of the concerns covered in Does high-frequency trading hurt retail investors? The evidence and Front-running.

Can an individual investor do high-frequency trading?

Not in any practical sense. It requires servers inside exchange data centers, direct exchange data feeds, specialized hardware and capital, and its profits per trade are fractions of a cent.

Summary

High-frequency trading is automated, short-horizon trading for a firm's own account that depends on speed. The SEC describes it by five traits rather than a legal definition, and groups its strategies into passive market making, arbitrage, structural and directional families. It grew with decimal prices and electronic, fragmented markets: between 2005 and 2009, trades in NYSE-listed stocks rose 662% while the average trade shrank 63%. Estimates put it at half or more of US stock volume. Research generally finds that its market making narrowed spreads in normal times, while its fastest strategies impose costs on slower traders and its liquidity cannot be counted on in a crisis.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. Which of these is NOT one of the five traits the SEC’s 2010 Concept Release said are often attributed to high-frequency trading firms?

    1. Ending the trading day as close to flat as possible
    2. Submitting many orders that are cancelled shortly after they are sent
    3. Holding concentrated positions for several months
    4. Using colocation and individual exchange data feeds
    Show the answer

    C. Holding concentrated positions for several months HFT is defined by very short holding periods. The five traits are fast programs, colocation and direct feeds, very short time frames for positions, many cancelled orders, and ending the day close to flat.

  2. A firm posts bids and offers all day in thousands of stocks, earning the spread and exchange rebates, and cancels most of its orders. Which strategy family is this?

    1. Passive market making
    2. Arbitrage
    3. Directional
    Show the answer

    A. Passive market making Passive market making rests quotes in the book and earns the spread when both sides trade. Because prices keep moving, the quotes are constantly replaced, which is why the SEC noted cancellation rates of 90% or more for this strategy.

  3. Between 2005 and 2009 the number of trades in NYSE-listed stocks rose 662% while the average trade size fell from 724 shares to 268. What does that pattern mainly show?

    1. Many more small investors started buying stocks
    2. Orders were increasingly sliced into small pieces and traded by machines
    3. The NYSE raised its fees on large trades
    Show the answer

    B. Orders were increasingly sliced into small pieces and traded by machines Many more, much smaller trades is the signature of automated trading: execution algorithms slicing large orders and high-frequency firms trading constantly in small sizes. It came as trading moved from NYSE’s floor to many electronic venues.

  4. What did research summarized by SEC staff conclude about high-frequency traders in the May 6, 2010 Flash Crash?

    1. They started the crash with a large sell order
    2. They did not trigger it, but their responses to the heavy selling worsened the volatility
    3. They were absent from the market that afternoon
    Show the answer

    B. They did not trigger it, but their responses to the heavy selling worsened the volatility The fall began with a large futures sell program. High-frequency traders first bought, then sold rapidly to each other and others, adding to the volatility. Because they have no duty to keep quoting, their liquidity cannot be counted on in stress.

Exercise

One task to do in Finabase. Nothing is graded or saved.

Open SPY’s one-day chart and look at the volume bars across the session. Find the minutes with the most volume and note when they happened. Then compare a typical midday bar with those in the first and last few minutes of trading.

Open SPY’s one-day chart
What you should find

Volume usually spikes at the open and the close, with a quieter middle of the day, and much of that activity is automated. A one-minute bar hides far more than it shows: high-frequency market makers replace their quotes constantly, with cancellation rates of 90% or more and many orders living a second or less.

Sources and further reading

  1. SEC — Concept Release on Equity Market Structure, Release No. 34-61358 (January 14, 2010), sections III and IV.B
  2. SEC Division of Trading and Markets staff — Equity Market Structure Literature Review, Part II: High Frequency Trading (March 18, 2014)
  3. SEC staff — Staff Report on Algorithmic Trading in U.S. Capital Markets (August 5, 2020)
  4. SEC staff — Report to Congress on Decimalization, as required by Section 106 of the JOBS Act (July 2012)
  5. SEC — Press release 2013-222 on Knight Capital’s August 1, 2012 trading incident (October 16, 2013)
  6. Matteo Aquilina, Eric Budish and Peter O’Neill — Quantifying the High-Frequency Trading "Arms Race", FCA Occasional Paper 50 (January 2020)