High Frequency Trading: The Speed Business
High frequency trading is algorithmic trading pushed to the physical limit: positions held for seconds or less, decisions made and orders placed in microseconds, and the whole business resting on being a few millionths of a second faster than the next firm. It is the most misunderstood layer of the market, described either as a parasite skimming every trade or as the invisible hand that halved everyone's trading costs. Both descriptions contain a piece of the record and neither survives contact with the full numbers. What the record shows is a specific business model with specific revenues: speed buys the right to be first in line at a known price, and being first at a known price earns fees, rebates, and information advantages that add up in fractions of a cent, millions of times a day. Understanding what that speed is actually for, who pays for it, and what the market gained and lost when it arrived, turns a vague villain story into a readable feature of the tape.
The Race in One Sentence
The entire industry exists because price changes propagate at finite speed, and whoever sees the new price first can act on it before the rest of the market adjusts. Every other feature of the business is an engineering consequence of that one fact. Firms house their computers inside the same data centers as the exchanges, paying for rack space measured in meters of cable, because light through fiber takes about five microseconds per kilometer and thirty kilometers of distance is an eternity in a race decided in microseconds. The firms spend fortunes on shorter routes between cities, on microwave towers that beam data in straight lines above the fiber curves, and on faster network cards, all to shave single-digit microseconds off a round trip. Nothing in this world is clever about markets. It is clever about physics and logistics, applied relentlessly to one narrow question: who hears the latest price first?
The prize for hearing it first is not ownership of the market. The prize is position in a race that repeats millions of times a day. When news or flow shifts the fair price of a stock, the fastest firms reprice their quotes before slower participants can respond, and in that gap sits the entire edge: they buy from whoever is still selling at the old price, or sell to whoever is still buying at it, and step aside before the slower world catches up. The margins are microscopic, which is why the volumes must be enormous and the costs of failure absolute. A high frequency firm that is second is not slightly less profitable. It is out of business.
Where the Speed Comes From
Speed has suppliers, and the supply chain is a market of its own. Exchanges sell colocated rack space and premium data feeds; the premium feed delivers the same prices as the public one, just sooner, which is enough. Telecom carriers sell dark fiber and microwave links between New Jersey and Chicago, and the records of that arms race are public: routes were shortened twice, first to around 13 milliseconds round trip, then to under 9, because whoever bought the shorter pipe outran the buyers of the longer one until the next shortening. None of this spending improves price discovery in any direct sense. It is purely positional spending, the financial equivalent of two neighbors bidding up the price of the same hilltop, and its bills are ultimately woven into the cost structure of every participant's fill.
| Claim about HFT | What the record shows |
|---|---|
| It widened trading costs | Spreads on liquid names compressed to about a penny, the tightest in market history |
| It provides deep liquidity | Displayed size thins fast: quotes cancel in microseconds under pressure |
| It is riskless for markets | Flash events now propagate in milliseconds across correlated venues |
| It is guaranteed profit | Firms die when their speed edge erodes; second place earns nothing |
A Worked Example: One Million Shares of Rebates
Speed is only half the business; the other half is the fee schedule. Exchanges pay a rebate to traders who add resting liquidity and charge a fee to traders who take it. A typical schedule pays roughly two tenths of a cent per share to the maker and charges roughly three tenths from the taker. Those fractions look like rounding errors until the volume multiplies them. Consider one high frequency desk providing liquidity in a single liquid stock. It posts bids and offers all day, and over a session it is filled on one million shares of maker flow. One million shares at a rebate of 0.002 dollars per share is 2,000 dollars for the day. Multiply by roughly 250 trading days and one desk, in one name, collects about 500,000 dollars a year in rebates alone, before a cent of directional profit is counted. Add several dozen liquid names, several desks, and the fee schedule becomes the business model: the firm is not primarily trading direction, it is operating a toll position at the top of the book, and speed is what keeps the toll position first in line.
The same arithmetic explains the behavior pattern that confuses human observers. Why does displayed size vanish the instant price starts to move? Because the maker rebate only pays when the quote is not filled adversely: a resting bid that gets run over by a falling market costs the firm its capital, and no rebate covers that. The rational response is to cancel the instant conditions turn, at microsecond speed. The liquidity the desk provides is real, but it is conditional liquidity: present when the market is calm, gone when it is needed. Retail traders experience this as the wick that slides through an empty book, or the stop that fills two ticks past where the chart said the size was. The size was never a commitment. It was a toll booth with a trapdoor.
The Market Impact, Both Sides
The measured gains are real and mostly accrued to ordinary participants' costs. Spreads on liquid equities compressed from several cents to around a penny as high frequency makers competed for rebates, and the implied cost of a round lot trade fell with them. Price discovery across the many fragmented venues tightened: prices agree between markets faster than any human era could manage, and the arbitrage that keeps a stock's futures, ETF, and cash quotes in line is now continuous rather than occasional. Those are documented, quantified benefits, and they arrived exactly as the industry promised they would.
The measured costs are equally real and harder to see on a quote screen. Displayed liquidity is thinner and less trustworthy than spreads suggest, because conditional liquidity evaporates under precisely the conditions where depth matters. Fragile structure amplifies stress: in the 2010 flash crash, market-wide prices moved 5 percent and fully reversed inside minutes, with high frequency makers withdrawing en masse at the worst moment and some trades executing at absurd prices in the vacuum. Adverse selection rose for everyone: any participant whose orders rest in the market is now trading against counterparts who know something they do not, which shows up as the consistent slippage between an order's arrival price and its fill. The honest ledger has profits and losses on both sides, and which side dominates depends on the order: small retail market orders arguably gained the most, while anyone working size or resting orders for any length of time pays more of the toll.
The speed layer is permanent infrastructure now, priced into every fill whether acknowledged or not. The next lesson turns from the industry to its residue on the charts: the specific, repeatable marks machine flow leaves in candlesticks, and how to read them without mythology.
High Frequency Trading Questions
How fast is high frequency trading, in actual numbers?
Order-to-response cycles run in the low microseconds: a few millionths of a second from seeing a price change to having new orders in the book. Round-trip latencies between the major trading centers dropped from around 13 milliseconds to under 9 through microwave and shortened fiber routes. The relevant comparisons are all physical: light through fiber travels roughly 5 microseconds per kilometer, which is why rack location inside the exchange data center is worth real money.
Do high frequency firms ever lose money?
They lose their entire business whenever the speed edge erodes. The margins per trade are fractions of a cent and depend on being first; a firm that is second at the same quote earns nothing, so competitive shortening of routes and hardware refreshes are existential, not optional. Individual bad days happen too, most famously when models misread conditions and firms trade into events they were built to avoid.
Is high frequency trading legal?
Yes, within extensive regulation. Speed itself is not an offense, and being faster at public price information is not insider trading. The regulated boundaries concern manipulation: quote patterns designed to mislead, such as placing and canceling orders to create false impressions of supply or demand, are illegal, and enforcement cases exist. The legal line separates competing on speed from fabricating the picture other participants see.
How does high frequency trading affect a small retail order?
Mostly through the price it receives. A small market order typically fills at a penny spread, which is better than at any time before the machine era, and fills fast because someone always answers at the touch. The costs to retail show up less in that fill and more in the environment: thinner displayed depth at exactly the moments size matters, and occasional air pockets in stressed sessions when the conditional liquidity withdraws all at once.