High-Frequency Trading: A Mixed Verdict
High-frequency trading is automated trading in which the edge comes from speed rather than insight — reacting to prices and to other participants' orders faster than anyone else, holding positions briefly and taking a small amount many times. The advantage is bought with capital: proximity, faster feeds and hardware.
How it works
High-frequency trading (HFT) is automated trading in which the edge comes from speed. Software reacts to prices and to other participants’ orders in microseconds, holds positions briefly, and takes a small amount many times.
The edge is latency, and latency is bought. Space beside the matching engine, direct data feeds, and hardware built for one narrow job.
That framing is the honest one. It is an engineering advantage bought with capital, not a superior read of the market.
It is a capital business rather than a retail one. The costs are fixed, large and continuous, so it concentrates among a few well-financed participants.
Much of it is automated market making. Quotes are posted on both sides of the order book, earning the bid-ask spread when both sides fill.
The rest is two familiar jobs: acting on price differences between related instruments or venues, and reacting to book changes faster than whoever caused them.
The fair balance sheet
So the honest verdict is mixed, not villainous. Automated market making has narrowed quoted spreads, and the same liquidity withdraws in the moments it is most wanted, and both of those are true at once.
Take the good side first, because it usually gets skipped. Spreads are tighter than they were, small orders fill more easily, and you collect that on every trade.
On a daily chart you are not competing with it. The edge being contested lives inside a single second.
Now the other side. That liquidity is quoted, not committed. It is cancelled instantly when the risk of trading against better-informed order flow rises, so depth thins exactly when depth matters.
The economics are paid in fractions. A fraction of a spread collected millions of times is a business, and some venues pay a rebate for resting orders while charging to take them.
That structure is invisible to most retail traders, and it shapes routing — including into dark pools and through payment for order flow.
Much of what prints on the tape is machines trading with machines. The volume column records transactions, not conviction.
In practice
None of it survives onto a daily chart. A microsecond advantage resolves long before a daily bar closes. On a one-minute chart it matters a great deal, where a fraction of a spread is a large share of the move.
Where a longer-term trader does meet it is the opening gap. Quoting is thinnest in the first minutes, so spreads widen and fills get worse. It carries no obligation to be present, which is one reason circuit breakers exist.
Resting orders are visible information, and that includes stops. A stop loss in the book is an order like any other, and clusters of them are inferable from level 2 data.
Say that factually, without attributing motive. A hidden order is the available answer, and deception belongs on the market manipulation page.
The cost that decides your outcome is your own. On this site’s shared history a round trip is 2% of a median bar’s range and 45% of the smallest — trivial over a swing, decisive over a scalp.
What to change in response
Three adjustments follow, and none involves getting faster. The first is to use limit orders rather than market orders wherever the situation allows it. A market order accepts whatever the book offers at that instant; a limit order sets the worst price you will take.
The second is to avoid the first minutes of the session. Quoting is thinnest then, spreads are widest, and the opening has not yet resolved into a stable trading range.
The third matters most. Choose a holding period long enough that a fraction of a spread stops deciding the outcome. If the move you are reaching for is barely larger than your round-trip cost, the problem is the horizon rather than the execution.
Competing on speed is not available to you, and that is a constraint rather than a grievance. The response is to trade where speed is not the variable being contested.
What high-frequency trading is not
- It is not a strategy you can run. The entry cost is infrastructure, not software you install.
- It is not a synonym for algorithmic trading. Only a small subset competes on latency.
- It is not market manipulation by definition. Speed is not deceit.
- It is not why a swing trade lost. Above a minute, the explanation is almost always duller.
When it fails
In a quiet market it is most of the activity, which is when it is least interesting. Machines swapping inventory inside a range says nothing about direction.
The first real failure is withdrawal under stress. Quotes are cancelled faster than they are replaced, depth collapses, and price travels further per unit of size than anyone modelled.
The second is that the benefit is unevenly spread. Narrow quotes in heavily traded names say little about a thin one, where the improvement never really arrived.
The third is the trader who tries to read it. Watching the book for machine behaviour at human speed is reading a record of decisions already made.
The fourth is misattribution. A stop triggered by an ordinary move gets blamed on machines, and the real cause — position size, or a stop placed inside the noise — goes unexamined.
The fifth is the argument itself. Both the outraged version and the reassuring one sell a simple story, when the benefit and the fragility are one mechanism seen from two sides.
The original data
research/broker-coverage.json scanned the 31,760 videos in research/search-study-corpus.jsonl.
Seven titles carry “high frequency”, at a median of 121,256 views — one of the highest medians
measured anywhere on this site. “Payment for order flow”, the mechanism by which your order actually
reaches one of these firms, returns zero.
research/series-measurements.json, via site/measure_series.py, puts a round trip at 2% of a
median bar. A firm at a small fraction of that cost has different economics, not the same game
faster. Measure your round-trip cost as a share of the move you are reaching for; if that share is
large, the fix is a longer horizon, not a faster one.
Related
Market makers are what most of this activity actually is, which is where the economics are set out. The order book is the surface it operates on, and the only part of it you can see. And liquidity is the thing being supplied and withdrawn, which is where the fair verdict comes to rest.
I am not competing with a machine that measures its edge in microseconds, and I stopped pretending otherwise a long time ago. On a daily chart the move I am trying to capture is far larger than anything speed can take from me. The trades where it did hurt were the ones I was rushing anyway, entering on a market order into a thin open for reasons I could not really name. Slowing down fixed more of that than any tool ever has.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.