What Algorithmic Trading Actually Is
Algorithmic trading is the execution of orders by computer programs following a defined set of rules, and it now accounts for the majority of volume in the largest markets. The phrase covers two jobs that look nothing alike from the outside. One job is delivery: taking an order that already exists and getting it filled with the least cost and the least notice, slicing it into pieces and working them through the session. The other job is decision: watching market data and deciding what to buy or sell, at what price, at what moment, inside a risk budget. Both are algorithms. Both follow rules written in advance. But the delivery program serves a decision already made, while the decision program is the decision. Nearly everything a chart watcher sees when machines dominate a tape comes from one of these two jobs, and keeping them separate is the first step toward reading a machine-dominated market instead of being confused by it.
Two Jobs, One Word
The execution algorithm is the older and quieter of the two. Its input is a parent order, perhaps 50,000 shares to buy, and a benchmark to beat: the volume-weighted average price of the session, the arrival price, or the close. Its output is a stream of child orders, sized and timed to blend into the market's own rhythm. It has no opinion. If the parent order says buy, it buys; its entire intelligence is spent on how. Buy a little faster when the tape absorbs it easily, pause when the book thins, drift with the session's volume curve so the order leaves no wake. The execution algo is measured in basis points of slippage, and a good one saves its user more than any single trade decision can.
The alpha algorithm is the decision maker. Its input is market data, news, statistical relationships, or any signal its designers can justify; its output is a position. Some run for days on end and trade a handful of times a week. Some hold for milliseconds. What unites them is the standard they are held to: a tested, quantified edge. An alpha program that loses money gets rewritten or retired, and the ones that survive are the ones whose rules kept working after the costs were counted. From the outside, on a chart, the alpha programs are the visible aggression: the sudden lifts, the fast sweeps, the moves that start without warning because a model decided the price was wrong a fraction of a second before any human could have.
| Attribute | Execution algorithm | Alpha algorithm |
|---|---|---|
| Job | Deliver a decided order into the market | Decide what to buy or sell |
| Input | A parent order and a benchmark | Market data and a tested model |
| Output | Hundreds of small child orders | Positions, entries, exits |
| Timescale | Minutes to days | Milliseconds to weeks |
| Scored on | Slippage against the benchmark | Profit after costs, risk-adjusted |
What Dominance Looks Like
Machine participation stopped being a niche decades ago, and its share is now the majority of volume in major equity and futures markets. On the chart, that dominance shows up less as chaos than as texture. Average print sizes shrink, because execution algos deliberately slice. Moves retrace faster, because decision programs stand ready at every price and absorb hesitation. The book at any single moment is thinner than it looks in screenshots, because displayed size can be canceled in microseconds. Days do not look louder; they look quicker. The auction still does its job, discovery still happens, but the unit of participation has changed from a person weighing a decision to a program weighing thousands of them per second, and the residue of that change is visible in almost every session's tape.
Dominance also changed what kind of behavior pays. Human hesitation, partial fills, and round-number spacing all leave patterns in the flow, and programs are built to work around exactly those habits. Volume now clusters at scheduled marks: the open, the close, scheduled rebalances, option expiries, because execution algos are benchmarked to time and their orders arrive on schedules. None of this requires a conspiracy to explain. It is the sum of thousands of programs following similar published ideas about how to slice an order, all running at once, all slightly amplifying the same rhythms.
A Worked Example: One Basket, One Afternoon
An asset manager decides at midday to add 50,000 shares of a stock trading at 80.00. The decision took weeks of research and one committee meeting; the delivery takes the rest of the afternoon, and machines do all of it. The parent order goes to an execution algorithm with one instruction: buy 50,000 shares, keep the slippage under the volume-weighted average price, finish by the close. The algo estimates the afternoon's volume curve, then starts slicing. Every twenty seconds it sends another child order of 100 shares. Five hundred child orders over two hours and forty minutes, each one small enough to sit inside the bid without moving it, each one timed to the volume rhythm so the participation rate stays constant. An observer watching the tape sees no order at all. They see a steady pulse of 100-lot prints on the bid, indistinguishable from a hundred other steady pulses, which is precisely the design.
Halfway through the program, the alpha side of the market notices something. The stock has been drifting down all afternoon, and to one decision program the drift has carried price below a level its model prices as fair. The alpha algo buys, fast: not 100 shares but 2,000, lifted off the offer in under a second, walking the price from 79.80 to 79.86. The tape prints a block of blue candles and every short-term trader sees a move. The two algorithms now meet without ever seeing each other: the execution algo, still pinned to its schedule, happily sells nothing and keeps bidding, and pays the 79.86 for its next slices as the drift resumes. By the close, the 50,000 shares are fully delivered at an average of 79.91, three cents inside the day's volume-weighted average price. The alpha program is showing a profit. The asset manager paid for the delivery like a utility bill. Neither program held an opinion about the other, and the chart records only their combined footwork: the pulse, the spike, the drift.
That afternoon is the market in miniature. Decisions arrive at random, deliveries arrive on schedule, and the tape is the sum. A trader who reads the session as a story about buyers and sellers in agreement is reading the wrong layer; the layer beneath has an institution patiently paying whatever the afternoon charges, and a model briefly correcting a price it disagreed with.
Where the Human Still Fits
The honest answer to what machines have taken is process, not edge. Programs execute better than people, and they decide faster, but the rules they run are written by people testing ideas against history, and the edges they exploit were found by people asking why a pattern should pay. What has genuinely closed is the middle ground: the discretionary trader who wins on quick fingers and full attention at the screen is competing directly with programs built for exactly that contest. The ground that remains is slower and less crowded: reading context that no model was trained to price, holding a position through discomfort a program would never feel, and supplying the patience that delivery algorithms borrow from their clients. The retail chart reader is not going to out-react a machine. They can still out-wait one, and the market still pays for waiting in the right place.
Machines dominate the tape by every measure of volume, and the dominance is structural: the delivery job exists because the largest participants need it, and the decision job exists because it pays. Neither is going away. The practical response is not nostalgia for a human tape but literacy in the machine tape: knowing which prints are delivery, which are decisions, and why the clock now matters as much as the price.
The fastest corner of the machine market deserves its own treatment. The next lesson narrows to high frequency trading: the speed race, the rebate business, and the measured impact on everyone else's fills.
Algorithmic Trading Questions
Is all trading now algorithmic?
Not all, but the majority of volume in major markets is, counting both execution and decision programs. Human-directed trading remains substantial in options markets, small caps, and discretionary funds, but even most manually decided orders are delivered to the market by execution algorithms. The machine share covers both the deciding and the delivering, which is why it dominates volume while individual people still matter at the margin.
What is the real difference between an execution algo and a trading algo?
Authority. The execution algorithm receives a decision that was already made and optimizes only the delivery: size of slices, timing, and cost. The alpha algorithm makes the decision itself, from data, under a tested edge. The execution algo is scored on slippage; the alpha algo is scored on profit. One is logistics, the other is judgment encoded in rules.
Do algorithms cause market volatility?
They change its texture more than its level. Machines absorb hesitation and retrace moves quickly, which smooths ordinary sessions, but they also move at identical speed when a shared signal fires, which concentrates volatility into shorter bursts. The evidence cuts both ways: calmer average days, faster extremes. The programs are amplifiers of whatever behavior their rules encode, not an independent source of panic.
How does a retail trader compete against algorithms?
By not competing where machines are strong. Out-reacting a program in milliseconds is impossible; out-waiting one is routine, because programs hold what their models price and no more. The durable edges left for the individual are context, patience, and timeframes longer than the machine attention span: holding through the discomfort a program cannot feel, and reading conditions no model was trained to price.