What Is Algorithmic Trading? A Plain-English Explanation

Strip away the marketing and algorithmic trading is one plain idea: running a fixed set of rules that turns market conditions into orders. Almost everything people believe about it describes something else.

Ask ten people what algorithmic trading is and you get ten pictures. A room of physicists racing each other in microseconds. A neural network that has learned the secret shape of the market. A black box you switch on and leave alone while it pays your bills. Those are, in order, a different business, a different claim, and a fantasy. None is the definition.

What follows is the definition without code and without hype, then most of the length spent on what algorithmic trading is not, because the misconceptions do more damage than the ignorance.

Start with the boring definition

An algorithm is a fixed set of rules that turns conditions into actions. A recipe is an algorithm. So is the instruction on a fire door. There is nothing computational about the word itself.

A trading algorithm is a fixed set of rules that turns market conditions into orders. That is the whole definition. Written out in English rather than code, a very simple one might read:

If the price of this instrument closes above its 50-day average, buy an amount sized to risk 1 percent of the account, place a stop below the recent low, and exit when either the stop is hit or the price closes back below the average.

That is a complete algorithm. It is deliberately trivial, and it is not a good one, but it has every part a real one has: a condition, a decision, a size, and an exit. Algorithmic trading is simply the practice of running rules like that: writing them down precisely enough that software can execute them without a human deciding anything in the moment.

Notice what the definition does not contain. It says nothing about speed, nothing about learning, and nothing about profit. Those are three separate claims that get bundled into the word, and they need unbundling one at a time.

It is not high-frequency trading

High-frequency trading, usually shortened to HFT, is a subset of algorithmic trading. Treating the two as synonyms is like treating Formula One as a synonym for driving.

HFT is defined by its speed and its infrastructure. Firms in that business rent rack space inside an exchange’s own data centre so their servers sit metres from the matching engine, a practice called colocation. They buy custom hardware and direct market data feeds to shave microseconds off the round trip between seeing a price and acting on it. Positions are opened and closed within seconds or fractions of a second, and the day usually ends with little or nothing held.

That is a business built on latency, capital, and engineering headcount, and no retail platform gives access to it. A retail algorithm has none of those properties. It runs on a laptop, a small cloud server, or a broker’s platform, reaching the market over an ordinary internet connection. Its decision points are minutes, hours, or days apart. Its edge, if it has one, cannot be speed, because it will always lose a speed race to a colocated machine. It has to come from the rules themselves: what the strategy chooses to trade, how much it risks, and when it stands aside.

So when someone describes retail automation in the language of HFT, ask which one they actually mean. The two share a family name and nothing else that matters.

It is not artificial intelligence

The second bundle to unpick is the assumption that automated means intelligent. The algorithm sketched above contains no learning of any kind. It does exactly what it was told, forever, until a person changes it. That is true of the overwhelming majority of retail trading automation: the rules were written by a human, tested by a human, and are executed literally by software that understands nothing.

Machine learning in trading is a real field, and some institutional systems genuinely do adapt their parameters from data. But it is a separate claim, and in retail products it is usually a marketing word rather than a description of the software. There is no way to tell from the outside, which is exactly why the word gets used.

So treat it as a question rather than a selling point. If a product claims to be AI-driven: what does it learn from, how often does it change its own behaviour, and what does the record look like across conditions the system had never seen. If the answers are vague, the claim is decoration. A strategy that plainly says “these are the rules and they do not change” is telling you more, not less.

It is not a machine that prints money

The third misconception is the expensive one. An algorithm does not know whether it is winning. It has no opinion about your balance and no ability to notice that the market has changed character. It applies its rules to whatever is in front of it, including conditions its author never considered.

Automation removes hesitation, fatigue, and the temptation to widen a stop because this one feels different. Those are real advantages, covered separately in how automation removes the human element. What automation does not remove is loss. A rule-following system takes losing trades, has losing weeks, and goes through drawdowns, because the rules are a bet about how markets behave and no bet wins every time. Consistency of execution is not the same as certainty of outcome.

The four parts of any trading algorithm

Every trading algorithm, however it is dressed, is made of the same four components. Naming them is the fastest way to evaluate one, because you can ask about each part separately.

• Signal. The condition that triggers interest: a price level, a moving average cross, a volatility reading, a time of day. This is the part people obsess over, and usually the least important of the four.
• Sizing. How much is put at risk per trade. Sizing determines what a losing run costs, which makes it the component most likely to decide whether an account survives. Our note on position sizing and the 1 percent rule covers the arithmetic.
• Execution. How the order reaches the market: order type, acceptable slippage, what happens if the price moves between signal and fill, or if the connection drops. Unglamorous, and the source of most of the gap between a backtest and a live account.
• Risk control. The rules sitting above the strategy: maximum exposure, daily loss limits, behaviour around scheduled news, when the system stops trading entirely. Quietly missing from most systems that fail badly.

A description of a strategy that only talks about signal is a description of one quarter of one.

How a retail trader actually gets access

There are three routes, and they demand very different things from you.

Build. Write the rules yourself, usually as an Expert Advisor on MetaTrader 5 or in Python against a broker’s interface. Full control, full responsibility. It requires programming ability and, harder than the coding, the discipline to test honestly.

Buy. Purchase someone else’s automated system and run it on your own account. The barrier is low, and so is the visibility: you are trusting rules you cannot read, usually supported by a record you cannot independently confirm.

Follow. Connect your account so that trades executed on a strategy account are mirrored into yours. You keep your money at your own broker and run no software yourself. What you give up is control over the rules, which makes the quality of the evidence the whole decision: an independently tracked record on a third-party service such as Myfxbook, covering a period long enough to contain bad conditions as well as good ones. Our guide on how to vet a strategy sets out what to check before any money is involved.

All three routes are legitimate. None removes the need to understand what you are attached to.

An algorithm inherits every flaw in its rules

This is the sentence worth keeping. Software does not improve a strategy. It removes the human variability from the execution of one, which makes a good set of rules more reliable and a bad set of rules more reliably bad.

If the rules were fitted to a period that happened to suit them, automation will apply that fit faithfully to a period that does not. If the sizing is too aggressive, automation will apply it without flinching through a losing run, exactly when a person would have hesitated. If the risk controls were never written, nothing stops the system at the point where stopping matters. The machine is not a second opinion. It is an amplifier for the judgment already inside it.

Which is why the useful question is never “is it automated”. It is “what are the rules, who wrote them, how are they sized, what stops them, and what does the verified record show”.

Key takeaways

• A trading algorithm is a fixed set of rules that turns market conditions into orders, and algorithmic trading is the practice of running them. That is the entire definition.
• High-frequency trading is a different business, defined by colocation and microsecond decisions. Retail automation has no access to it and cannot compete on speed.
• Automated does not mean intelligent. An AI claim is a separate assertion, to be checked rather than assumed.
• Every algorithm has four parts: signal, sizing, execution, and risk control. The last two decide survival more often than the first decides profit.
• Retail access comes in three forms: build, buy, or follow. Each still requires you to understand what you are attached to.
• An algorithm inherits every flaw in the rules it was given. Consistent execution is not the same as a good outcome.

This article is educational and is not financial advice, a recommendation, or an offer. Trading foreign exchange and CFDs carries a substantial risk of loss, including the loss of your entire investment, and losses are a common outcome. Any figures used above, such as the 1 percent risk in the sample rule, are illustrative only and are used to explain a mechanism. They are not results, projections, or typical outcomes of any strategy. Past performance is not a guarantee of future results. Consider your own circumstances and seek independent professional advice where appropriate.