What it is
Algorithmic trading means deciding and executing trades through a defined set of rules rather than discretion. The rules can be simple or elaborate; what makes it algorithmic is that they are specified in advance and applied consistently.
The appeal is measurability. A discretionary trader cannot easily say why last month was worse than the one before. A systematic one can test the rules against history and know what to expect.
The main strategy families
- Trend following — enter in the direction of an established move, exit when it fails. Low win rate, large winners.
- Mean reversion — trade the expectation that price returns to an average after stretching away. High win rate, occasional large losses.
- Breakout — trade a move out of a established range or level.
- Momentum — trade instruments already moving strongly, expecting continuation.
- Market making — quote both sides and earn the spread. Requires infrastructure most retail traders do not have.
Backtesting, and how it lies
A backtest replays a strategy over historical data. It is essential and it is also the easiest thing in trading to get wrong.
The classic failures: not charging spread and commission; using data that would not have been available at that moment; testing on a period so short it captures one market condition; and — most common — tuning parameters until results look excellent, which fits the strategy to that specific history rather than to the market.
Walk-forward validation
The honest approach is to fit parameters on one window of data and evaluate them on the *next* window, which the fitting never saw. Repeat across the history and you get a picture of how the strategy performs on genuinely unseen data.
Results usually degrade substantially at this step. That degradation is the useful information — it is the difference between a strategy and a description of the past.
The numbers that actually matter
- Profit factor — gross profit divided by gross loss. Above 1.0 is profitable; below 1.3 is thin.
- Maximum drawdown — the worst peak-to-trough fall. What you have to survive emotionally and financially.
- Trade count — under a hundred trades, results are largely noise.
- Out-of-sample profit factor — the same measure on data never used for fitting. The only one worth much.
- Expectancy per trade — average profit per trade after costs.
Common questions
Is algorithmic trading profitable?
It can be, and most attempts are not. The common failure is a strategy fitted to historical data that has no edge on new data. Out-of-sample validation is what separates the two, and it is the step most often skipped.
Do I need to know how to code?
To build your own, yes — or you commission it. Some platforms offer visual strategy builders, though they constrain what you can express. Either way you need rules precise enough to leave no ambiguity.
How much data do I need to backtest?
Enough to include different market conditions — trending, ranging, volatile and quiet. Several years on higher timeframes, and enough trades that the result is not noise. A hundred trades is a reasonable floor.
Trading carries risk and you can lose money. Nothing on this page is financial advice. Past performance does not indicate future results.