What is a trading algorithm? Rules, signals and code explained

A precise definition

A trading algorithm is a deterministic set of rules that maps market data to an action. Given the same inputs, it always produces the same output — that reproducibility is the whole point. It removes emotion, hesitation and inconsistency, and it makes the strategy testable: you can run it over years of history on the backtester and measure exactly how it would have performed. A hunch cannot be backtested; an algorithm can.

The four parts of every algorithm

Strip any trading algorithm down and you find the same four components. An entry rule decides when to open a position. An exit rule decides when to close it — usually a stop-loss and take-profit. A position-sizing rule decides how much to risk, covered in position sizing. And a universe defines which markets it watches. Miss any one and you do not have a complete, runnable algorithm.

The anatomy of a signal

A signal is the output of the entry/exit logic: a value that says long, short or flat. It is built from indicators computed on price — a moving average, an RSI, a breakout level. The key discipline is that a signal must use only data available at decision time; peeking at the close of a candle before it forms is the classic look-ahead bias that makes a worthless algorithm look brilliant in a backtest.

A real, readable example

python · sma_algo.pyimport pandas as pd

def signal(df, fast=20, slow=50):
    """Return 1 (long) or 0 (flat) for each bar — a complete entry algorithm."""
    df['fast'] = df['close'].rolling(fast).mean()
    df['slow'] = df['close'].rolling(slow).mean()
    # shift(1): only act on the PREVIOUS closed bar — no look-ahead
    df['pos'] = (df['fast'] > df['slow']).shift(1).fillna(0).astype(int)
    return df

That is a full trading algorithm in six lines: an entry/exit rule expressed as a position of 1 or 0 per bar. Bolt on a sizing rule and a venue and it becomes a runnable bot.

Algorithm vs bot vs strategy

People use these words loosely, so pin them down. A strategy is the idea (“follow trends”). An algorithm is that idea written as exact, testable rules. A bot is the running program that executes the algorithm live against an exchange API. You design a strategy, encode it as an algorithm, and a bot runs it.

Where to start

Begin with one simple, transparent algorithm — an SMA crossover is the canonical first one — and test it honestly on the backtester before adding any complexity. Resist the urge to stack indicators; every extra rule is another knob to overfit. A simple algorithm you understand beats a complex one you do not.

Frequently asked questions