Your first backtest

Every strategy, from a hobbyist's to a hedge fund's, is the same four parts: a signal, an entry, an exit, and a size. We'll snap them together into the classic beginner strategy — a moving-average crossover — and watch it run.

Every strategy is four parts

Strip away the jargon and a trading strategy answers four questions, in order:

Miss any one and you don't have a strategy yet. Beginners obsess over the signal and forget the other three — but as you'll see, exit and size decide your fate at least as much as the entry does.

A worked example: the moving-average crossover

The "hello world" of backtests. A moving average (MA) is just the average price over the last N days, recomputed each day — it smooths the noise so you can see the trend. We use two: a fast one (say 10 days) that reacts quickly, and a slow one (say 50 days) that moves ponderously.

• Signal: the fast MA is above the slow MA → the recent trend is up.
• Entry: buy the moment the fast MA crosses above the slow MA (the "golden cross").
• Exit: sell the moment it crosses back below (the "death cross").
• Size: risk a fixed slice of the account on the position.

Run that rule over years of history, tally the trades, and you have a backtest. Feed the trade results into Module 2's scoreboard — CAGR, Sharpe, drawdown — and you can finally answer "would this have worked?"

Size is not an afterthought

Two traders run the identical crossover signal. One risks 2% of the account per trade; the other risks 30%. Same entries, same exits — wildly different fates. The reckless one can be wiped out by a normal losing streak the careful one shrugs off. Position sizing turns a signal into a survivable strategy, and it's the difference between a backtest that compounds and one that goes to zero on trade 14.

This is its own discipline, and we built a calculator for it: the Position Size & Risk of Ruin tool runs a Monte Carlo of your edge to show how sizing changes your odds of survival. Module 6 goes deep on it.

Vectorized vs. event-driven

Two ways to actually compute a backtest, and it's worth knowing the words:

• Vectorized: calculate all the signals across the whole history at once, in one sweep. Fast, simple, and what nearly every beginner tool (and most of our teardowns) uses. The risk: it's easy to accidentally let future data leak in (hello, look-ahead bias).
• Event-driven: step through time one bar at a time, as if live, only ever seeing the past. Slower and more code, but far more realistic — it's how serious systems are validated before real money.