A trading system is an idea that has been forced to become specific: rules complete enough that two people following them would take the same trades. Building one is less about finding a magic combination and more about honest bookkeeping on a simple idea.
Start from an edge hypothesis, not from indicators
Every durable system is a sentence before it is a ruleset: "after a strong impulse, the first pullback tends to continue", "ranges compress before they break", "currencies with hawkish repricing trend for days". The sentence names WHO is on the losing side and why the behaviour should persist. Systems assembled backwards — stacking indicators until the backtest looks good — skip this step, and it shows later: when the rules stop working, there is no underlying idea to check against, only settings to twiddle.
The five components
- Filter (when you are allowed to hunt): the market condition your edge needs — trending by structure, above the 200-period average, volatility above a floor. The filter's job is keeping you out of the environments where the idea has no prey.
- Setup (what must be true): the checkable conditions — pullback to the prior breakout level, RSI reset, session window.
- Trigger (the exact go): one unambiguous event — the close beyond the level, the break of the trigger bar. If two readers could disagree, it is not a trigger yet.
- Exit, both sides: the invalidation stop and the profit mechanism (fixed R multiple, structure trail, time stop). Exits define the P&L distribution; systems differ here more than at entries.
- Sizing: fixed-fractional off the stop distance. Already covered in Module 10 — the system inherits it, never improvises it.
Expectancy: the only verdict
A system's quality is one number: expectancy = (win% × average win) − (loss% × average loss) — the expected profit per trade, best expressed in R (multiples of what you risked). A 40%-win-rate system whose winners average 2.5R and losers 1R earns (0.40 × 2.5) − (0.60 × 1) = +0.40R per trade: excellent. A 70% win rate with 0.5R winners and 1R losers is (0.70 × 0.5) − (0.30 × 1) = +0.05R: barely alive, and dead after costs. Win rate alone tells you nothing — it must be read against the payoff ratio, and beginners chasing high win rates routinely build systems on the losing side of that trade-off because winning often feels like an edge. Spread and swap costs come out of expectancy directly; a +0.1R edge on a 10-pip target does not survive a 1.5-pip spread.
The overfitting trap
Take any losing ruleset, add conditions until the backtest is green, and you have not found an edge — you have memorized history. Signs you are doing it: parameters tuned to odd precision (why 13.5 and not 14?), rules that exist to dodge specific past losses, performance that collapses when one setting moves slightly. Defenses: keep the rule count small (five conditions is a lot); demand the idea works roughly across a band of settings, not exactly at one; test on data you did not build on (build on 2023–24, verify on 2025); and distrust any equity curve that looks too smooth — real edges are lumpy, because the market only pays when the pattern's losers are taking the other side.
From rules to trust
Once the rules are written, the sequence is fixed: backtest for plausibility, then trade it small — 0.1–0.25% risk — for 30–50 live trades, executing EXACTLY as written. The live sample teaches what no backtest can: slippage reality, and whether you can actually follow the rules when money is attached. Grade yourself on rule-adherence first, expectancy second; a profitable month of broken rules is worse news than a losing month of perfect execution, because only one of them is repeatable. Scale risk only after both numbers pass. That is the whole craft: a falsifiable idea, honest arithmetic, and a sample size before conviction.