"Buy when it looks oversold" is not a strategy. A machine can't run it — there is a judgment call hiding in every word: looks how, oversold by what measure, buy how much? A quantitative strategy removes the judgment by breaking the idea into a handful of explicit parts, each a rule with no room for interpretation.
However complex a strategy looks, it is the same six parts. To make that concrete, we'll take one real strategy apart — Adaptive Trend, run on Bitcoin over 2016→2026 in the backtest lab.
1. Hypothesis — an economic reason, not a chart pattern
Before a single line of code, one question: why should this work? Adaptive Trend's answer is that markets tend to keep moving in the same direction for longer than expected — information is priced in gradually, and flows chase performance. That is an economic claim you can argue about, not a shape someone spotted on a chart.
This is the part most people skip, and skipping it is why most "strategies" are really just curve-fits. If you can't state the why in one sentence, you don't have a hypothesis — you have a coincidence that held up in the past.
2. Signal — the hypothesis, made numeric
The hypothesis becomes a rule the machine applies to every candle, without exception. Here the signal is a crossover of the 50-day and 200-day exponential moving averages: when the fast average crosses the slow one, the trend counts as confirmed. "The trend is up" (open to debate) becomes EMA(50) > EMA(200) (a fact). No candle is skipped, no exception is made because this time feels different.
3. State — long, short, or cash, and nothing in between
At any moment the strategy is in exactly one of three states: long, short, or cash. Over ten years that produced 134 long trades and 61 short trades, with the strategy exposed to the market 84% of the time. Constraining the states is the point — it removes the half-positions and the "let me wait one more day" that discretion invents. A rule that can only pick from three answers can't talk itself into a fourth.
4. Volatility sizing — hold the risk constant, not the position
How much to hold is its own rule. Adaptive Trend estimates volatility with an EWMA, then scales the position so that annualized portfolio volatility stays near a fixed 25% target, capped at 5% of capital per trade. The effect is subtle but decisive: risk stays roughly constant instead of ballooning exactly when markets turn violent.
A fixed number of coins is a moving amount of risk; a fixed amount of risk is a moving number of coins. Sizing by volatility chooses the second — the same reason a good driver eases off the accelerator on ice.
5. Exit — a rule that breathes with volatility
The exit is a trailing stop at 2.5× the 20-day Average True Range. Because ATR expands in turbulent markets and contracts in calm ones, the stop automatically widens to give a real trend room to run, and tightens to lock in gains as things settle — without ever removing the protection. The exit adapts to the weather instead of being a fixed distance that is too tight in a storm and too loose in the calm.
6. Cost and evaluation — subtract reality first
A 0.1% fee per execution reads like a rounding error. Across 195 trades it came to $21,594 — more than double the starting capital. This is the part beginners leave out and the part that quietly kills live strategies: no backtest number means anything until costs are deducted. A result measured before fees isn't optimistic — it's fiction.
What it actually did
Put those six rules together, run them on ten years of Bitcoin, and this is the result — the equity curve and the drawdowns underneath it:
The headline numbers:
| Metric | Value |
|---|---|
| Sharpe | 1.47 |
| CAGR | 38.10% |
| Max drawdown | 28.99% |
| Profit factor | 1.95 |
| Win rate | 44% |
The biggest number is not the point
Over the same decade, simply buying and holding Bitcoin returned far more — 14,638% versus the strategy's 2,926%. So did the strategy fail? Only if "most return" is the goal, and it rarely is. Buy-and-hold earned that number through gut-wrenching volatility and drawdowns deep enough to shake most people out at exactly the wrong moment. What a quantitative strategy sells is not the biggest number — it is the best return per unit of risk.
That gap — Sharpe 1.47 versus 1.05 — is the entire thesis. And notice something that looks like a contradiction: the strategy loses on 56% of its trades and still wins. It can, because the average win (18.33%) is nearly three times the average loss (6.71%). Win rate is one of the most misleading numbers in trading; how much you make when you're right versus how much you lose when you're wrong matters far more than how often.
Why every part has to be explicit
Read the six parts again and notice what is missing: nowhere is there a "use your judgment." That absence is deliberate, and it is what makes the strategy testable. The moment any part contains a human decision — "exit when it feels toppy" — you can no longer backtest it honestly, because you will unconsciously make that call with the benefit of hindsight. Explicit rules are the only thing that let history give you an answer you didn't rig.
Three traps that void the result
Even a fully specified strategy can be quietly broken:
- Lookahead — using information the rule wouldn't have had in real time (tomorrow's close, a later-revised data point). The backtest looks brilliant and falls apart live.
- A signal you can't compute in real time — if the condition needs data that only exists after the fact, the strategy can't be traded, only admired.
- Sizing that ignores volatility — a fixed position size lets risk swing wildly; the same trade that is prudent in a calm market is reckless in a panic.
The checklist
Before you write a line of code, pressure-test the idea against the six parts:
- Hypothesis — can you state the economic why in one sentence?
- Signal — is the condition numeric and computable on every bar?
- State — is every possible position defined, with nothing in between?
- Sizing — does the size hold risk constant, not exposure?
- Exit — is there an explicit rule for getting out, written before you need it?
- Cost — have you subtracted realistic fees before judging anything?
If any answer is "it depends," that is where the discretion — and the self-deception — is hiding. Fix it there, before it costs you.
The six parts turn an idea into something testable. The next questions are how you judge the result honestly — which return you should even look at, what each risk-adjusted metric quietly hides, and the ways a backtest can still lie to you. Those are the pieces ahead.