Selectivity

A property of a strategy's rules: how often the entry conditions fire. A selective (picky) strategy generates few entries; a permissive one generates many. Selectivity is unrelated to per-trade confidence.

In plain English

When a backtest report shows few trades, it is tempting to interpret that as "the strategy was careful, only entering high-quality setups." This is wrong.

A strategy has no concept of confidence. It has rules — "fast EMA crosses slow EMA AND ADX > 25" — that either match the current candle or do not. Every entry is treated identically by the engine. The strategy does not know which trades are better than others.

What looks like selectivity is really just the strategy's rules happening to fire rarely, driven by:

• Interval. 1d candles produce ~365 opportunities per year; 1h produces ~8760.
• Filter strictness. Each added condition (ADX, RSI threshold, volume) shrinks the set of candles that qualify.
• Symbol behavior. Symbols with cleaner trends fire EMA-cross signals less often than choppy ones.

A selective strategy can have real per-trade edge — but you cannot tell from the data alone whether the few observed trades are skill or luck. See sample size and statistical significance.

Why it matters for this fleet

Trade counts in this dossier swing wildly with interval and EMA pair, purely structurally — not because some strategies are "choosier." At the two extremes: a 50/200 pair on daily candles produced just N=3 trades (id478), while a 9/21 scalp on 1-minute candles produced N=10,574 (id628). The 50/200-on-daily pair averages only ≈28 trades across the fleet — structurally few, because daily candles cross rarely.

No strategy on the 1d interval is "more confident" than a 1h strategy — it just has far fewer candles to look at. And selectivity buys nothing on its own: that 9/21 1-minute scalp fires the most of any pair, yet the 9/21 pair is 0% profitable fleet-wide.

Selectivity vs. confidence — a useful re-framing

The user's instinct ("few trades = high confidence") collapses these four into one. Keeping them separate is the foundation of clean strategy evaluation.

Examples from the live fleet

• id478 (EMA 50/200 · BTC · 1d · 2× · long): just N=3 trades — looks extremely selective. But the rules are simply a 50/200 EMA cross on daily candles, with no "confidence filter" anywhere. The rarity is structural: daily candles cross the 50/200 pair very seldom over the window. Three trades is too few to conclude anything (win rate could be anywhere from 20% to 94%).
• id511 (EMA 21/50 · BTC · 1h · 2× · long): N=469 trades — looks permissive. Same kind of EMA-cross logic, just a tighter pair on 1h candles, so it has far more crossings to react to. It is not less selective in any meaningful sense; it simply operates on shorter candles.

The right comparison is not "selective vs. permissive." It is "high-N reliable vs. low-N speculative." And note the worst offender: the 9/21 scalp fires the most of any pair, yet is 0% profitable — proof that frequency tells you nothing about per-trade quality, in either direction.

Practical implications

• Don't reward strategies for having few trades. Reward them for high N AND positive metrics.
• If you want true selectivity, you have to design it explicitly — e.g. add a regime filter ("only trade when ADX > 30") and measure whether the filtered subset performs better than the unfiltered one (sub-population edge, not absolute edge).
• A strategy that fires only a handful of times over the window (like id478's 3 trades) isn't selecting trades — the market is selecting it.

Related

• statistical significance
• sample size
• edge
• ema cross
• market regime

Sources

• wiki/qa-sessions/2026-05-17-session.md#q2 (first asked here)
• Fleet trade counts by interval