An informal measure of how much a market's short-term price wiggle (noise) overwhelms its longer-term direction (trend). Trend-following systems work when this ratio is low; they fail when it is high.
Noise-to-Trend Ratio
An informal measure of how much a market's short-term price wiggle (noise) overwhelms its longer-term direction (trend). Trend-following systems work when this ratio is low; they fail when it is high.
In plain English
Imagine two stocks that both gain 50% over a year:
- Stock A drifts up smoothly with small daily moves.
- Stock B gains 50% net, but along the way it crashes 30% twice and rallies 40% three times.
Both have the same trend. But Stock B has much more noise — short-term price action that isn't part of the underlying move. A trend-following signal looking at Stock B will see "fake reversals" inside the larger uptrend and trigger losing trades.
That ratio — how much noise crowds out the trend — is what separates BTC from ETH from SOL in this fleet.
Why it matters for this fleet
[[EMA Cross]] is essentially a noise filter: the slow EMA smooths out the noise so a crossover means something. If the noise is too big, the smoothing doesn't help — crossovers fire on noise, not on trend.
- BTC has the lowest noise-to-trend ratio of the three. Its drawdowns inside an uptrend tend to be 10–20% — uncomfortable but recoverable, and the 4h EMA cross can ride through them.
- ETH historically trades with higher beta to BTC plus its own idiosyncratic moves — more noise per unit of trend.
- SOL has the highest noise-to-trend ratio of the three. Its intraday and intraweek swings frequently exceed 15%, even during a bull market.
But the sharpest noise lever in this dossier is not the symbol — it is the pair-and-interval speed. A fast EMA pair on a fast candle samples more of the wiggle and less of the trend; a slow pair on a slow candle does the opposite. That speed axis dominates everything else.
How to estimate it (rough heuristics)
- Average True Range (ATR) / price ratio over a period: higher = noisier.
- Number of EMA crosses per month on a fixed period: more crosses ≈ more noise.
- Realized volatility / trend slope: classic signal-to-noise measure.
(The simulator doesn't have these as first-class metrics yet, but they can be computed from the candles table.)
Examples from the live fleet
The pair-speed axis sorts the fleet almost perfectly by noise:
- The 9/21 scalp pair (the fastest pair in the dossier) is 0% profitable at 2× leverage — not one row clears break-even — with a median profit factor of 0.49 (gross wins ÷ gross losses; below 1.0 means it loses money). The extreme case is id628 (EMA 9/21 · BTC · 1-minute candles · 2× · short): win rate 10.6% (the share of trades closing in profit), profit factor ~0.11, and a −98% drawdown. At one-minute resolution there is essentially no trend left to follow — the crossover is firing on pure noise, and it blew up.
- The 50/200 macro pair (the slowest) sits at the other end: it carries the highest raw profit factor in the fleet, 1.68. Fewer crossings, each one filtering out far more of the wiggle, so the handful of signals it does fire are cleaner. (The catch, covered in sample size: "fewer signals" also means too few trades to prove an edge — none of the 50/200 rows are edge-significant.)
The lesson is the ratio itself: speed up the pair or the candle and you raise the noise-to-trend ratio, and the signal degrades from a coherent (if rare) trend filter into a noise generator.
Related
Sources
wiki/qa-sessions/2026-05-17-session.md#q1 (first asked here)/api/analytics perSymbol data