DEV Community

Gueta Quant
Gueta Quant

Posted on Originally published at guetaquant.com AI-assisted

ATR Stops Without Magic Numbers: Why 2.5 ATR14 Is a Convention, Not a Law

Stop = Price ± (k × ATR14) fits in one line. The problem was never the formula — it is
choosing k, and knowing what the number does and does not mean.

Educational content: methodology only, no signals, no return claims.

By **Mahdi Goodarzi* (g.dev/mahdigoodarzi), Gueta Quant —
Spanish guide: gestión de riesgo con ATR.*


1. ATR is a range measure, not a probability

This distinction gets lost in most tutorials. ATR measures the average absolute range of
price. It is not a standard deviation (σ) and it assumes no normal distribution.

That matters because financial return distributions have heavy tails. So a 2.5×ATR stop is
a tolerance distance around the average range. It is not a 95% confidence interval, and
nobody can derive a loss probability from it without additional distributional assumptions
that are themselves questionable.

2. The k trade-off is real and unavoidable

Context Common k Consequence
Scalping ~1.5× ATR More stops hit, smaller loss per stop
Swing ~2.0–2.5× ATR Middle ground
Positional trend ~3.0–3.5× ATR Fewer stops, larger loss when one triggers

Low k means paying the noise cost: you get stopped by ordinary volatility. High k
means paying the severity cost: each stop that does trigger takes a bigger bite, so you
need a higher reward-to-risk ratio just to keep expectancy positive. Above roughly 3.5×
ATR, that requirement becomes hard to satisfy honestly.

2.5×ATR14 is the most common convention in intraday and swing models. It is a
convention, not an optimum, and it should be calibrated per instrument by examining the
distribution of your own stop-out distances.

3. The rule that does generalise: inverse sizing

Position size should be inversely proportional to volatility. When ATR doubles, size
halves; when ATR compresses, size can grow — holding the monetary risk constant.

def position_size(capital, risk_pct, atr_value, value_per_unit, k=2.5):
    """Monetary risk is fixed; distance is volatility-derived."""
    risk_money = capital * risk_pct
    stop_distance = k * atr_value
    if stop_distance <= 0 or value_per_unit <= 0:
        raise ValueError("distance and unit value must be positive")
    raw = risk_money / (stop_distance * value_per_unit)
    return raw          # caller applies broker min/step, and must re-check margin
Enter fullscreen mode Exit fullscreen mode

Two things that script deliberately does not do: it does not round up to the broker's
minimum volume (that silently multiplies risk on small accounts), and it does not verify
margin. Those are separate checks.

4. Static vs trailing

  • Anchored to entry: computed once at open, never moves. Simple, and it lets you plan size in advance.
  • Trailing: recomputed as price moves. Reduces give-back, but introduces a rule you must define before the trade — otherwise you are moving stops after seeing the outcome, which is outcome conditioning, not risk management.

5. What to check before trusting any k

  1. Does the instrument's volatility regime match the sample you calibrated on?
  2. Are stop-outs dominated by noise rather than by thesis invalidation?
  3. Does the reward-to-risk still clear after your real spread and commission?
  4. Does the resulting size pass a margin check separately?

If any answer is unclear, the stop distance is a guess with a formula attached.


Study material under Colombia SFC Decreto 2555 de 2010 — financial-engineering education,
not investment advice. No signals, no managed accounts. Related:
MT5 lot sizing ·
verify stop loss and margin separately in MT5 ·
risk management foundations.

Top comments (0)