ATR (average true range)

Definition

ATR (Wilder 1978) is the smoothed average of the true range — each bar’s high-low span extended to include any gap from the prior close. It measures how much an instrument MOVES per bar in its own price units, gaps included. ATR is this KB’s workhorse scale unit: stops, targets, ladder spacing, and position sizes are quoted in ATR multiples so that one parameterization transfers across instruments.

How it works / structure

• True range: TR_t = max(H−L, |H−C_prev|, |L−C_prev|) — the three-way max is what captures gaps that high-low alone misses.
• Formula: ATR_t(N) = Wilder-smoothed average of TR over N bars (14 convention); the platform’s atr_14_pct divides by price for a percentage form comparable across instruments.
• Parameters (engine-executable): window N, smoothing convention (pinned), absolute vs percentage form, and the derived uses — stop distance (k × ATR, mgmt-stop-loss), target distance, tranche spacing (mgmt-scaling), and size-normalization (risk per trade ÷ ATR-scaled stop = shares/contracts — risk-fixed-fractional).
• Not direction: ATR carries no directional information — it is pure magnitude, which is why it composes cleanly with any directional signal.

When it applies

Everywhere scale matters: stop/target placement that adapts to each instrument’s noise floor, volatility-normalized sizing across a book (risk-volatility-targeting uses realized vol or ATR interchangeably at this granularity), regime description (regime-volatility expansion/compression), and replay parameter transfer across instruments.

Risk profile & failure modes

• Backward-looking in vol shifts: ATR lags regime breaks — the bar that triples true range is averaged against 13 quiet ones; post-shock stops sized on pre-shock ATR are too tight (and vice versa).
• Gaps still exceed it: ATR includes past gaps but bounds nothing — a 1-ATR stop does not limit loss to 1 ATR on a 3-ATR gap (mgmt-stop-loss gap caveat).
• Session-definition sensitivity: futures ATR differs by session convention (RTH vs 24h bars); mixing conventions breaks parameter transfer (ms-sessions-auctions).

Evidence & limits

Wilder (1978) is the construction source. ATR itself makes no predictive claim — it is a measurement, and its descriptive validity is definitional. The evidence questions attach to its uses (do ATR-scaled stops beat fixed stops? — a replay question per strategy; volatility clustering, which makes recent ATR informative about near-future ATR, is among the most robust facts in financial econometrics and is covered in regime-volatility).

Falsifiable-thesis examples

Illustrations only, not signals:

• “X’s ATR(14)%, elevated to twice its one-year median, will remain above that median for at least 10 more sessions (clustering)” — falsified by the ATR series.
• “Switching system S from fixed 2% stops to 2-ATR stops improves replay Sharpe this year” — falsified by the paired replay.

Cross-references

• Primary consumers: mgmt-stop-loss, mgmt-scaling, risk-fixed-fractional, risk-volatility-targeting
• The phenomenon behind its usefulness: regime-volatility (clustering)
• Band sibling: indicator-bollinger-bands (standard-deviation scale instead of true-range scale)
• Platform binding: atr_14_pct