Skip to main content

ATR (Volatility & Stops)

Updated Jun 24, 2026 at 2:35pm

Research Draft High 1,368 words

Average True Range (ATR) is a pure volatility gauge, not a directional indicator. Developed by J. Welles Wilder and published in his 1978 New Concepts in Technical Trading Systems, it answers a single question: how much does this instrument typically move in one bar, right now? In a swing context its value is almost entirely instrumental — you rarely trade ATR directly. Instead you use it to translate a chart idea into the two numbers that actually govern survival: where the stop goes and how big the position is. The core tension is that ATR adapts the stop to current noise (good — it stops widening you out during calm and gives room during chaos), but that same adaptiveness means a wider stop forces a smaller position, and a stop placed purely by volatility math can ignore where price structurally should not go.

How it's calculated

ATR is built on True Range (TR), the greatest of three values for the current bar (per StockCharts ChartSchool and Macroption):

1. Current High − Current Low 2. |Current High − Previous Close| 3. |Previous Close − Current Low|

Components 2 and 3 capture gaps — a bar that opens far from yesterday's close has a large true range even if its own high-to-low is narrow. This is why ATR, not the raw daily range, is the honest volatility measure for stocks (which gap on earnings/news).

ATR is then Wilder's smoothed average of TR, default 14 periods. The first value is a simple average of the first 14 TRs; thereafter:

> ATR = [(Prior ATR × 13) + Current TR] / 14

Wilder's smoothing uses a 1/n factor (heavier history weighting than a standard EMA's 2/(n+1)), so ATR reacts more slowly than the same-length EMA. Because ATR is in price units (e.g. "$1.40"), it is not comparable across stocks. ATRP (ATR Percent) = ATR / Close × 100 normalizes it for cross-ticker comparison — a $1.40 ATR on a $200 stock (0.7%) is far calmer than $1.40 on a $20 stock (7%).

How it's used in practice (the setups)

1. ATR stop placement. The standard swing convention is to set the initial stop 2–3× ATR(14) from entry on a daily chart (commonly cited across StockCharts, IG, and OANDA; treat the multiplier as a regime/style choice, not a law). Tighter (1.5–2×) for mean-reversion or low-vol blue chips; wider (2.5–3×+) for momentum/small-cap names that need room. The logic: a stop closer than ~1 ATR sits inside normal one-bar noise and will be hit by random wiggle.

2. Chandelier Exit (trailing stop). Created by Chuck LeBeau, popularized via Alexander Elder. For a long:

> Chandelier Stop = Highest High (22) − ATR(22) × 3

The stop trails up from the highest high since entry and only ratchets higher, never lower — it locks in trend gains while leaving a 3-ATR buffer for pullbacks. Defaults are 22-period (≈ one trading month) and 3× multiplier. Use it to manage a winning swing, not to time entries — LeBeau and ChartSchool both stress it is an exit-only tool that needs a separate entry trigger.

3. ATR / volatility-based position sizing. The decisive use, and the one Van Tharp argues matters more than the entry itself. Fix dollar risk first, let volatility set share count:

> Shares = (Account Equity × Risk %) / (ATR × Stop Multiplier × $/point)

This is the Turtle traders' "N" model (Dennis/Eckhardt defined N as the 20-day EMA of True Range and sized each "Unit" so a 1N move ≈ 1% of account equity; Van Tharp formalized the general case as the "percent volatility" model): size so a 1-ATR adverse move costs a constant fraction of equity. Result — you automatically take fewer shares of a wild stock and more of a quiet one, equalizing risk across very different instruments.

4. Target and expectancy framing. Quote profit targets in ATR multiples too (e.g. risk 2 ATR to make 3–4 ATR) so reward:risk is measured in the instrument's own units rather than arbitrary dollars.

Failure modes a pro watches for: (a) ATR is lagging — it spikes after a vol event, so a stop set the day after an earnings gap may be far too wide; (b) on a low-volume or recently-quiet stock ATR understates real risk before a catalyst; (c) blindly trusting the ATR distance over an obvious structural level — if support is 1.2 ATR away, that level, not the 2-ATR formula, is where price "should not" trade.

Adoption, debate & evidence

ATR is near-universal: it ships in every charting platform and underlies Keltner Channels, the SuperTrend indicator, Chandelier Exit, Parabolic-SAR-style work, and most volatility-targeting quant systems. As a volatility measurement it is essentially uncontested — there is no serious argument that ATR mis-measures realized range.

The contested part is whether stops add expected value. The key academic reference is Kaminski & Lo, When Do Stop-Loss Rules Stop Losses? (SSRN 968338; J. of Financial Markets 2014): mathematically, if returns follow a random walk, a simple stop-loss rule always lowers expected return (you pay the friction for nothing). Stops add value only when returns are positively serially correlated — i.e. in momentum/trending regimes (for a mean-reverting strategy with negative autocorrelation, a stop hurts expected return to first order). Their empirical test on monthly U.S. data (1950–2004) found stop rules added on the order of 50–100 bps per month during stop-out periods versus buy-and-hold. Horizon caveat — important for swing trading: that study used monthly sampling; the authors note stop-losses showed little value at short-term sampling frequencies and helped mainly at longer intervals. So Kaminski-Lo supports the mechanism (stops help in momentum, hurt in mean-reversion) but does not directly validate daily-bar swing stops — treat it as a directional argument, not proof for short-horizon use. Folklore vs measured: the popular "2× ATR / 22-period 3× ATR" defaults are conventions, not optimized constants. The "best multiplier" backtest figures circulating online (specific win rates, expectancy per trade) are vendor backtests on particular samples and periods — useful as direction, not as proven edge; they are sample-specific and rarely out-of-sample validated. The robust, well-supported claim is narrower: volatility-scaled stops out-survive fixed-dollar/fixed-percent stops because fixed stops cluster their hits in high-VIX periods, getting whipsawed out precisely when noise is highest.

Strengths & limitations

Works best: sizing positions to equalize risk; setting stops that breathe with regime; trailing trend-followed swings (Chandelier). Its single greatest contribution is consistent risk — Van Tharp's central claim.

Fails / misused: ATR gives no direction and no signal — it cannot tell you to enter. It lags volatility spikes. The #1 misuse is treating the ATR-multiple stop as a substitute for structure: placing a stop at "entry − 2 ATR" when that lands in the middle of a support shelf (so noise hits it) or just beyond an obvious level everyone can see (so it gets hunted). ATR sets the minimum sensible distance; chart structure should set the actual level, with ATR as the sanity floor. Second misuse: comparing raw ATR across tickers — always use ATRP for that.

Sources

Disputes flagged: (1) Stop-loss expected-value is regime-dependent (Kaminski-Lo) — not a universal edge — and their evidence is on monthly data, so it does not directly prove daily-bar swing stops. (2) Specific "best multiplier" win-rate/expectancy figures from vendor backtests are sample-specific and not treated as established.