Keltner & Donchian Channels

Three ways to wrap price in an envelope

Bollinger Bands use standard deviation of returns to build a volatility envelope around a moving average. That's one answer to "how wide should the channel be?" There are two other historical answers — older than Bollinger — that professional traders still use every day.

Keltner Channels (Chester Keltner, 1960) use ATR instead of σ. Smoother, less reactive to sudden noise.

Donchian Channels (Richard Donchian, 1960s) use N-bar high and low — no math at all, just extremes. The cleanest, most literal channel you can draw.

Why three? Because "how much room does price need?" has three reasonable answers:

Kaufman ran all three (plus percentage-based bands) on the same chart and concluded:

The first two methods of calculating bands, the percentage of trendline and percentage of price, are almost identical, very smooth, and are the farthest from the center. The next band closer to the moving average is the average true range. It moves slightly farther apart when prices are more volatile. The band closest to the trendline is the standard deviation, which is more sensitive to price volatility. — Perry Kaufman, Trading Systems and Methods

Bollinger = tight and reactive. Keltner = medium and smooth. Donchian = stepwise and literal.

Keltner Channels — 1960 original

Chester Keltner published his method in How to Make Money in Commodities (1960). The formula Kaufman reproduces:

A 10-day moving average of typical price, bracketed by the 10-day average daily range. Kaufman's aside on why 10 is there at all:

In an era before the pocket calculator, who knows how much impact that convenience had on Keltner's choice? [10 periods lets you divide by shifting a decimal.]

Kaufman's signal rule, per Keltner's original: "A buy signal occurs when the price crosses above the upper band and a sell signal when the price crosses below the lower band; positions are always reversed." Same stop-and-reverse structure as Parabolic SAR, different width math.

Keltner Channels — modern Linda Raschke variant

The Keltner you'll find on every modern charting platform is not Keltner's 1960 version. Murphy attributes the modernization directly:

The Keltner channels were originally developed by Chester Keltner in his 1960 book. Linda Raschke, a very successful commodity trader, has reintroduced them to technicians. In her modification, the bands are also based on the average true range (ATR), but the ATR is calculated over 10 periods. This ATR value is then doubled and added to a 20 period exponential moving average for the plus band and subtracted from it for the minus band. — John Murphy

So:

Three differences from the original:

1. Middle: EMA(20) instead of SMA(10). Longer, smoother, more recency-weighted.
2. Band width: Wilder's ATR instead of average daily range. ATR captures gaps.
3. Scaling: 2 × ATR instead of 1 × avg(H−L). Wider.

Raschke's version is what the TTM Squeeze (below) assumes, what virtually every backtest uses, and what our keltnerChannels() function implements.

Donchian Channels — the purest breakout

Richard Donchian was working at Hayden Stone in the 1960s when he formalized what he called the 4-Week Rule. Kaufman:

Richard Donchian created what now seem to be very simple trading methods, but in the 1960s while working at Hayden Stone, he implemented a number of technical systems that were groundbreaking. They included moving average and breakout systems. At the time, any systematic trend-following method was state-of-the-art and profitable.

And the 4-Week Rule, verbatim:

Go long (and cover shorts) when the current price exceeds the highs of the previous four calendar weeks. Sell short (and liquidate longs) when the current price falls below the lows of the previous four calendar weeks. — Kaufman

4 weeks = 20 trading days. The channel is defined by the extremes of that window:

No smoothing, no volatility scaling, no parameters beyond N. Price cannot exceed the upper line without redefining it — the moment it makes a new N-bar high, it is the upper line.

Murphy on N = 20:

The 4 week, or 20 day, trading cycle is a dominant cycle that influences all markets. This may help explain why the 4 week time period has proven so successful.

Murphy's blunter take:

The 4 week rule is simple, but it works.

The validation — Donchian's 60-year track record

This is the clean statistical case for a channel breakout. Kaufman's testing results:

23-market N-breakout test, 2000–2017 (kaufman.txt:8611):

The average 'best' calculation period was 93 days, typical of a macrotrend approach. Of the 23 tests performed on each market, seven of the markets had tests that were all profitable... and the average percentage of profitable tests was 79%.

79% of parameter variations were profitable. For comparison, Kaufman's MACD crossover test had only 5–8% of parameter combinations profitable (see our MACD lesson). The breakout rule is fundamentally more robust.

Corn futures, 1961–present (Kaufman Figure 8.26): Donchian's original 5-and-20-day moving average rule produced positive cumulative P&L for 60 continuous years. Kaufman: "Although the rate of return has slowed, it seems remarkable that a simple method could have been consistently profitable for 60 years."

And the famous Playboy's Investment Guide quote Kaufman preserves:

Childishly simple… was recently discovered to rank premiere among a dozen widely followed mechanical techniques.

The independent 1970 Dunn & Hargitt study ranked Donchian's 4-Week Rule first among all mechanical commodity trading systems tested (Murphy murphy.txt:3774).

Turtle Trading — Donchian in its famous form

Kaufman's direct statement: "Readers will recognize that this is the basis for the Turtles' trading method."

Richard Dennis and William Eckhardt's 1980s Turtle experiment ran two Donchian systems in parallel:

System 1 — aggressive:

• Enter long when today's high > 20-day high
• Exit long when today's low < 10-day low
• (Mirror for shorts)
• Skip entry if the previous S1 trade was profitable (unless it lost ≥ 2×ATR)

System 2 — conservative:

• Enter long when today's high > 55-day high
• Exit long when today's low < 20-day low
• (Mirror for shorts)
• No skip filter

Position sizing was 2×ATR from entry (full ATR position-sizing — which you learned in the ATR lesson). This is the archetypal Donchian breakout system, and its 4-decade track record is why "trend-following" is a real strategy category and not just folklore.

We'll cover the Turtle rules in their own lesson in the Strategies unit. For this lesson, just notice: every Turtle system is a Donchian channel, nothing more.

Play with it

Toggle between Keltner, Donchian, Both, and TTM Squeeze. Notice:

• Donchian has the stepwise structure — its upper line only moves when a new N-bar high is set. In long rallies the upper line looks like a staircase.
• Keltner curves smoothly with the EMA. The ATR bands expand in volatile moves and contract in quiet ones, but always smoothly.
• Flipped to Both, you see both at once. Donchian (blue) always sits outside Keltner (orange) in range-bound markets — because Donchian can't tighten mid-range, only step outward.
• TTM Squeeze adds Bollinger Bands. Watch for periods when the Bollinger lines pull inside the Keltner lines — that's the compression signal.

The TTM Squeeze — Bollinger inside Keltner

This deserves careful sourcing. The "TTM Squeeze" name is John Carter's (2008 book Mastering the Trade) — he trademarked it, sells a paid indicator under that name, and the mechanic is widely used in retail. Our reference library does not contain his book.

What our books do contain: the same mechanic attributed to a different author. Kaufman (kaufman.txt:12825):

Williams suggests that other indicators can be combined to capture volatile moves after a price contraction: A standard 20-day, 2 standard deviation Bollinger band; A 20-day Keltner Channel; A 21-day Chaikin Oscillator to monitor the flow of funds. To enter a long position… The Bollinger bands narrow so they are fully inside the Keltner Channel while the Chaikin oscillator is below zero. The Chaikin Oscillator crosses above the zero line.

The logic: Bollinger contracts faster than Keltner (σ of returns reacts faster than ATR-smoothed ranges). When σ drops enough that Bollinger falls inside Keltner, volatility is unusually compressed. The premise is that compression precedes expansion — a burst is coming.

The honest verdict: none of our reference books run a statistical test on this pattern. It's widely believed, widely traded, and Kaufman/Williams both endorse it conceptually. Treat it as a hint to watch for breakouts, not a standalone signal. A squeeze can compress for weeks and then break the "wrong" direction; direction requires another filter.

How to choose between the three

No book in our library says "use X for Y." The convention synthesized from Kaufman and Murphy:

• Bollinger Bands — use when you care about statistical deviation of closes. Good for mean-reversion fades at 2σ touches in range-bound markets. Most reactive to single-bar volatility spikes.
• Keltner Channels — use when you want a smoother envelope. ATR-based bands don't blow wider on one noisy candle. Better for trend-following when you want the EMA as a true mid-line.
• Donchian Channels — use when you want a breakout signal. Price crossing a Donchian line is a new N-bar high/low. This is the channel that turns into a trading system with the least modification.

The three don't compete — they answer different questions. Pros typically overlay at least two on the same chart.

Hidden traps

• Thinking Keltner ≈ Bollinger. They use different math and respond differently to spikes. A 2σ Bollinger band and a 2×ATR Keltner band reach different widths in the same market.
• Using Donchian in ranges. Same failure mode as SAR and MA crossovers — Donchian breakouts whipsaw repeatedly in a 3-month consolidation. Murphy's fix: widen N (8-week instead of 4-week) in choppy markets.
• Running a continuous Donchian always-in-market system. Murphy: "continuous systems have a basic weakness. They stay in the market and get 'whipsawed' during trendless market periods." The Turtles' fix was asymmetric entry/exit (20-day entry, 10-day exit) — this filters some whipsaws at the cost of earlier exits.
• Over-optimizing N. Kaufman's 2000–2017 test found ~93 days was the average best, but that's a meta-optimum. Don't grid-search N per market — you'll curve-fit.
• Confusing the middle lines. Keltner's middle is an EMA. Donchian's "middle" is just (upper + lower) / 2 — a mathematical artifact, not a real trend line. Trading pullbacks to Donchian's middle doesn't carry the same meaning as pullbacks to Keltner's EMA.
• TTM Squeeze as standalone signal. It's a compression warning, not a direction call. Pair with something that gives direction (MACD, price structure, Chaikin Oscillator per Williams's original formulation).

Quick check

What you now know

• Three envelopes: Bollinger (σ-based), Keltner (ATR-based), Donchian (extreme-based). Three different math, three different widths, three different reactive characters.
• Keltner 1960 original: SMA(10) of typical price ± SMA(10) of (H−L). Keltner picked 10 partly because of decimal-shift division convenience (pre-calculator era).
• Keltner modern (Raschke): EMA(20) ± 2·ATR(10). What every charting platform plots. ATR captures gaps; EMA is smoother than SMA.
• Donchian 4-Week Rule: Buy on 4-week (20-day) high; sell on 4-week low. The earliest recorded N-day breakout system.
• 79% of N-day breakout parameter variations were profitable across 23 markets 2000–2017 (Kaufman). One of the most robust statistical results in the TA literature.
• 60 years of continuous positive P&L for Donchian's 5-and-20-day MA system on corn futures (Kaufman).
• Turtle Trading is Donchian channels. System 1: 20-day entry / 10-day exit. System 2: 55-day entry / 20-day exit. Plus 2×ATR sizing and stops.
• TTM Squeeze (John Carter, 2008) is Bollinger-inside-Keltner compression. Kaufman has the same mechanic via Williams; the "TTM" branding is modern retail, not sourced here.
• Use Bollinger for reversion, Keltner for smoother envelopes, Donchian for breakouts. Pros overlay multiple.
• Always-in Donchian systems whipsaw in ranges. The Turtle fix: asymmetric entry/exit periods.

Next: Ichimoku Cloud — the Japanese multi-component indicator that bundles support, resistance, direction, and momentum into a single glance.