Double trend exhaustion

Williams %R, in one formula

Larry Williams introduced %R in 1973 as a momentum oscillator for futures markets. The whole indicator boils down to one equation:

Read in English: take the highest high over the past n bars and the lowest low over the same window. Where does today's close sit inside that range? Express it as a percentage, then flip the sign so the indicator is bounded between −100 (close at the low) and 0 (close at the high).

That's the whole tool. Nothing fancy underneath: no machine learning, no proprietary normalization, no Fed model. Just “where in the recent range did we close, on a fixed scale.”

Close to −100 means exhausted

The convention is that %R values near −100 indicate an oversold market — the close is pinned to the bottom of its recent range — and values near 0 indicate an overbought market. Practitioners typically draw two horizontal threshold lines (commonly −80 and −20) and treat readings beyond them as the working extremes.

Williams %R is in the same family as RSI and Stochastics. All three answer the same general question (“is price stretched relative to its recent range?”) with slightly different math. %R uses the simplest version: the close inside the high-low envelope, no internal smoothing, no rate-of-change calculation. That simplicity is a big part of why it has lasted.

Noisy or laggy — pick one

Run %R on one window and you're stuck with a tradeoff. A short window reacts quickly. It catches every minor pullback, including the ones that mean nothing, and it pings into “oversold” on routine 2–3% dips. A long window ignores small wobbles but lags badly: by the time the indicator confirms an oversold turn, the bounce is already weeks old.

Every momentum tool runs into this same dilemma. The plain answer is that no single window length is good at both rejecting noise and being on time. You either accept false alarms, or you accept being late.

EMA over the raw line

Raw Williams %R is jagged. A single down day can push it from −60 to −95 and back. Acting on the raw value means acting on noise. The standard fix is to smooth %R with an exponential moving average (EMA), which weights recent values more heavily than older ones. An EMA damps the day-to-day chop without adding much lag. The line still bends at the same time, it just stops shaking so much.

The indicator smooths both of its %R lines this way. The shapes stay the same, but the noise floor drops and the threshold crossings start to mean something.

The short clock and the long clock

The core idea behind the indicator is to quit choosing between “noisy and fast” vs “clean and slow” — and to require both at once. The engine runs %R on two windows and smooths each:

When both lines sit at or below the deep-oversold threshold, the market is exhausted on the short clock and the long clock simultaneously. One alone is noise. Both together is real stress.

Oversold first, then the turn

Deep-oversold by itself is a warning, not a signal. Markets can stay oversold for weeks. The actual trigger is the exit: the bar where both smoothed %R values rise back above the deep-oversold threshold after having sat below it on the prior bar.

That moment — not the descent into oversold — is what gets passed up to the regime gate and the confirmation gate. This matters because waiting for the turn rather then the entry gets rid of a whole class of false alarms automatically. Markets that go deeply oversold and stay there do not fire. Only the ones that bend back up do.

A dip-buying tool, not a top-caller

Williams %R is symmetric — it computes overbought extremes near 0 as cleanly as oversold extremes near −100. The indicator ignores the overbought half entirely. The reason is empirical: on broad-index daily charts, oversold reversals carry strong forward expected returns; overbought reversals do not. Markets can stay overbought for years (look at any structural bull run) without producing exploitable reversals. Equities drift up over time, and that asymmetry is the reason.

So the engine only reads the side where the math has paid off historically. It is a dip-buying tool, not a top-calling tool.

The reversal is one of three gates

A raw reversal candidate by itself is not the customer-facing signal. The full product layers two more gates on top: a four-state regime classifier that throws out reversals in calm or correction markets, and a confirmation gate that requires more evidence in bear regimes than in panic regimes.

Williams %R, doubled, is what fires the candidate. The regime classifier decides whether to listen. The confirmation gate decides whether the candidate is strong enough to publish.