Williams %R mean reversion: when oversold actually means buy.

Most "oversold" indicators don't work for buying. Stochastic oscillators, RSI extremes, MACD crossovers — tested rigorously, they fire false signals as often as they fire real ones. The pattern is consistent enough that experienced traders learn to ignore raw oversold readings entirely.

Williams %R is different — not because the indicator itself is special, but because of the specific filter conditions that make it useful. When applied carelessly, %R produces the same noise as any other oscillator. When applied with strict trend and volume filters, it becomes a sharp tool for catching short-term reversals in already-strong stocks. This article shows you the difference.

What Williams %R measures

Williams %R, developed by Larry Williams in 1973, is a momentum oscillator. It measures the current closing price relative to the highest high and lowest low over a lookback period (typically 14 days).

The formula:

%R = (Highest High − Close) / (Highest High − Lowest Low) × -100

The output ranges from 0 to -100. A value of -20 means today's close is in the top 20% of the recent range. A value of -80 means today's close is in the bottom 20% of the recent range.

The traditional interpretation:

• Above -20 (close to 0): "overbought"
• Below -80 (close to -100): "oversold"

The traditional trade: buy oversold (%R < -80), sell overbought (%R > -20).

This traditional approach loses money over time. Across years of backtests, raw %R signals have hit rates close to 50%, with frequent whipsaws and large drawdowns during trending periods. Buying oversold readings in a downtrending stock is just buying a falling knife. Most retail traders who try this approach quit within months.

Why mean reversion fails without filters

The fundamental problem with raw oscillator-based mean reversion: it assumes prices will revert to the mean. But prices don't always revert. They sometimes trend. They sometimes break down structurally. An oversold reading in a stock that's genuinely declining isn't a buy signal — it's a stage marker on a longer fall.

Oversold doesn't mean cheap. Oversold means "recently sold off." In a real downtrend, oversold can stay oversold for weeks while the stock keeps falling.

This is why every successful mean-reversion approach in equity markets shares one common feature: it only takes signals in stocks that are already in established uptrends. The strategy isn't betting that the stock will reverse from a downtrend — it's betting that a temporary pullback within an uptrend will resolve back up.

This is the only context where mean reversion has consistent edge. In trending stocks, short-term oversold readings often coincide with normal pullbacks that find buyers and resume the trend. The %R signal isn't predicting a reversal — it's timing a re-entry into an existing trend.

The Williams %R swing-trading rule set

Here's a strict, filtered implementation that has historically held up better than naive oversold-buying:

Why this filtered version works

Three structural reasons:

1. The trend filter eliminates the worst signals

Buying %R oversold in any random stock is a coin flip. Buying %R oversold in stocks that have just made 50-day highs is a different statistical population entirely. You're buying weakness in stocks that just demonstrated strength — the regression-to-trend dynamic genuinely operates in this group.

2. The 2-4 day persistence requirement filters single-day noise

A single-day %R spike below -80 is often just intraday volatility that gets bought back the same day. By requiring the indicator to stay oversold for 2+ days, you ensure the pullback is real — meaning supply has actually been available at lower prices, and that supply will be exhausted before the trade fires.

3. The exit target is realistic

Mean-reversion trades shouldn't target new highs. They should target the midpoint of the range. A %R reading crossing back above -20 (overbought) is the natural exit — it means the pullback has fully resolved, and any further upside requires the stock to begin a new advance, which is a different setup entirely. Exiting at -20 captures the mean-reversion edge cleanly without overstaying.

Where Williams %R fails

Failure mode 1: Trend changes

The strategy assumes the existing uptrend will resume after the pullback. When a stock is actually transitioning from Stage 2 to Stage 3 (topping), pullbacks that look like normal mean-reversion setups instead resolve into deeper declines. You'll see a series of trades that "almost worked" before the stop is hit. Mitigation: a sequence of 2-3 consecutive Williams %R losses in different stocks often signals the broader market regime is changing — pause new entries and reassess.

Failure mode 2: Gap-down events

Mean-reversion stops are typically 4-7% below entry. A negative news event or earnings surprise can gap a stock 8-15% lower overnight, exceeding the stop. You exit at a much worse price than planned. There's no perfect mitigation — gap risk is intrinsic to overnight positions — but avoiding entries before scheduled events (earnings, RBI policy days) reduces the exposure.

Failure mode 3: Strong-trend exhaustion

In stocks that have moved very far above their moving averages (extended runs), %R oversold readings can occur during the natural cooling-off period of an exhausted advance. The pullback that looks like a mean-reversion setup is actually the start of a deeper correction. Mitigation: avoid entries when the stock is more than 15-20% above its 50-day SMA. Such extreme extension often precedes deeper retracements that don't fit the strategy's profile.

Williams %R vs RSI — why one over the other

RSI (Relative Strength Index) is more popular than Williams %R, but functionally they capture similar information. Williams %R has two minor advantages for swing traders:

• %R is more responsive. It moves faster than RSI in the same conditions, meaning entries can be slightly earlier — useful for short-duration mean-reversion trades where timing matters.
• The fixed range (0 to -100) makes thresholds more interpretable. RSI thresholds (typically 30/70) are slightly less intuitive than %R's -80/-20 because %R is symmetric around -50.

For practical purposes, an RSI-based version of the same strategy with appropriately adjusted thresholds (e.g., RSI < 30 / RSI > 50 reversal) would behave very similarly. The choice between %R and RSI is more about personal preference than meaningful edge difference.

Performance considerations

The filtered Williams %R approach is positive-expectancy in trending markets, particularly for stocks that have a history of mean-reverting after pullbacks. Specific performance depends on the strictness of the trend filter, the universe tested, and the broader market regime.

A general observation: mean-reversion strategies tend to have higher hit rates but smaller average winners than trend-following strategies. A typical filtered %R implementation might win 50-60% of trades, but average winners might only be 1.2-1.5x average losers. The math still works out positive, but the per-trade payoff is more modest than with breakout-based strategies. The trade-off is more frequent activity and smoother equity curves.

What this means for you

Williams %R, applied with strict filters, gives you a way to participate in pullbacks within trends. It complements the breakout strategies covered earlier — TPB, Darvas Box, VCP — which all buy strength. %R buys weakness within strength, which is structurally a different setup with different timing characteristics.

For traders running multiple strategies, %R signals tend to fire on different days than breakout signals, providing genuine diversification within a single account. A stock that just had a TPB breakout won't generate a %R buy signal until it has a meaningful pullback days or weeks later — meaning the same stock can be traded twice through different strategies.

The next strategy article covers Bollinger Band Squeeze — another approach that combines volatility and momentum, but with a focus on detecting low-volatility periods that precede explosive moves.