Stochastic RSI vs Williams %R

Michael Whitman · Trading since 2017, full-time since 2020 · Published 2026-09-09 · Updated 2026-10-04 · Reviewed 2026-10-04

These two are often filed together because both are bounded and both are read as overbought or oversold. They are built from different inputs, though, and measured on three decades of one fund they behave differently in ways a chart reader can count.

What each one is

Williams percent r places the close inside the recent high-low range. With the standard 14-bar setting it is −100 × (highest high of 14 bars − close) ÷ (highest high − lowest low), so it runs from −100 at the bottom of the range to 0 at the top. There is no smoothing. The ratio by itself reads 0 at the top of the range and 1 at the bottom, upside down from price; the −100 multiplier, which is how StockCharts’ ChartSchool writes Larry Williams’ indicator, flips it so the line rises with price and moves the decimal. Williams %R covers it in full.

The stochastic relative strength index applies the same range idea to an indicator. First the 14-bar relative strength index is computed with Wilder’s smoothing. Tushar Chande and Stanley Kroll, who introduced it in their 1994 book The New Technical Trader, defined it as (today’s reading − lowest of the last 14 readings) ÷ (highest − lowest), a value from 0 to 1 with no smoothing. Most charting platforms multiply that by 100 and add two averages: %K, a 3-bar average of the raw value, and its signal line %D, a 3-bar average of %K. This page uses that platform form. Stochastic RSI covers the construction.

One reads the bar’s high and low; the other never sees them. The relative strength index is built from closes only, so a long wick that closes back in the middle registers on Williams percent r and not at all on the stochastic version.

Where they differ

What the range is measured against. Williams uses the actual extremes of the bars. The stochastic version uses the highest and lowest indicator readings, which can be a narrow range even in a week when price moved a lot.

How often each sits on its boundary. Because the raw stochastic version measures a reading against its own last 14 values, any new high in the relative strength index puts it at exactly 100. Williams percent r only reaches exactly 0 when the close is the high of the whole 14-bar range.

Whether anything is smoothed away. Williams shows every jump. The stochastic version’s %K and %D iron those out, which makes crosses easier to act on and hides how noisy the raw reading was.

How you are meant to act. Williams gives a level and leaves the decision to you. The stochastic version gives two lines and a crossing, a discrete event that a written rule can name.

Where Stochastic RSI and Williams %R agree

Both are bounded, so neither can report that a strong move has become stronger once it is at the top of its scale.

Both are read as reversal warnings, and the data below does not support that reading on either. An extreme reading means the recent move was strong, and on a fund that drifted upward, strength more often came before more of the same than before a turn.

And both cost a round trip each time a reading is acted on. The faster a tool enters its zones, the more round trips it suggests, which is where the counts per year in the data section matter.

Which one to use: Stochastic RSI or Williams %R?

Run williams percent r when you want the state of price itself. It uses the highs and lows, so a rejection bar shows up, and nothing is averaged away.

Run the stochastic relative strength index when you need a discrete event, a %K and %D cross, for a rule you can write down and test. Accept that its raw form spends long stretches pinned at 0 or 100.

Run either one as a description, not a trigger, when the trade is against the trend. Neither tool’s oversold or overbought entries separated what SPY did next from an ordinary day.

And run both together only when you know what each adds. They are related, not identical: on SPY the daily changes correlated 0.50, so a disagreement between them is information about the difference between price range and indicator range, not a vote.

The original data

The test. Yahoo Finance SPY daily bars, high, low and close as traded (not adjusted), from 2 January 1996 to 2 October 2026, 7,739 sessions. Both tools were computed exactly as written above: Williams percent r over 14 bars, the relative strength index over 14 with Wilder’s smoothing, the stochastic version over 14 readings with a 3-bar %K and 3-bar %D. The zones are the usual ones: −20 and −80 for Williams, 80 and 20 for %K. The test starts in 1996, in line with this site’s other SPY tests, because SPY’s 1993 to 1995 bars are less reliable: 9.36% of their opens repeat the previous close, against 2.53% from 1996 to 2000. The earlier closes are used only to start the relative strength index.

How alike are they? Across 7,738 day-to-day changes, the change in Williams percent r and the change in %K correlated 0.50. Against the raw, unsmoothed stochastic version the correlation was 0.85, so much of the gap is the smoothing, and the rest is the different input. A correlation of 0.50 is not “the same half of the time”: the two moved in the same direction on 62.68% of days, and the 0.50 measures how closely the sizes of their daily moves line up as well. On the 1,081 days Williams was at −80 or below, %K was at 20 or below on 70.03% of them; on the 1,559 days %K was there, Williams agreed on only 48.56%.

Pinned at the boundary. The raw stochastic version sat at exactly 100 on 1,293 days (16.71%) and at exactly 0 on 992 (12.82%), together 29.53% of all sessions. Williams percent r sat at exactly 0 on 59 days (0.76%) and at exactly −100 on 27 (0.35%), 1.11% together. Even smoothed, %K sat at exactly 100 on 344 days and at 0 on 181.

What followed a zone entry. For each tool, an entry is the first close inside a zone after a close outside it. Entries were kept only when at least 10 sessions apart, and each was checked against SPY’s close 10 sessions later, dividends included. Across all 7,729 sessions with 10 sessions after them, SPY was higher on 61.43%.

All four are within chance (normal approximation to the binomial, two-sided). Oversold entries were followed by a slightly higher share and overbought entries by a slightly lower one, but none of the gaps is large enough to tell apart from an ordinary day on samples this size. Every entry with 10 sessions after it is in the zone entries file and every daily value in the daily readings file.

Fewer videos name the stochastic version, and they draw more views. In the 24,971-video search study behind this site, Williams percent r is in 35 titles across 34 channels at a median of 6,222 views; the stochastic relative strength index is in 27 titles across 23 channels at a median of 30,393.

Why the RSI version pins and the price version hovers

A range of readings is narrower than a range of prices. The relative strength index moves slowly from bar to bar because of its smoothing, so its own 14-bar high and low can sit close together, and any reading at the edge of that range scores exactly 0 or 100.

A close equal to the highest high of 14 bars is far rarer than an index reading at the top of its last 14. Williams percent r needs the close to equal the highest high of the window, and most strong closes still sit a little under the day’s own high or an earlier high. So it spends long stretches near the top without touching it.

That is why the stochastic version needs %K and %D at all. Without them, more than a quarter of sessions would sit exactly on the edges.

What the %K and %D smoothing costs

It converts a level into an event, and it delays the event. The %K and %D cross happens after the raw reading has already turned, the same trade every smoothed indicator makes: fewer false moves, each one later.

And it hides the condition that should slow you down. A raw oscillator slamming between its boundaries is telling you the market is disorderly. Smoothed, the same stretch can look like a series of orderly crosses.

When it fails

The characteristic failure is reading two tools as two votes. When both are in a zone it feels like confirmation, but on SPY a large share of lower-zone days were shared by both, because both respond to the same falling closes. The days they disagree are the useful ones, and they say something about the input (price range against indicator range), not about direction.

A second failure is using either one to exit a trend. On a fund that rose over most of the period, upper-zone entries were followed by gains about as often as any other day. Selling the first overbought reading meant selling into the more common outcome.

A third is treating a pinned 100 as extra strength. The raw stochastic version reaches exactly 100 whenever its input sets a 14-bar high, which on SPY happened on about one session in six. A reading that common is not a signal by itself.

A fourth is tuning the lookback until the last reversal is caught, which fits the sample and nothing beyond it.

And a fifth is ignoring the cost of acting on every entry. Either tool enters its zones dozens of times a year on a daily chart, and each entry acted on is a round trip paid before anything is known.

Related

Williams %R covers the raw range-position reading and why it matches the fast stochastic. Stochastic RSI covers the twice-derived construction and the signal line. And RSI covers the measure the stochastic version is built from.

This page is educational, not financial advice. Test every idea on your own charts before risking money.