RSI vs Williams %R: Differences, Data and Which to Use

Michael Whitman · Trading since 2017, full-time since 2020 · YouTube creator, 137K subscribers · Published 2026-09-08 · Updated 2026-11-01 · Reviewed 2026-11-01

The RSI weighs recent gains against recent losses and smooths the result; Williams %R only asks where today’s close sits between the 14-session high and low. That one difference decides almost everything else.

On SPY, the S&P 500 fund, the two flagged “overbought” on the same day just 17.35% of the time since 1993, and nearly every disagreement was %R calling overbought while the RSI did not.

Most comparisons of these two stop at “%R is faster, RSI is smoother.” This one counts how often they actually agree, works both formulas through on two real SPY sessions from this autumn, and ends with a plain rule for which one to put on your chart.

What each one is

The relative strength index compares average gains with average losses over 14 sessions, each average smoothed by Wilder’s method, and maps the ratio onto 0 to 100. J. Welles Wilder set it out in his 1978 book, with 70 and 30 as the usual lines. The relative strength index covers the calculation step by step.

Williams %R measures where the close sits in the recent range, counting down from the 14-session high: (highest high - close) / (highest high - lowest low) x -100. A close at the high reads 0; a close at the low reads -100. StockCharts gives 0 to -20 as overbought and -80 to -100 as oversold. Williams percent R covers it in full.

That second formula is the fast stochastic inverted. The stochastic oscillator measures the close up from the low; %R measures it down from the high and multiplies by -100, so %R = %K - 100 exactly. That identity held on every one of 14,250 S&P 500 sessions checked.

Where they differ

What drives the reading. A string of small gains moves the RSI slowly, because the average gain has to outweigh the average loss by a wide margin. %R needs only one close near a fresh 14-session high to jump to the top of its scale.

Which way the scale runs. The RSI counts up from weakness. %R counts down from strength, so its “high” readings are its numerically smallest, and a rule copied from one to the other points at the wrong extreme unless the thresholds are rewritten.

A worked example: two SPY sessions, both formulas

Monday 21 Sep 2026: %R says overbought, the RSI says neutral. SPY closed at $773.50. Over the 14 sessions to that day the high was $774.89 and the low $749.60, a range of $25.29.

• Williams %R = ($774.89 - $773.50) / $25.29 x -100 = $1.39 / $25.29 x -100 = -5.50, deep inside the 0 to -20 overbought band.
• The RSI that day was 58.43, nowhere near 70.

The close was within $1.39 of its two-week high, which is all %R reports. But the prior two weeks had held plenty of down days, including a close of $754.05 on 16 Sep, so average losses were still large and the RSI sat just above its midpoint.

Friday 23 Oct 2026: %R says oversold, the RSI says no. SPY closed at $759.97 against a 14-session high of $781.62 and low of $758.68, a range of $22.94.

• Williams %R = ($781.62 - $759.97) / $22.94 x -100 = $21.65 / $22.94 x -100 = -94.38, oversold.
• The RSI read 42.09, still a long way above 30.

Same pattern in reverse: a close near the bottom of a narrow range pins %R, while a balanced mix of recent gains and losses keeps the RSI in the middle. Neither reading is wrong. They answer different questions, and on SPY these splits are the norm rather than the exception.

How often do RSI and Williams %R agree?

Rarely, and the disagreement runs one way. Over 8,437 SPY sessions, there were 3,309 days when at least one oscillator read overbought. Both did on 574 of them. %R alone did on 2,714. The RSI alone did on only 21.

On the oversold side the overlap was thinner still: 110 shared days out of 1,148 on which either was oversold (9.58%), 1,023 days for %R alone and 15 for the RSI alone.

Read the other direction, the RSI almost never stands alone. On 96.47% of the RSI’s 595 overbought days, %R was overbought too, and on 88.00% of its 125 oversold days, %R was oversold too. In 8,437 sessions the two never pointed at opposite extremes on the same day.

So “wait for both to agree” mostly means “wait for the RSI”. Adding %R as a confirmation to an RSI 70 adds almost nothing, because %R was already there. Adding the RSI as a filter to %R removes 2,714 of %R’s 3,288 overbought days, about four in five.

Which one to use: RSI or Williams %R?

Use the relative strength index when you want an extreme to be rare. It reached 70 or above on 7.05% of SPY days and 30 or below on 1.48%, so a reading there marks an unusual stretch.

Use Williams %R when you want to know where the close sits in its recent range, such as whether a pullback has reached the bottom fifth of the last two weeks. Read that way, its frequent extremes are a description, not a warning.

Do not run %R beside the stochastic. They are the same line with the scale flipped.

The original data

Main sample: SPY daily highs, lows and closes from Yahoo Finance (split-adjusted, not dividend-adjusted), 29 Jan 1993 to 30 Oct 2026. The first 60 sessions only settle Wilder’s averages, so counts run from 27 Apr 1993: 8,437 sessions.

RSI(14) uses Wilder’s smoothing; %R(14) uses the 14-session high and low including the current day. Overbought means RSI 70 or above, %R -20 or above; oversold means RSI 30 or below, %R -80 or below.

On 47.17% of SPY sessions neither oscillator was at either extreme. The two disagree far more often than they agree, and almost all of the disagreement is %R flagging days the RSI leaves alone.

A longer history: the S&P 500 since 1970. The same settings on Yahoo’s ^GSPC bars, 14,250 sessions from 31 Mar 1970 to 2 Oct 2026, gave the same shape. %R sat at -20 or above on 5,005 days (35.12%) and entered that zone 20.39 times a year; the RSI sat at 70 or above on 1,229 days (8.62%) and entered it 5.73 times a year.

On the oversold side, %R sat at -80 or below on 2,174 days (15.26%) and the RSI at 30 or below on 395 (2.77%).

Which one gets there first: QQQ since 1999. On 6,890 QQQ sessions from 4 Jun 1999 to 23 Oct 2026, %R was already at -20 or above at 148 of the RSI’s 149 moves up to 70 (99.33%), and on 134 (89.93%) it had been there since an earlier session. Of %R’s 545 overbought stays, the RSI reached 70 during only 120 (22.02%), and of its 456 oversold stays the RSI reached 30 during only 54 (11.84%). Put simply, when the RSI reaches 70, %R is almost always there already.

Recomputed on Nasdaq’s own QQQ bars for 2,454 sessions from 20 Jan 2017, %R’s overbought share came out at 43.24% against 43.15% on Yahoo’s, and the RSI’s at 12.63% against 12.67%. The two feeds agree to within a tenth of a percentage point, so the gap is not a quirk of one data source.

What these counts are not. They describe when each oscillator reaches its lines and whether the other agrees. They say nothing about what price did next; no signal was tested. Every SPY day’s RSI and %R is in the SPY oscillator file, and the S&P 500 days are in the S&P 500 RSI and %R file.

When it fails

The costly failure is fading every %R overbought reading as if it were an RSI 70. Move an “overbought means sell” rule from the RSI to %R and on SPY it fires on almost two days in five, because a rising market keeps closing near its two-week high.

Treating agreement as two signals. An RSI 70 that %R confirms is close to automatic: %R agreed on 96.47% of those SPY days.

Reading %R’s zero as weakness. Zero is a close at the very top of the range.

Treating a long %R stay as exhaustion. On the S&P 500, %R’s longest run at -20 or above lasted 27 sessions.

Reading a quiet RSI as a quiet market. On 23 Oct 2026 the RSI’s 42.09 looked unremarkable, yet SPY had closed $21.65 below its two-week high, near the bottom of the range. If the question is “how far has price pulled back inside its recent range?”, the RSI is the wrong tool and %R answers it directly.

Swapping one for the other without moving the lines. A screen built on “RSI below 30” that is switched to %R must become “%R at -80 or below”, not “%R below 30”, which is every reading %R can ever give, because its scale tops out at 0.

Questions people ask

Is Williams %R better than RSI?

Neither is better in general; they measure different things. %R is the better tool for “where is the close in its recent range?” The RSI is the better tool when an extreme should be rare. For a full walk-through of %R on its own, see how to use Williams %R.

Which is faster, RSI or Williams %R?

Williams %R. It has no smoothing, its median daily move on the S&P 500 was 9.34 points against 2.73 for the RSI, and on QQQ it was already overbought at 148 of the RSI’s 149 moves up to 70.

Can you use RSI and Williams %R together?

You can, but expect them to disagree most of the time: on SPY they shared only 17.35% of overbought days. If you combine them, know that the RSI is doing almost all the filtering.

Why is Williams %R negative?

Larry Williams measured the close down from the high and multiplied by -100, so the scale runs from 0 at the top of the range to -100 at the bottom.

Is Williams %R the same as the stochastic?

It is the fast stochastic’s %K minus 100, with no %D smoothing line. Same information, flipped scale.

Related

The relative strength index covers Wilder’s averages of gains and losses. Williams percent R covers the inverted range reading measured here, and stochastic covers the calculation %R turns upside down.

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