What Is RSI? The Relative Strength Index Worked by Hand and Tested on Five Funds
Michael Whitman
·
Trading since 2017, full-time since 2020
·
YouTube creator, 137K subscribers
·
Published 2026-09-05
·
Updated
·
Reviewed
RSI, the relative strength index, compares the size of recent up moves to recent down moves and reports the result on a scale of 0 to 100. Readings above 70 are called overbought and below 30 oversold, yet on SPY since 1994 what followed either crossing was within chance of an ordinary day.
Covered on this page: TradingView.
RSI, the relative strength index, is a 0-to-100 score of how the average recent gain compares with the average recent loss. Above 70 is called overbought and below 30 oversold, but tested on five funds since 1994, neither line, nor the popular “buy at 40 in an uptrend”, was followed by anything chance could not explain. What RSI does show clearly is which way the trend has been running.
How it works
Each close is sorted into a gain or a loss, the two are averaged, and the averages are compared. J. Welles Wilder Jr. published the method in his 1978 book New Concepts in Technical Trading Systems and used 14 periods, which is still the usual default.
- Split each change. A higher close is a gain and its loss is zero; a lower close is a loss, written as a positive number, and its gain is zero.
- Start the averages. The first average gain is the plain mean of the first 14 gains, and the first average loss the plain mean of the first 14 losses.
- Roll them forward. Every later average = (yesterday’s average × 13 + today’s value) ÷ 14. This is Wilder’s smoothing: older bars never fully drop out, they just keep losing weight.
- RS = average gain ÷ average loss, and RSI = 100 − 100 ÷ (1 + RS). Equal averages give 50; a run with no losses at all gives 100.
Everything RSI knows comes from the closes on your chart. It has no idea where support is, what volume did or how far price sits from its last swing.
A worked example
SPY, 30 October 2026. At the close on 29 October ($769.36) the smoothed average gain was about $2.058 and the average loss about $2.011, so RSI(14) read 50.57: gains and losses almost exactly balanced. On 30 October SPY closed at $755.72, a fall of $13.64.
- New average gain = ($2.058 × 13 + $0) ÷ 14 ≈ $1.911
- New average loss = ($2.011 × 13 + $13.64) ÷ 14 ≈ $2.842
- RS = $1.911 ÷ $2.842 ≈ 0.672
- RSI = 100 − 100 ÷ 1.672 ≈ 40.2; unrounded, the script gives 40.20
One session moved the reading ten points, because a single loss more than six times the average loss entered the loss side at a weight of one fourteenth. The same arithmetic tells you what oversold would have taken: to close the next session at an RSI of 30, SPY would have had to fall another $21.01, or 2.78%, in one day.
SPY was still above its 200-day average that day ($728.85), so 40.20 was the kind of reading that uptrend guides call support. The sections below test what such readings were followed by. The entry itself comes from the chart, not the panel: market structure supplies the price at which an idea is wrong, which RSI cannot.
RSI in an uptrend and a downtrend
The most useful thing RSI shows is where it lives, and that depends on the trend. Fidelity’s indicator guide says RSI tends to stay between 40 and 90 in an uptrend and between 10 and 60 in a downtrend; Wikipedia attributes slightly tighter ranges, 40 to 80 and 20 to 60, to the analyst Andrew Cardwell. Neither page shows a count, so here is one.
Splitting SPY’s 8,263 sessions since January 1994 by whether the close was above or below its 200-day simple average:
- Above the average (6,192 sessions): RSI between 40 and 90 on 95.24%, between 40 and 80 on 94.85%. Under 30 on 18 sessions, 0.29%.
- Below the average (2,071 sessions): between 10 and 60 on 93.38%, between 20 and 60 on 92.95%. Over 70 on 4 sessions, 0.19%.
The other four funds landed in the same place. Above the 200-day, QQQ, IWM, GLD and TLT spent between 93.51% and 94.53% of sessions in the 40 to 90 band; below it, between 91.24% and 92.25% in the 10 to 60 band.
Part of this is built in, and it matters. A close above a 200-day average usually follows a run of gains, and gains are exactly what lift RSI, so the ranges describe the trend rather than add a second opinion about it. Constance Brown’s point, as Investopedia summarizes it, is that oversold in an uptrend sits well above 30. On SPY the count backs that up: in a rising market, 30 almost never came.
Is an RSI of 30, or 40, a buy signal?
Not on its own, in any of the five funds. The test takes the first close beyond each line, counts a crossing only if it comes at least 20 sessions after the last one counted (so no two 20-session windows overlap), and asks whether the fund was higher 20 sessions later. The comparison is every session of the same kind: all sessions for 30 and 70, sessions above the 200-day for the dip to 40.
| Fund (from) | First close under 30 | First close over 70 | Under 40, above 200-day | Any session |
|---|---|---|---|---|
| SPY (Jan 1994) | 27 of 40, 67.50% | 56 of 87, 64.37% | 53 of 74, 71.62% | 63.95% (uptrend 65.99%) |
| QQQ (Dec 1999) | 23 of 36, 63.89% | 51 of 80, 63.75% | 42 of 61, 68.85% | 61.92% (uptrend 64.26%) |
| IWM (Mar 2001) | 30 of 42, 71.43% | 31 of 51, 60.78% | 34 of 52, 65.38% | 59.58% (uptrend 57.55%) |
| GLD (Sep 2005) | 23 of 33, 69.70% | 37 of 60, 61.67% | 27 of 41, 65.85% | 55.65% (uptrend 55.64%) |
| TLT (May 2003) | 19 of 39, 48.72% | 17 of 38, 44.74% | 14 of 35, 40.00% | 49.77% (uptrend 48.87%) |
Every one of the fifteen results is within chance of its base rate. The smallest p values, using the normal approximation to the binomial, were 0.10 for GLD under 30 and 0.12 for IWM under 30; with fifteen tests, one or two near 0.1 is what luck alone tends to produce. The dip to 40 came out above its base in four funds and below in TLT, which fits a small effect or none, and samples of 35 to 74 cannot tell the two apart.
The overbought line looked even less like a warning. After SPY’s 87 spaced crossings above 70 it was higher 20 sessions later 64.37% of the time, against 63.95% for any session. Stays above 70 were short: 179 separate stretches, a median of 2 sessions, 73 of them a single day and the longest 18.
Settings, the 50 line and divergence
Length decides how often the lines are touched. The same SPY sessions, scored at different lengths:
| RSI length | Sessions over 70 | Sessions under 30 | Either |
|---|---|---|---|
| 2 | 41.83% | 27.41% | 69.24% |
| 5 | 26.72% | 11.75% | 38.47% |
| 9 | 14.27% | 4.60% | 18.87% |
| 14 | 7.10% | 1.51% | 8.62% |
| 21 | 2.67% | 0.40% | 3.07% |
| 25 | 1.48% | 0.28% | 1.75% |
At 2, SPY was “overbought or oversold” on more than two sessions in three, so the words stop meaning extreme; that is why the RSI(2) strategy uses far tighter lines. At 25 the lines almost never come. A good setting is one whose lines are rare enough to mean something on your timeframe, and 14 is still the reference.
The 50 line needs no threshold to argue about. Above it, average gains are larger than average losses; below it, the reverse. Used that way RSI is a filter on direction rather than a signal.
Divergence is price making a higher high while RSI makes a lower one. It says the second push was built from a weaker mix of gains and losses than the first, which is true and worth noticing, but a move can slow, pause and carry on.
The original data
RSI(14) here is computed with the formula above on daily closes from Yahoo Finance, downloaded 1 November 2026, prices as a chart shows them (not adjusted for dividends). SPY’s file starts on 29 January 1993 and its first RSI is on 19 February 1993; statistics run from 3 January 1994 to 30 October 2026, 8,263 sessions, so the seed has long worn off and a 200-day average exists on every one. The other funds start where Yahoo’s history starts.
How often SPY reached the lines. RSI closed above 70 on 587 sessions (7.10%) and below 30 on 125 (1.51%). The highest reading was 87.03 on 26 January 2018 and the lowest 16.70 on 8 August 2011. In the 250 sessions from 3 November 2025, 21 closed above 70 and 3 below 30, with a high of 78.54 on 14 May 2026 and a low of 27.30 on 30 March 2026; SPY was above its 200-day average on 237 of them.
The ten-point ranges behind the trend chart. On sessions above the 200-day average, SPY’s RSI(14) fell in the ranges from 20-30 up to 80-90 on 0.29%, 4.47%, 18.27%, 33.22%, 34.33%, 9.03% and 0.39% of days; on sessions below it, in the ranges from 10-20 up to 70-80 on 0.43%, 4.73%, 24.43%, 37.13%, 26.65%, 6.42% and 0.19%. The chart labels these to the nearest whole percent.
The base rates. SPY closed higher 20 sessions later on 5,271 of 8,243 sessions with a close that far ahead (63.95%, median change +1.31%); counting every crossing instead of spaced ones gives 49 of 67 under 30 (73.13%) and 123 of 179 over 70 (68.72%), still within chance (p = 0.12 and 0.18).
A second price feed agrees. Recomputing RSI from Nasdaq’s own SPY closes for the 2,471 sessions from 3 January 2017 gives the same 22 crossings under 30 and the same 64 over 70 on the same dates. The two feeds’ closes differ by more than a cent on 2 sessions (the largest $1.42), which moves RSI by at most 1.37 points and changes no crossing. Every session, crossing and band is in the daily SPY file, the crossing list for all five funds and the band table.
In the search study behind this site, 594 of 24,971 videos carry RSI or relative strength index in the title, with a median of 5,377 views, and 77% never pass 50,000. The other oscillator people reach for, Williams %R, spends far longer past its own extreme line, a gap measured in RSI vs Williams %R.
When it fails
Selling the first overbought reading
On SPY the 20 sessions after a cross above 70 looked like any other 20 sessions. A strong market is made of up closes, and a rule that sells the first high reading sells into exactly the closes RSI is counting.
Waiting for 30 in a rising market
Above the 200-day average, SPY’s RSI went under 30 on 18 sessions in almost 33 years. A plan that needs 30 to buy will usually sit in cash through the whole advance.
Buying 40 because a guide called it support
The dip to 40 in an uptrend came out ahead of its base rate in four of five funds, but never by more than chance allows. Treat the reading as a description of a pullback, then let the chart say where it fails.
Reading the panel instead of the chart
The same reading at a major level and in the middle of nowhere means two different things, and the panel cannot tell them apart. A moving average has the same blind spot.
Questions people ask
What is a good RSI number?
There is no good or bad number, only a description: above 50 means recent gains have outweighed losses. With SPY above its 200-day average, readings from 40 to 90 covered 95.24% of sessions.
Is RSI below 30 a buy signal?
Not on the evidence here. Across five funds, a first close under 30 was followed by a 20-session result within chance of any session; for SPY, 67.50% higher against 63.95%.
Is an RSI of 40 good to buy?
In an uptrend 40 is a normal pullback reading. On SPY, 71.62% of first dips to 40 above the 200-day were higher 20 sessions later, against 65.99% for any uptrend session, a gap within chance.
Related
Moving averages share RSI’s character: a smoothed summary of recent price that describes rather than predicts, and the lag is easier to see there.
Market structure gives what RSI cannot, a specific price at which the read is wrong.
How to use the RSI walks through reading the panel in a range and in a trend, one step at a time.
And risk management is the gap between an indicator reading and a position size, a gap no oscillator closes.
Treat this the same way as a moving average, as context rather than a trigger. A high reading tells you the recent bars have been mostly up, which you can see by looking at them. What it never tells you is where to get in or where you are wrong, and those are the only two decisions that cost money.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money. Some links on this page earn a commission if you buy through them. It costs you nothing and it does not decide what appears here or in what order — how these pages are made is set out in our methodology.