RSI — the Relative Strength Index
The answer
RSI compares the size of the average gain to the size of the average loss over the last 14 days, and turns the ratio into a number from 0 to 100. It measures how one-sided recent price movement has been. A reading of 25 does not mean the stock is cheap. It means almost all of the recent daily moves have been down, which in a downtrend is exactly what you would expect.
Why this costs you money
"RSI is below 30, so it is oversold, so it is a buy." That sentence has cost retail traders more money than any other sentence in technical analysis, and the reason is straightforward once you see the arithmetic.
RSI falls below 30 when average losses have been much larger than average gains over the last 14 days. That is the definition of a downtrend that is currently working. So the sentence, translated honestly, reads: "this stock has been falling hard and persistently, so I will buy it."
Nobody would say it that way. The indicator lets them say it a different way, and the different way sounds analytical.
In a market that is genuinely falling, RSI goes below 30 and stays there. It can sit between 20 and 35 for weeks. A trader who buys the first reading buys, then sees a 12% loss, then sees RSI go to 22, decides that is even more oversold, and buys more. The position size grows while the thesis gets weaker. This is the single most reliable way to turn a small loss into an account-ending one, and the indicator is what gives the trader permission at every step.
The mirror image costs almost as much. RSI goes above 70 in a stock that has just started a genuine multi-month advance. The trader sells, or short sells. The stock rises another 60% while RSI stays above 70, and the reading going higher makes the trader more confident they are right.
RSI does not measure value. It measures one-sidedness. In a trend, one-sidedness is the trend, and the extreme reading is confirming it.
How it works
You need the mechanism, because the mechanism is what tells you where it breaks.
For each of the last 14 days, note whether the close was higher or lower than the previous close, and by how much.
- Add up all the gains and divide by 14. That is the average gain.
- Add up all the losses, as positive numbers, and divide by 14. That is the average loss.
- Divide the average gain by the average loss. Wilder called this RS, for relative strength.
- Convert it to a 0 to 100 scale:
RSI = 100 − (100 ÷ (1 + RS)).
In practice platforms use a smoothed running average rather than a plain one, so old days fade out gradually rather than dropping off. The principle is the same.
What the numbers actually mean
| RSI reading | What it literally says | What it does not say |
|---|---|---|
| 50 | Average gains equal average losses over 14 days | Anything about direction of the next move |
| Above 70 | Gains have been about 2.3 times losses | That the stock is expensive |
| Below 30 | Losses have been about 2.3 times gains | That the stock is cheap |
| 100 | Every one of the last 14 days closed up | That a reversal is due |
Read the middle column again. RSI 30 is a statement about the last 14 days, not about the company, the valuation or the future. A stock can be at RSI 25 and be the most expensive share on the exchange.
The 3 things people do with it
1. The 70 and 30 lines. Wilder's original suggestion. Covered above. It is the most used and the least reliable of the 3.
2. The 50 line as a trend divider. RSI above 50 means average gains exceed average losses. This is a simple, honest trend filter and it is far more useful than the extremes. In a healthy uptrend RSI tends to hold above 40 on pullbacks and reach 80 or more on advances. In a downtrend it tends to fail near 60 and reach 20 on declines.
3. Divergence. Price makes a higher high and RSI makes a lower high. This says the second advance was made with less one-sided buying than the first. It is the most interesting RSI reading and also the most abused. Covered in detail below.
What it tells you, and what it does not
RSI tells you how one-sided the last 14 days have been. That is all it has, and it is more than it sounds, because your eye cannot extract it from a candle chart.
Four things it does not tell you.
It does not tell you about value. The word "overbought" invites a valuation reading and there is no valuation input in the formula.
It does not tell you the trend has ended. An extreme reading in a trend is evidence the trend is strong. This is the reverse of the textbook and it is the most important sentence on this page.
It does not have universal levels. 70 and 30 are 1978 suggestions for commodity futures. A stock in a strong uptrend may not touch 30 for 2 years. Waiting for 30 on such a stock means waiting for a change of regime, not a pullback.
It does not survive a gap intact. RSI uses close-to-close changes. A stock that gaps 20% on results records that as 1 large up day. The reading jumps. No continuous buying pressure produced it.
Divergence, honestly
RSI divergence is real and it is a warning, not a trigger. Here is the honest version.
In a strong trend, divergence appears repeatedly and most of the instances mean nothing. Price makes a higher high on less momentum, which is normal in a mature trend, because a trend's largest percentage moves usually come early. A trader who acts on every divergence in a strong trend will be short throughout a sustained advance.
Divergence becomes usable only when price confirms it — when the trend structure itself breaks, with a lower high and a lower low, after the divergence appeared. The divergence tells you where to look. Price tells you when.
The decision rule
Use RSI to measure the strength of a move, and only use its extremes as a trade signal when price is contained.
If the longer moving average is flat and price has repeatedly turned near the same 2 levels, an RSI extreme near one of those levels is usable.
If the longer moving average is sloping, treat an extreme in the direction of the slope as confirmation of trend strength and never as a reversal signal. In that case use the 40 to 50 zone instead: in an uptrend, an RSI pullback to 40 that holds is the pullback, and the extreme reading is not.
Unless the extreme reading was produced by a gap or by a circuit-locked series of days. Then it is a measurement artefact and it should be ignored entirely.
Try this now
This is the counting exercise, and it is the most valuable 5 minutes in this cluster. Do not skip it because you think you know the answer.
- Open a 1 year daily chart of a stock you actually hold or watch. Add RSI (14).
- Mark every date where RSI closed below 30. Write down how many there were. If there were none, pick a different stock — one that fell during the year.
- For each of those dates, look at the price 10 trading days later. Was it higher? Mark it worked or did not work.
- Now write down the fraction. For example: 3 worked out of 9.
- Now do the important part. Go back and look at the 200-day moving average on each of those dates. Split your list into 2 groups: readings that happened while the 200-day average was rising, and readings that happened while it was falling. Write down the fraction for each group separately.
What you should see. The combined fraction in step 4 is usually unimpressive, often close to a coin toss. The split in step 5 is where the lesson is. On most charts, oversold readings that happened while the long-term average was rising worked considerably more often than the ones that happened while it was falling.
That is the whole article in 2 numbers, on your own stock, in your own handwriting. RSI below 30 in an uptrend is a pullback. RSI below 30 in a downtrend is the downtrend.
Bonus, 60 seconds. Change the RSI period from 14 to 7 and count the sub-30 readings again. You will get many more. Change it to 21 and you will get far fewer. The signal count is a function of a setting you chose, which means the setting was never a fact about the market.
Three real cases
1. The S&P 500, September 2008 to March 2009 (United States) — oversold for 6 months Through the financial crisis decline, the S&P 500 reached deeply oversold RSI readings repeatedly and continued falling to its low on 9 March 2009. Each oversold reading was arithmetically correct: losses were overwhelming gains. Each one was also a description of a functioning downtrend. A trader buying every sub-30 reading from September onwards accumulated a position at steadily worse prices for around 6 months.
2. The NIFTY 50, 2021 (India) — overbought for most of a year The Indian market rose strongly through most of 2021, with the NIFTY 50 reaching successive record highs before a correction late in the year. Through the advance the index spent extended periods with daily RSI above 70, and RSI pullbacks generally stopped in the 40s rather than reaching 30. That behaviour — holding above 40, reaching 80 — is the signature of a strong uptrend, and it is the reason "wait for oversold" kept an entire generation of Indian retail traders out of a large advance.
3. J. Welles Wilder Jr., 1978 (United States) — where the numbers came from Wilder introduced RSI in New Concepts in Technical Trading Systems, published in 1978, alongside ATR, the Directional Movement system and Parabolic SAR. He chose a 14-period lookback and identified 70 and 30 as levels of interest. He was writing about commodity futures, calculating by hand, before personal computers existed. Those defaults are now the shipped setting on essentially every charting platform worldwide, applied to instruments and timeframes he never tested. The 14 is a convention, not a discovery, and the fact that everybody uses it is the only argument in its favour.
The question that resolves it
A novice sees RSI at 28 and asks: is this oversold?
An expert asks: is the long-term average rising or falling?
Both are looking at the same 28. On a chart with a rising 200-day average, 28 is a pullback inside an uptrend and it is one of the better entries an indicator will hand you. On a chart with a falling 200-day average, 28 is the downtrend functioning correctly, and buying it is standing in front of the move.
The indicator cannot answer the question. You have to bring the answer to it.
What would make this wrong
The claim is that RSI extremes work differently in trends and in ranges, and that sub-30 readings in a downtrend are not buy signals.
You would falsify it with your own count. Run the step 5 split above on 10 different stocks. If oversold readings in falling markets were followed by higher prices as often as those in rising markets, this article is wrong and you should believe your data.
You would falsify the "RSI measures one-sidedness, not value" claim by finding 2 stocks with identical RSI values and showing that the RSI told you something about their relative valuation. It cannot, because there is no valuation term in the formula, but seeing it on 2 real charts is more convincing than reading it.
The honest limits are 4.
First, RSI is not useless in downtrends. Bullish divergence at a final low is a real phenomenon and it does precede some reversals. The problem is that it also appears many times before the reversal, and there is no way to know from the indicator which instance is the last one.
Second, the counting exercise measures hit rate only. A strategy with a 30% hit rate can be profitable if the winners are large. Do not conclude from a low count that the tool has no use.
Third, 10 trading days is an arbitrary window. Change it to 5 or 20 and the fractions change. Try several. If the result is stable across windows it is more trustworthy.
Fourth, this article makes no claim that RSI produces profit anywhere. It claims you can know what RSI measures and where it stops being true. Those are different things and only the second is being taught.
In India
Indian platforms ship RSI at 14 with 70 and 30 lines drawn in. NSE stocks, index charts and commodity charts all use the same setting because it is the default.
Three Indian conditions change what the number means.
Circuit-locked series break the calculation. A share outside the derivatives segment carries a daily price band, commonly 2%, 5%, 10% or 20%. When a stock is locked at its upper band for several days, every daily gain is identical and capped. RSI reads extremely high. It is measuring an exchange rule, not accelerating demand. The same applies at the lower band, where a stock nobody can sell produces a very low RSI. In both cases the trader looking at RSI is looking at a number the market did not produce.
Thin volume produces false extremes. Below the most liquid few hundred NSE names, a single order can create a 5% or 6% day in a stock that usually moves 1%. That one day moves RSI a long way. Before acting on an RSI signal in a mid or small-cap Indian share, look at the volume on the days that produced the reading. If 1 or 2 days did most of the work, the signal is 1 buyer's opinion.
The whole session is visible. Indian markets trade 9:15 to 15:30 with a pre-open auction from 9:00 to 9:08 and no regular pre-market session. The daily close-to-close change RSI uses covers a continuous auction. That is a genuine advantage over US data and it means an Indian daily RSI reading is a more honest record of what actually traded.
In the United States
The same 14 and the same 70 and 30, on a much deeper market, with 2 differences that matter for the reading.
Extended-hours trading puts holes in the input. US shares trade before 9:30 and after 16:00 Eastern time, and most platforms exclude that from the standard daily candle. Most large US companies report results after the close. So the entire reaction to a results announcement typically arrives as a single close-to-close change in the RSI calculation. A stock that moved 18% overnight gives RSI one enormous up day. There was no sustained buying pressure across a session — there was a repricing while almost nobody was trading. RSI cannot tell the difference and neither can you, from the indicator alone.
No daily price bands. US individual stocks have Limit Up-Limit Down pauses rather than daily caps, so a single session can carry a very large move in full. American RSI readings reach extremes faster and unwind faster than Indian ones on comparable news, and any threshold copied from an American book will trigger more often on a US chart.
The US also offers the deepest testing ground. If you want to run the counting exercise on 200 stocks rather than 2, US price history is where the data exists and it is mostly free.
Where they differ, and what that tells you
The difference is where the distortion sits.
Indian price bands compress the input series. A move that wanted to be 30% arrives as 20%, then 20%, then whatever is left. RSI sees several large days instead of 1 enormous one, so the reading climbs steadily and stays extreme longer.
US extended-hours trading hides part of the input series. A move that wanted to be 18% arrives as 1 close-to-close jump with no visible path. RSI sees a single spike.
What that tells you is which false signal to expect in each market. In India, expect RSI to sit pinned at an extreme for a run of days on a stock that is circuit-locked, and expect that to look like an unusually powerful trend. Check for band locks before you believe it.
In the United States, expect RSI to jump across the 70 or 30 line in 1 session after results, with no accumulation of pressure behind it. Check whether the reading was produced by a gap before you believe it.
There is a wider point. RSI's whole logic depends on the idea that a series of closes reflects continuous negotiation between buyers and sellers. Both markets interrupt that negotiation, in opposite ways, and the indicator reports the interruption as if it were momentum.
Carry this
- RSI measures how one-sided the last 14 days have been. It contains no information about value.
- In a trend, an extreme reading is confirmation of strength. In a range, it is a position within the range. The 200-day average tells you which you have.
- 14, 70 and 30 are 1978 suggestions. Change the period and count the signals again. The number that changes was never a fact.