What Is RSI? The Gauge Everyone Reads Backwards

What the number is actually made of

J. Welles Wilder Jr. published RSI in 1978, in a book called New Concepts in Technical Trading Systems, alongside three other indicators that also went on to outlive him. The arithmetic is less intimidating than the name. Take the last 14 bars. Add up how much price rose on the up days and average it. Do the same for the down days. Divide the first by the second, and you have a ratio of recent buying pressure to recent selling pressure. Then squash that ratio onto a 0-to-100 scale, so it never runs off the top of the chart.

That squashing step is the whole reason RSI feels so readable. A raw ratio is unbounded and awkward — gains running nine times losses is a big number, and gains running eleven times losses is a bigger one, and neither tells you much at a glance. Bounded between 0 and 100, it becomes a gauge. Gauges invite you to read them like fuel gauges. This is where the trouble starts.

Because notice what is not in the calculation: the price. Not the absolute price, not the price a year ago, not earnings, not any notion of value. RSI cannot see any of it. It sees 14 bars of movement and asks one question — how lopsided was that? A share at $12 and a share at $1,200 produce the same reading if their last 14 bars had the same shape.

70 and 30 are a suggestion, not a law

Wilder proposed 70 and 30 as the thresholds, and roughly half a century later they are still the defaults on every charting platform on earth. It is worth being clear about what that means: they are one man's suggested settings from 1978, not a property of markets. There is nothing in the mathematics that makes 70 special. It is a line someone drew, everyone copied, and almost nobody has interrogated since.

And Wilder did mean them as reversal signals. This is worth saying plainly, because a lot of modern writing quietly implies he knew better and the rest of us misread him. He did not; the secondary record has him treating a reading above the upper band as a sign that a reaction or reversal was imminent. The trend-aware reinterpretation of RSI came later, from analysts like Constance Brown and Andrew Cardwell, who spent the next two decades pointing out that in a real trend the thing simply refuses to come down.

So the standard reading is not a folk corruption of the original. It is the original. It is just also, on the evidence of the last fifty years of price data, the wrong way round.

The pin

Here is the behaviour that breaks the fuel-gauge intuition. In a sustained uptrend, RSI does not touch 70, ring a bell and retreat. It goes above 70 and stays there — for days, sometimes weeks — because the ingredient it measures, one-sided upward movement, is exactly what a sustained uptrend is made of. The gauge is not malfunctioning. It is reporting, accurately, that buyers have been winning nearly every session. That is what a trend is.

On the chart below the reading sits above 70 for 30 consecutive bars. Price gains 17.7% across that stretch. Anyone who sold the first time the gauge went "overbought" watched the entire move from the sidelines, and had the indicator confirm they were right to sell on every single one of those 30 bars.

This is not an unlucky chart. It is a structural property of a bounded momentum gauge. The indicator is compressed into a fixed range, so once movement is strongly one-sided the reading saturates near the top and further strength cannot register anywhere else. Compare the range section of the same series, where the longest run above 70 is 3 bars. Same instrument, same formula, same threshold — 3 bars versus 30. The difference is not the indicator. It is what the market was doing underneath it.

Run the experiment yourself

Below is the rule everyone is taught, wired up so you can fire it. Sell every time the gauge crosses above 70, and see what price did next. Then flip the regime and run the identical rule again.

The contrast is instructive, and it is also the trap. In a range the rule looks superb — price oscillates between two levels, so a signal near the top is usually followed by a move down, and the indicator appears to have called it. That experience is where the folklore comes from. Nearly everyone who swears by overbought and oversold learned it on charts that happened to be ranging.

The problem is that you only know which regime you were in afterwards. A range is a trend that has not started yet, and a trend is a range that broke. Reading the gauge tells you nothing about which one you are standing in — and, as we are about to see, the published research is a good deal less kind to the range case than the widget is.

What the research actually found

The honest version of this section is more interesting than the one you usually get, which is either that technical analysis is astrology or that RSI is a licence to print money. Neither survives contact with the literature.

Start with the one that cuts against the sceptics. Chong and Ng tested RSI rules on sixty years of the London FT30 and found they generated returns above buy-and-hold in most cases. That is a real, peer-reviewed result and it should not be waved away.

But the same authors went back six years later with five indices and twenty-seven years of daily data, and the picture broke apart in a specific and revealing way. Across fifteen tests, not one buy-minus-sell spread was significantly positive. The leg that consistently failed was the sell leg — the overbought one. On the German DAX, after RSI came back down through 70, the index went on to rise a further 1.05% over the following ten days, a result significant at the 5% level. Selling overbought was not merely unhelpful. It was a statistically significant way to lose money.

The mechanism had already been named. In 2003, testing technical rules on Singapore data, Wong, Manzur and Chew wrote that RSI is a counter-trend indicator and that if used in a trending market it often becomes entrenched near one end of the range for days, or even weeks, giving false indications of a market top or bottom. Their own study is quietly damning on the point: they tested four RSI rules, published the results for the one based on the 50 centreline, and dropped the three overbought/oversold variants with the note that the results were mixed.

Then in 2020 Muruganandan did the test this article is really about, splitting eighteen years of Indian market data into bull, bear and sideways cycles and running RSI through each. In the bull phases the indicator fired three to four times more sell signals than buy signals — the pin, counted — and those sell signals lost money in three of the four. Nothing in his results reaches statistical significance, and he is careful to say so; his own summary is that RSI failed to deliver a positive return even before transaction costs.

One more finding is worth the space, because it is faintly comic. Anderson and Li tested the 30/70 threshold on eleven years of dollar-Swiss franc and found it lost money — while 20/80, 25/75, 35/65 and 40/60 all made money. Their proposed explanation is that 30/70 fails precisely because it is famous and everybody uses it. Treat that as a raised eyebrow rather than a finding: it is one currency pair, no transaction costs, no significance testing, and the winning thresholds were chosen by searching the same data they were then measured on — which is the exact error that surveys of this literature spend their time warning about.

So what is it good for?

Reading the section above, you would be forgiven for concluding the indicator is worthless. That is not where the evidence lands either. What consistently comes out of these studies is narrower and more useful: the extremes are the weak part, and the 50 centreline — RSI used as a trend gauge rather than a reversal gauge — is the version that keeps surviving the tests. The Singapore study's centreline results were strongly significant where its overbought results were not.

That reframing is the practical takeaway. Above 50, gains have been outweighing losses; below 50, the reverse. Read that way RSI stops pretending to time turns and starts describing which side is winning, which is the question it can actually answer. A reading of 78 is not a sell signal. It is the indicator telling you, with some emphasis, that this thing is going up.

It is also a legitimate confirmation tool. If your reason for being in a trade comes from structure — a level, a trend, a pattern — then a momentum gauge agreeing with you is mild corroboration, and a momentum gauge flatly disagreeing is worth a second look at your thesis. What it is not is a reason on its own.

And one honest caveat about transaction costs, which the papers keep raising and popular write-ups keep skipping. Several of these studies exclude commissions and note that including them would erase the edge entirely. The Singapore paper points out that local retail commission was around 1% against per-signal returns of roughly 0.2%. An edge that vanishes at the first brokerage fee was never an edge.

Divergence, honestly

No article about RSI is complete without divergence, so here it is, along with an admission you rarely see attached to it. Divergence is when price makes a higher high but RSI does not, which is read as the move running out of fuel — the new peak in price achieved with less one-sided buying than the last one. Bearish divergence at a top, bullish divergence at a bottom.

The idea is coherent and it follows directly from what RSI measures. What it does not have is evidence. Searching the academic literature for empirical tests of RSI divergence as a reversal predictor turns up almost nothing of quality — one recent conference paper describing itself as preliminary, on intraday currency data, with no significance testing. That is the entire haul. Everything else is practitioner assertion repeated until it acquired the texture of fact.

This is not the same as saying divergence does not work. It is saying nobody has shown that it does, which is a different and more uncomfortable claim, and one worth holding in mind whenever you see a chart annotated after the event with the three divergences that preceded a crash and none of the eleven that preceded nothing.

The same idea wearing three costumes

RSI is not alone, and once you see what it is doing the family resemblance is hard to miss. MACD takes the gap between two moving averages — one fast, one slow — and reads momentum off how that gap is changing. The stochastic oscillator asks where today's close sits inside the recent trading range, from the bottom of it to the top. Different arithmetic, same project: compress recent price behaviour into one readable number.

Which means they inherit the same blind spot. Below, all three run over the same stretch of the same series, and all three saturate at the same moment for the same reason — because the market underneath them was moving decisively in one direction, which is the one condition every momentum gauge handles badly.

That last point is the practical one. A chart carrying RSI, MACD and stochastic looks thoroughly researched and is mostly redundant — three views of the same underlying quantity, agreeing with each other because they are computed from the same thing, and giving a false impression of independent confirmation. If you want a second opinion, ask a different question: volume, structure, the trend on a longer timeframe. Not the same question in a different font.

How to read it without fooling yourself

1. 1Read the number as speed, not altitude. High means moving up fast lately. It has never meant expensive.
2. 2Treat 50 as the line that matters. Which side is winning is a question RSI can answer; when the winner will change is not.
3. 3Check the timeframe before you check the level. A 14-bar reading on a daily chart and on a five-minute chart are different measurements with the same name.
4. 4Expect it to pin in a trend, and treat that as information about the trend rather than a countdown to its end.
5. 5Never take a signal from it alone. If your only reason is a number crossing a line someone chose in 1978, you do not have a reason.

None of that is as satisfying as a gauge that tells you when to sell. It is, unfortunately, what the indicator can support. The version everyone learned — 70 sell, 30 buy — is the one part of it the evidence consistently refuses to back, which is a slightly awkward outcome for the most quoted number in technical analysis.