Nonzero Lab Research · Evidence · 2026-08-24 · 6 min read

RSI Overbought and Oversold: Fifty Years of Misreading

RSI rescales momentum onto a 0–100 scale; 70/30 were the author's defaults. In trends RSI pins to extremes — reading thresholds as reversals misreads it.

What does RSI actually compute?

RSI was introduced by Wilder (1978), with a default lookback of 14 periods. What it does can be said in one sentence: it takes the relative strength of up moves versus down moves over a window and compresses it onto a fixed 0–100 scale.¹

That is all it is — momentum expressed on a bounded scale. No line on that scale inherently means "about to reverse." The numbers 70 and 30 were reference lines the author drew in his book to mark "extended" regions. Reference lines, not laws.

The bounded scale has a subtler side effect: it makes extremes look symmetric and rare. The zones above 70 and below 30 each occupy a clean third of the visible range, as if marked out for "excess." But where the reading sits depends on the character of the recent market, not on any law of equilibrium — in a strong trend, living in the upper zone is the norm, not the exception.

It is worth being explicit about the division of labor. RSI answers the question "how strong have up moves been relative to down moves lately." It never answers, and cannot answer, "at what strength should the move turn." That second question belongs to market structure, not to the indicator's definition.

Where did 70 and 30 come from?

The spread of those two lines is worth pausing on. Nearly every charting package draws 70/30 as default horizontal levels, and generations of traders learned "sell above 70, buy below 30" without ever reading the original text. A rule of thumb, passed through a software default, acquired the status of a law of nature.

The pattern is the same one seen with moving averages: whatever in-sample edge simple rules show tends to decay once data mining is accounted for — the optimistic opening of Brock (1992)², the out-of-sample failure in Sullivan (1999).³ The default thresholds never even cleared the first bar; they were conventions from day one.

The inertia of defaults deserves its own sentence. Once a parameter is drawn on every chart, it stops being treated as a parameter and starts being treated as the coordinate system itself. Almost nobody asks where the coordinate system came from — and those who do ask open the same defaults. That is how a fifty-year misreading sustains itself.

Pinned to the rail in trends

Traders who treat 70 as a sell signal keep meeting the same puzzle in trending markets: RSI crosses above 70 and then stays there for weeks while price grinds higher — a strong crypto leg, a post-earnings drift in a large-cap stock. This is not the indicator failing. It is arithmetic: persistent one-way movement mechanically pushes relative strength toward the top of the scale.

Pinning at an extreme is not a siren that says "so overbought it must crash." It is another way of saying "this market is trending." Traders who read thresholds as reversal signals get punished by the same mistake on repeat in trending regimes: shorting strength too early, catching weakness too early.

The mirror image is just as real. In a genuinely ranging market, the 70/30 zones do frequently coincide with the edges of the range. The same reading carries opposite meanings in the two environments — the problem was never the indicator, it is that you never said which environment you were using it in.

Two worlds

"Overbought should fall" smuggles in a worldview: prices mean-revert, and extremes precede regression. In a trend-continuation world, extremes precede more extreme. Both worlds exist, and the same RSI print means opposite things in each.

Neither worldview is an identity. The trader who is "a mean-reversion person" will be right in one kind of month and bloodied in the other; the same holds for "a trend person." The durable skill is not allegiance to a world but bookkeeping across both — knowing which regime your current results actually came from.

So the right question is not "does price fall after RSI crosses 70." It is "in the kind of environment I trade, what tends to happen after RSI touches an extreme?" The environment is a variable, not background scenery.

The hard part is the first step: you cannot declare the regime in advance. Looking at a finished chart and saying "this was a trend" is hindsight — at the time, you did not know. What you can do is two things: freeze the rule beforehand, and afterwards split the results by regime. The first keeps the rule from drifting with your mood; the second lets the data tell you how often each world showed up.

On the profitability of technical analysis as a whole, the evidence has always been mixed: in the 95 modern studies surveyed by Park (2007), 56 reported positive results, 20 negative, and 19 mixed.⁴ Inside that ledger, expecting one default threshold line to hold up on its own mistakes the granularity of the evidence.

Turning "regime" into a variable

You cannot declare the regime in advance; you can only measure it after the fact — which is exactly what blind testing is for. With the ticker and the date hidden, each session is a stretch of market with no name and no ending. RSI touches an extreme, you act, the reveal is recorded. After enough sessions, "overbought inside a trend" and "overbought inside a range" are two separate piles of data in your hands, not one blur of intuition.

One discipline matters more than any other here: the split by regime must be defined before you look at the results. Deciding afterwards which sessions "counted as trending" is just narrative sneaking back in through the side door. Freeze the grouping rule first — a moving average, a channel, anything fixed — then let the numbers land where they land.

How to accumulate that sample, and how large it must be before it means anything, is covered in How Many Trials Do You Need?. Why reviewing finished charts cannot substitute for this process is covered in Why Your Chart Review Lies to You.