RSI: Overbought, Oversold, and Divergence

The Relative Strength Index (RSI) measures the speed and balance of recent price changes on a 0–100 scale: readings near the top mean recent gains have overwhelmed losses, readings near the bottom mean the opposite. What it does not do is tell you to sell at 70 and buy at 30 — the most common way beginners misuse it. Understanding what RSI actually computes makes clear both why the simple version fails and where the indicator genuinely earns its place.

What RSI Actually Measures

RSI, developed by J. Welles Wilder in the 1970s, looks at the last N periods (14 by default) and compares the average size of up-moves to the average size of down-moves. The formula compresses that ratio into 0–100:

If every one of the last 14 candles closed higher, RSI approaches 100.
If gains and losses have been roughly equal in size, RSI sits near 50.
If losses have dominated, RSI sinks toward 0.

So RSI is a momentum gauge: it tells you how one-sided recent movement has been, over a rolling window. On a daily chart, RSI(14) summarizes about three weeks of pressure. On a 1-hour chart, it summarizes 14 hours — which is why the "same" RSI value means very different things on different timeframes, and why checking RSI on multiple timeframes gives conflicting readings that are all technically correct.

Worth internalizing: RSI is computed purely from closing prices. It contains no information that is not already in the candles — it is a summary, not a second opinion.

Why 70/30 Is Not a Signal by Itself

The convention says RSI above 70 is "overbought" and below 30 is "oversold," with the implication that price is due to reverse. Here is the problem: overbought is what strong trends look like.

When a coin breaks out and runs from $10 to $16 over two weeks, RSI will cross 70 early in the move and can stay above it for the entire rally. Selling the first touch of 70 — or worse, shorting it — means exiting or fighting the very best trends. In a genuine bull trend, daily RSI can hold above 70 for weeks; in a capitulation decline, it can pin below 30 while price halves. "Oversold" is not a floor; assets in downtrends get oversold and then more oversold.

The behavior of RSI is regime-dependent, and this is the key practical insight:

In ranges, mean reversion works. Price oscillates between support and resistance, and RSI touching 70 near range resistance or 30 near range support is a reasonable fade setup — the level, not the RSI number, is doing most of the work.
In trends, RSI's zones shift. In uptrends, pullbacks often bottom with RSI around 40–50 rather than 30, and RSI spends most of its time in the upper half. In downtrends the mirror holds: rallies stall with RSI near 50–60. Some traders explicitly use these shifted ranges as a trend filter: an RSI that can no longer reach 70 on rallies is quietly telling you the character of the market has changed.

So the honest use of 70/30 is as a condition, not a trigger: "RSI is above 70" means "this move is fast and one-sided — a poor place to chase, and only interesting to fade if structure agrees."

Divergence: RSI's Most Useful Pattern

Divergence occurs when price and RSI disagree at the extremes, and it is where RSI adds something a bare chart shows less clearly.

Bearish divergence: price makes a higher high, but RSI makes a lower high. Example: a token rallies to $4.20 with RSI hitting 78, pulls back, then pushes to a new high of $4.45 — but RSI only reaches 66. Price went higher; momentum did not. The second leg was weaker per dollar gained, which often precedes a deeper pullback.

Bullish divergence: price makes a lower low while RSI makes a higher low. A coin capitulates to $0.80 with RSI at 22, bounces, then slides to $0.76 — but RSI bottoms at 31. Selling pressure was exhausting even as price ticked lower.

Practical rules that keep divergence honest:

It is a warning, not a trigger. Divergence says the current leg is losing force; it does not say the reversal starts now. Strong trends routinely print two or three divergences before actually turning — "divergences" is often just what a maturing trend looks like on the way up.
Demand confirmation from structure. A bearish divergence becomes actionable when price then breaks a higher low or a trendline — the tools from earlier in this path. Divergence plus a structure break is a real setup; divergence alone is a note in the margin.
Compare like with like. Draw divergence between clear swing extremes on closed candles, on one timeframe. Hunting for divergences across mismatched wicks and half-formed candles produces whatever answer you wanted.
Higher timeframes carry more weight. A daily divergence that took six weeks to form means more than a 15-minute divergence that formed over lunch.

Settings, Timeframes, and Practical Use

The default RSI(14) is fine, and there is a real argument for leaving it alone: it is the setting most other participants watch, and shared reference points are part of why technical levels function. Shortening the period (e.g., RSI(7)) makes the line jumpier and produces more overbought/oversold readings, most of them noise; lengthening it (RSI(21)) smooths at the cost of even later readings. As with moving averages, tuning the parameter until past signals look great is overfitting, not research.

A sensible way to slot RSI into a process built on this path so far:

Use structure first. Identify trend vs. range, mark your levels.
Use RSI as a condition check. In a range: is price at an edge with RSI stretched? In a trend: is a pullback reaching the zone where RSI has repeatedly bottomed (say 40–50 in an uptrend)?
Watch for divergence at your levels. A bullish divergence forming exactly at major support is confluence; the same divergence in the middle of nowhere is trivia.
Let structure trigger and invalidate. Entries come from price reclaiming or breaking levels; stops live beyond structure. RSI never needs to be the reason you enter or the reason you hold a losing position ("but it's oversold").

RSI also underpins many automated strategies and screeners precisely because it is bounded and comparable across assets — a portfolio scan for "daily RSI under 30" is a common way to surface capitulation candidates for further analysis, not a buy list.

Key Takeaways

RSI compresses the ratio of recent average gains to losses into 0–100; it is a momentum summary of closing prices, not new information.
70/30 are conditions, not signals: strong uptrends live above 70 and crashes live below 30, so fading those readings blindly fights the best moves.
RSI behavior is regime-dependent — mean reversion works in ranges, while in trends the oscillation range itself shifts (pullbacks bottoming near 40–50 in uptrends).
Divergence between price extremes and RSI extremes is the indicator's best pattern, but it warns rather than triggers — require a structure break to act, and weight higher timeframes more.
Keep the default RSI(14), use it as confluence at levels you drew first, and never let "oversold" justify holding a position whose structural invalidation has already hit.