The trade sounds mechanical: RSI crosses above 70, so the asset is overbought, so it must fall — sell. Some version of this rule has probably lost more beginner money than any single bad indicator reading, because it quietly replaces one question (“what is the momentum?”) with another it cannot answer (“when does the trend end?”).
RSI is a useful gauge with a narrow job description. The trouble starts the moment “overbought” is read as “sell now” — and the trouble is structural, not a matter of settings.
RSI is a momentum oscillator (0–100) comparing the strength of recent up-moves against down-moves over a lookback window — 14 bars by default. “Overbought” (above 70) and “oversold” (below 30) describe stretch relative to recent movement, not a coming reversal. In strong trends, RSI can live above 70 for weeks; fade-trading every overbought reading means repeatedly selling a rising market. The readings are most useful inside ranges at the edges, as pullback filters aligned with the trend, and as one observation among several — never as a standalone reversal signal.
What the number actually measures
RSI compares the average size of recent up-closes with the average size of recent down-closes over a lookback window — 14 bars in the default setting — and maps the ratio onto a 0–100 scale. High readings mean recent gains have dominated recent losses; low readings mean the opposite.
Read carefully, that is a description of momentum right now — how one-sided recent trading has been. It is not a forecast. The window is also part of the answer: the same market will show different RSI values at 7, 14, and 21 bars, because each setting samples a different slice of history. A reading is always “RSI(14) says…”, never “the market says…”
Why “overbought means sell” fails
The 70/30 bands come from the observation that readings beyond them are statistically stretched — moves that strong relative to the recent window. Stretch describes intensity, not direction. A market can be intensely bought for weeks when the reason for buying keeps being confirmed.
This is the trap’s mechanics: in a strong uptrend, momentum stays one-sided, so RSI pins in the upper region and every pullback is shallow. A trader selling each overbought reading is systematically selling the strongest part of the move — sometimes correctly timed for a small dip, cumulatively run over by the trend. “Overbought can stay overbought” is not a caveat; it is the standard behavior in the exact conditions where the reading appears most.
The mirror image ruins bottoms: in a genuine downtrend, oversold readings repeat, and each one tempts an early catch of a falling knife.
| Day | Close | Change | 5-day RSI |
|---|---|---|---|
| Day 0 | $10.00 | — | — |
| Day 1 | $10.25 | +0.25 | — |
| Day 2 | $10.45 | +0.20 | — |
| Day 3 | $10.70 | +0.25 | — |
| Day 4 | $10.60 | −0.10 | — |
| Day 5 | $10.90 | +0.30 | 90.9 (first enters overbought) |
| Day 6 | $11.15 | +0.25 | 90.9 |
| Day 7 | $11.05 | −0.10 | 80.0 |
| Day 8 | $11.35 | +0.30 | 81.0 |
| Day 9 | $11.60 | +0.25 | 91.7 |
| Day 10 | $11.90 | +0.30 | 91.7 |
This example uses a 5-day RSI for a hand-computable walkthrough (the standard default is 14 days; the logic is identical, just over a longer lookback). From Day 5's first overbought reading through Day 10, RSI stayed above 70 on all six readings while price kept climbing from $10.90 to $11.90 — a gain of about 9.2%.
Acting on 'overbought, sell' the moment RSI first hit 90.9 on Day 5 would have meant missing the next five days of gains — price ran from $10.90 to $11.90, about 9.2%, while RSI never dropped back below 70 the entire time. That is what 'selling the strongest part of the move' looks like in numbers.
Where the readings are genuinely useful
At the edges of a range. When price is oscillating between defined horizontal levels and the trend is absent, overbought near resistance and oversold near support describe real fading opportunities — the environment where mean-reversion logic has its best odds.
As a pullback filter with the trend. In an uptrend, RSI dipping back toward the 40–50 area and turning up again often coincides with the pullback ending — a timing observation in the direction the trend already established, not a reversal call against it.
As one observation among several. A divergence — price making a new high while RSI makes a lower one — is worth logging: momentum is fading relative to price. It weakens the trend’s claim; it does not end it. Divergences can stack two or three deep before anything happens, which is why they are evidence to be weighed, never a trigger on their own.
| Reading | Common use | Honest assessment |
|---|---|---|
| RSI > 70 in a strong uptrend | Sell — it must fall | Fails repeatedly; momentum is the point, not a warning |
| RSI < 30 in a strong downtrend | Buy the bottom | Same failure mirrored; downtrends stay oversold |
| RSI > 70 at range resistance | Fade toward range mid | Reasonable — edges of ranges are its home turf |
| RSI resets near 40–50 in an uptrend | Time a pullback entry with the trend | Its most reliable trading role |
| Bearish divergence at a major level | Log it, tighten stops | Weakens the trend claim; does not end it |
“It’s overbought, so it has to come down.” Replace it with: “it’s overbought — momentum has been one-sided; in a range that matters, in a trend it doesn’t mean what I want it to mean.”
- I know my RSI’s lookback setting — and that a different setting would show a different number.
- I have classified the market state first: trend or range.
- In trends, I use RSI with the direction (pullback timing), not against it (fading).
- In ranges, I combine edge readings with the horizontal levels that define the range.
- Divergences are logged as observations, not executed as signals.
Frequently Asked Questions
Should I change RSI from the default 14?
You can — shorter periods react faster and false-fire more; longer periods smooth out. But no setting fixes the core misuse: a momentum description is not a reversal forecast at any parameter.
Why did the “70 means overbought” rule work on some charts and not others?
Because it works in ranges and fails in trends — and different charts, or the same chart in different months, are in different states. The state decides the rule’s validity, not the indicator.
Are RSI divergences useless then?
No — they are useful observations that momentum is fading relative to price. Treat them the way you treat any weakening: reduce conviction, tighten stops, wait for structure to confirm. Just do not let a single divergence be the whole trade.
