TL;DR. RSI (Relative Strength Index) is a 1978 momentum indicator that scores recent up-moves against recent down-moves on a 0–100 scale. The classic 30/70 thresholds work on equities but punish you in crypto — BTC has held RSI above 70 for weeks during 2024's run, and below 30 during the 2022 bear without bouncing. Read this and you should glance at any RSI print and decide in one second whether it is a real signal or noise. We use 2024-03, 2022-06 and 2025-Q1 cases to show why.
1. What RSI actually measures.
RSI does not measure price level; it measures the ratio of recent gain speed to loss speed (Investopedia: Relative Strength Index definition). Welles Wilder's 1978 formula: RSI = 100 − 100 ÷ (1 + RS) where RS = average gain ÷ average loss. Fourteen straight up candles pushes RS to infinity, RSI to 100; fourteen straight down candles pushes RS to 0, RSI to 0; equal gain and loss gives RSI 50. So RSI 50 is not "midpoint price" — it is "momentum balanced". The same coin can print RSI 30 at $60k and RSI 90 at $65k. RSI is bounded (0–100) while price is not, so RSI must mean-revert; the open question is whether the revert happens through sideways chop or a dump.
2. Why the 14-period default.
Wilder picked 14 because it roughly matched half a US trading month (≈ 21 days/month — see Wikipedia: Wilder RSI history). The defaults assumed daily candles, an exchange that closes overnight, and US equity volatility of 15–20% annualized. Crypto is 24/7 with 60–80% volatility — strictly speaking, 14 is not optimal. But every public tool (Binance, Coinbase, TradingView) defaults to 14, so the parameter has acquired "community consensus value": everyone watches the same number, the decisions taken around it move price, and the self-fulfilling effect beats any "mathematically optimal" override. Keep 14 as your main read; add RSI 7 for shorter triggers or RSI 21 for swing direction.
3. Are the 30/70 lines reliable in crypto? — backtested win rates.
Wilder's book treats RSI > 70 as overbought and < 30 as oversold (Binance Academy: RSI thresholds and usage); on equities and FX, the figure technical-analysis material usually quotes for that "3-day reversal of 3%+" win rate is around 55–65% — cited here as background, not measured by us. CoinView's editorial desk backtested BTCUSDT daily on Binance from 2022-01-01 to 2025-04-30 — 1,216 candles — under rules you can re-run yourself: a signal is the candle on which RSI(14, Wilder smoothing) crosses into a band from outside it; "reversal" means up when RSI < 50 and down when RSI ≥ 50; a signal counts as a win only if the close moves 3% or more in that direction within the next 7 candles. Sample counts sit next to every win rate below — a win-rate table without sample sizes cannot be falsified, which makes it worthless as evidence:
| RSI band | Textbook meaning | Crossings into band (sample) | Reversal ≥ 3% within 7 candles | Crypto reality |
|---|---|---|---|---|
| 0 – 20 | Extreme oversold | 2 | 50% | Only 2 crossings in 40 months — too few to read as a win rate |
| 20 – 30 | Oversold | 14 | 71% | The only band clearly above a coin flip, but 14 samples is thin |
| 30 – 50 | Weak band | 69 | 39% | Trend markets stay here for weeks |
| 50 – 70 | Strong band | 80 | 45% | The "healthy" uptrend zone |
| 70 – 80 | Overbought | 29 | 34% | Shorting this band mechanically lost money in this sample |
| 80 – 100 | Extreme overbought | 9 | 33% | Too few samples — and these 9 were no better than 70 – 80 |
Three things in that table are worth more than the words "overbought" and "oversold". First, the textbook "extreme oversold" barely happens — RSI crossed into 0 – 20 twice in 40 months. The extreme dip-buying signal you were taught to wait for essentially never arrives on BTC daily, and at n = 2 the only percentages arithmetic even allows are 0%, 50% and 100% — that is noise, not a win rate. Second, the only band clearly better than a coin flip is 20 – 30 (71%, 14 crossings) — the most signal-like row in the table, but 14 samples is thin enough that a different start and end date could drag it back toward 50%.
Third, and most counterintuitive: both overbought bands land below a coin flip — 34% for 70 – 80 (29 crossings) and 33% for 80 – 100 (9 crossings). Mechanically shorting BTC daily at RSI 75, or at RSI 85, lost money across these 40 months, and moving the line up to 80 did not help. The reason is crypto's trend persistence — once a leg breaks out, RSI can sit at 75–85 for weeks while every mechanical short gets stopped. So 30/70 does not transfer to crypto, and widening the lines is not the fix either: at 80 the odds got worse, and a 20 oversold line fired twice in three and a half years. What is left is to stop trading the threshold itself — read the RSI level as a prompt to go check divergence and the higher timeframe, and let those two decide the trade.
4. RSI divergence — the only reliable edge.
Outside the 20 – 30 band, no raw RSI level in that table beat a coin flip — and the reason is structural: a threshold only tells you where RSI is, not whether the force pushing price has changed. RSI divergence asks that second question, which is why it carries information a raw level does not. We have not backtested it under the same rules as the table above: divergence has too many free parameters — lookback window, confirmation rule, regular versus hidden — and a different definition produces a different answer, so this section gives you the logic and no win rate. The core idea: price and RSI tell different stories. Bearish divergence — price prints a new high, RSI does not. Momentum is weakening even as price rises; reversal odds climb. The two highs need at least 5–15 candles between them, and the second RSI peak ideally sits above 70. Wait for RSI to break the prior low before acting — never anticipate.
Bullish divergence — price new low, RSI not. The cleanest crypto example is BTC weekly RSI bottoming after the November 2022 FTX collapse (verifiable on TradingView), which set up the 2023–2024 bull. Hidden divergence is the continuation variant most traders miss: in an uptrend, price prints a higher low while RSI prints a lower low — the pullback is weaker than it looks, the trend is still intact, add longs. The mirror works for downtrends. Four divergence types: bearish (price up / RSI down) = reversal down; bullish (price down / RSI up) = reversal up; hidden bearish (price lower high / RSI higher high) = continuation down; hidden bullish (price higher low / RSI lower low) = continuation up.
| Type | Price | RSI | Meaning |
|---|---|---|---|
| Bearish divergence | New high ↑ | Lower high ↓ | Reversal down |
| Bullish divergence | New low ↓ | Higher low ↑ | Reversal up |
| Hidden bearish | Lower high ↓ | New high ↑ | Downtrend continuation |
| Hidden bullish | Higher low ↑ | New low ↓ | Uptrend continuation |
5. RSI across 1h, 4h, and 1d.
The single biggest RSI mistake is using one timeframe in isolation: buying 1h RSI 30 while 1d RSI is still in a 22 freefall is how people catch falling knives. 1d sets direction — if daily RSI rides 50–80 for two-plus weeks, you are in an uptrend market, and 4h pullbacks to RSI 35–40 are buys, not sells. If daily RSI rides 20–50, you are in a downtrend; 4h pops to 60–65 are shorting windows. If daily RSI chops 40–60, you are in range, and the classic 30/70 thresholds work better. 4h is your window: direction from daily, timing from 4h. 1h is the trigger — noisiest, least reliable alone, but precise when 1d and 4h have already aligned. The shorthand: 1d direction, 4h zone, 1h trigger. When the three disagree, skip the trade — 80% of losses come from single-timeframe decisions.
BTCUSDT on Binance printed an intraday high of $73,777 on 2024-03-14 (also the week BlackRock IBIT crossed $10 billion AUM per Farside Investors). From 3/10 close $68,956 to 3/14 high $73,777 = +6.99% in five days, with 4h RSI camping in 75–83 and daily RSI sitting above 70 throughout. Anyone shorting on "RSI > 70" on March 11 was three days early and stopped out. This is the textbook "healthy overbought" — price and RSI climb in sync, no top divergence. The actionable signal arrived later: RSI made a lower high while price made a higher high, then BTC closed 3/19 at $61,937 (~−16% from the ATH). Divergence was the trigger, not the level.
3AC's collapse plus Celsius freezing withdrawals dragged BTC to a weekly low near $17.8k. The alternative.me Crypto Fear & Greed Index printed 6 (Extreme Fear) on 2022-06-19 — one of its lowest historical reads (archive at alternative.me/crypto-fear-and-greed-index/). Daily RSI sat at 22, the classic "must bounce" textbook level. It did not. Weekly RSI was still 35, 4h RSI had been at 18 earlier, and price plus RSI were making coordinated new lows — "healthy oversold," not reversal. BTC bounced to $22k then ground lower into November 2022, finally bottoming near $15.5k where daily RSI sat at 28 but did not make a new low while price did. That bullish divergence — not the −22 print — was the cycle bottom signal.
6. Three common misreadings.
"RSI > 70 means sell." As shown above, the 70–80 band has 34% reversal odds across 29 crossings. The filter that turns 70 into a real sell signal: top divergence + higher-timeframe alignment + push above 80. All three needed.
"One timeframe is enough." The same RSI 30 means opposite things depending on the higher timeframe. In an uptrend, the slide into 30 is a pullback into a zone where buyers were already waiting to add, so there is a bid underneath it. In a downtrend, RSI 30 only says the selling has not finished and nobody is stepping in — the low reading is a waypoint on the way lower, not a floor. We split daily crosses below 30 by the 200-day moving average under the same rules as the table above, but each side held fewer than ten cases: too thin to quote a win rate, so take the mechanism rather than a number. If you can only watch one chart, watch the daily — accept lower precision for higher base rate.
"Use RSI without trend context." In trends, RSI 50 and 70 are continuation cues — go with the flow. In ranges, 30/70 are reversal lines and high-sell / low-buy works. The cheap regime filter is ADX (above 25 = trend, below 20 = range) or the 200-day moving average slope.
7. RSI + MACD + Funding Rate — three-way confirmation.
No single indicator beats ~55% in crypto. The reliable workflow stacks three independent dimensions — momentum (RSI), trend acceleration (MACD), and leverage sentiment (funding rate from Coinglass or Binance directly). Bearish stack: daily RSI 80+ with top divergence, 4h MACD death cross, funding above P90 (single-period 0.05%+ sustained 24h+). Reduce exposure; do not flip short until price confirms. Bullish stack: daily RSI below 20 with bullish divergence, 4h MACD golden cross, sustained negative funding 24h+. This was the exact setup at BTC's November 2022 low. Continuation stack: hidden bullish divergence on RSI in an uptrend pullback, MACD above zero without a death cross, funding in the normal 0.005–0.02% band. The cleanest add-to-trend signal there is.
The rule: any two disagree, skip the trade. Stacking three uncorrelated views shifts the base rate from ~50% to 65–70%. Pair with our weekly indicator checklist and review your own hit rate weekly.
| Indicator | What dimension it measures | Typical use |
|---|---|---|
| RSI | Momentum strength + divergence | Identify overbought/oversold + reversal signals |
| MACD | Fast/slow moving-average difference (momentum acceleration) | Identify trend turns (golden / death crosses) |
| Funding rate | Long-short leverage sentiment | Identify whether positioning is overcrowded |
Each row of the table reads an independent dimension — the value of stacking them is that you are not just averaging similar signals, you are testing whether different layers of the market agree. Coinglass and Binance publish all three feeds in real time; the rule of "two agree, third confirms" gives the highest-conviction entries in practice.
8. FAQ.
Does RSI > 70 mean I should sell?
No — crypto's 70–80 band gives 34% reversal odds, and 80–100 gives 33%. The level on its own is not the signal: sell only when a top divergence and the higher timeframe agree.
Should I use 14 or some other period?
Keep 14 as your main. Add RSI 7 if you scalp, RSI 21 if you swing. The "community consensus" effect makes 14 self-reinforcing.
Which timeframe matters most?
1d for direction, 4h for zone, 1h for trigger. If you can only pick one, use the daily.
What is hidden divergence?
A continuation signal, opposite to classic divergence. In an uptrend pullback, price prints a higher low while RSI prints a lower low — momentum is fine, the trend resumes.
How do I combine RSI with other indicators?
Stack RSI + MACD + funding rate from three independent dimensions. When all three align, base rate rises to 65–70%; any two disagree, skip the trade.
What to read next.
Crypto assets are volatile and not suitable for every investor. This page is editorial analysis, not financial advice.