The Relative Strength Index was published by J. Welles Wilder Jr. in 1978. Almost every charting platform ships it as a default. Almost every retail trader has been taught the same one-line rule: buy below 30, sell above 70. And almost every backtest of that rule on real markets shows it loses money.
That's not because RSI is broken. It's because the one-line rule throws away most of the information the indicator carries. This article walks through what RSI actually computes, what its readings mean in different market regimes, and the divergence pattern that does the heavy lifting in confluence engines like Stryqe's.
The calculation
RSI is a normalised momentum oscillator that compares the average size of up-moves to the average size of down-moves over a fixed lookback. The default lookback is 14 candles. The output is bounded between 0 and 100.
RSI = 100 − (100 / (1 + RS))
If every candle in the lookback closed up, average loss is 0, RS is infinite, and RSI is exactly 100. If every candle closed down, RS is 0 and RSI is exactly 0. Most of the time the value drifts somewhere between 30 and 70, with occasional excursions to the extremes.
One subtlety: Wilder's original method used an exponential smoothing of the gains and losses, similar to an EMA but with a different smoothing constant. Modern implementations sometimes use a simple moving average of gains and losses instead. The two methods produce slightly different readings — within a few points typically — and either is acceptable. Stryqe's engine uses the simple average variant for computational efficiency on rolling 2000-candle windows; the difference vs Wilder's smoothing is negligible at typical scales.
What "overbought" actually means
The first thing to internalise: RSI does not measure how high a price is. It measures how aggressive the recent up-moves have been relative to the recent down-moves. A coin can be "overbought" (RSI > 70) at any absolute price level — you can be overbought at $20, then again at $25, then again at $35 in the same uptrend. That's exactly what happens in a trending market.
This is the source of the failure of the simple 30/70 rule. Selling RSI > 70 in an uptrend is selling strength. Buying RSI < 30 in a downtrend is catching a falling knife. Wilder himself warned about this — his 1978 book includes a chapter explaining that RSI levels mean different things in different trend regimes — but the warning is routinely ignored in retail education.
In an uptrend, RSI typically oscillates between 40 and 80. In a downtrend, it oscillates between 20 and 60. The "neutral" centre line shifts based on trend, and dips below 40 in an uptrend (or rallies above 60 in a downtrend) are often the higher-probability tradable signals.
How Stryqe handles the trend-dependence
Stryqe's RSI signal is not a simple level check. The engine computes RSI(14) on the same candle data, then applies trend-aware thresholds rather than the canonical 30/70:
| Trend regime | Oversold cutoff | Overbought cutoff |
|---|---|---|
| Above 200-EMA (uptrend) | 30 | 75 |
| Below 200-EMA (downtrend) | 20 | 65 |
In an uptrend, the engine waits for genuine washouts (RSI < 30) before flagging oversold, and is generous with what it considers overbought (75). In a downtrend, it raises the bar for "oversold" to 20 — anything above that is just a normal pullback in a downtrend — and lowers the "overbought" trigger to 65, because a 65 reading in a downtrend is structurally extreme.
Even with these adjusted thresholds, the engine treats RSI level alone as one of the lowest-weighted signals (default weight 6 out of 100) — explicitly because RSI levels in isolation are unreliable. The real value comes from the divergence layer.
RSI divergence — the part that works
Divergence is the disagreement between price direction and RSI direction. It comes in two forms:
- Bullish divergence: price makes a lower low, but RSI makes a higher low. The new price low is being made on weaker downside momentum than the previous low.
- Bearish divergence: price makes a higher high, but RSI makes a lower high. The new price high is being made on weaker upside momentum than the previous high.
The interpretation is causal: if a new price extreme is being made on declining momentum, the trend that created the extreme is fatiguing. Divergence is a leading indicator of trend exhaustion in a way that simple level readings are not. Empirically, divergence-based signals show meaningfully higher hit rates than pure level signals on the same data.
How Stryqe detects divergence
The engine looks at the most recent 30 candles and compares the lowest low (and highest high) in the recent 15-candle window against the lowest low (highest high) in the older 15-candle window. The exact rule:
bearish = (recent_high.price > older_high.price) AND (recent_high.rsi < older_high.rsi − 3)
The +3 RSI delta requirement is important. Without it, tiny noise-level RSI differences would trigger divergence flags constantly. Requiring at least a 3-point RSI move in the opposite direction filters out coincidental noise and keeps only structurally meaningful divergence.
The bullish divergence flag is also gated by the 200-EMA filter — bullish divergence in a downtrend is not treated as a buy signal, because divergence-based reversals against a strong downtrend have historically failed at high rates. Bearish divergence has no such filter, because in a strong uptrend, bearish divergence is a leading indicator of a top that does occasionally arrive.
Common RSI mistakes
1. Trading the level on every dip
RSI dropping to 35 in an uptrend is normal — that's the lower end of the trend's typical range. Treating every drop below 50 as "oversold" produces a flood of fake signals. The fix is the trend-aware threshold: only treat < 30 (or < 20 in downtrends) as a real oversold reading.
2. Ignoring the time-frame
RSI(14) on a 1-minute chart and RSI(14) on a 4-hour chart are different animals. A 1-minute RSI of 28 might persist for two minutes; a 4-hour RSI of 28 might persist for two days. Stryqe operates on hourly candles for crypto signals, which is a balance between noise (1-minute) and lag (daily). Choose your timeframe deliberately and stick to it.
3. Using RSI alone
This is the cardinal sin. Even with trend-aware thresholds and divergence detection, RSI in isolation is one signal of many. Confluence engines exist precisely because every individual indicator has a hit rate well below what most retail content claims. RSI's contribution is incremental — it improves a confluence score, but does not stand alone.
Backtests of single-indicator strategies (RSI levels, MACD crosses, MA crosses) on real crypto data consistently show win rates between 45 and 55%, after costs. Once you subtract friction, most of these are coin-flip strategies. Confluence — requiring multiple independent signals to agree — is what moves the needle into actually profitable territory.
4. Mistaking RSI for a price predictor
RSI is a momentum oscillator. It tells you something about the recent shape of price movement. It does not tell you where price is going. An RSI of 80 doesn't predict a top; it tells you the recent up-moves have been aggressive, and that aggressive up-moves are sometimes followed by mean reversion and sometimes by continuation. The next candle's outcome depends on factors outside the RSI calculation entirely.
When RSI is most useful
Three contexts where RSI carries real information:
- Genuine extremes after capitulation events. RSI < 20 (or > 85) in any regime is rare and usually marks a short-term reversal point. These are the readings worth attending to.
- Divergence at a structurally important level. Bearish divergence as price retests a major resistance is a higher-conviction signal than divergence in mid-range.
- Confirming a trend change. When RSI breaks out of its prior range (above 70 for the first time in a downtrend, below 30 for the first time in an uptrend), that's structurally meaningful.
The bottom line
RSI is a decent indicator embedded in a terrible cultural rule. The 30/70 buy/sell heuristic was never how Wilder intended the indicator to be used, and it consistently underperforms even simple trend-aware variants. The two upgrades — trend-aware thresholds and divergence detection — are not advanced techniques. They are the minimum needed to make RSI useful, and they're cheap to compute.
Even with those upgrades, RSI is one signal among many. The point of a confluence engine is that no single indicator is reliable on its own, and the proper question to ask is never "what does RSI say?" but "what do RSI, MACD, Bollinger, volume, and trend say together?"