How to Set Up RSI Divergence Alerts for Crypto Trading

Divergence alerts are easy to generate and hard to make useful. Most implementations fire constantly, mostly on noise, and fire earliest on exactly the signals that will not survive to the candle close. This article covers the four settings that separate a usable divergence alert from a notification stream you will mute within a week.

• Regular divergence signals exhaustion; hidden divergence signals continuation. They are opposite trades.
• Both pivots must be confirmed, which means the signal is necessarily late — an alert that isn't late is reading an unclosed candle.
• Pivot distance matters: two swings 200 bars apart are not a divergence, they are a coincidence.
• Divergence is a warning that momentum is fading, not a reversal — it needs a trigger and a trend filter.
• An unfiltered divergence alert on 30 pairs across 4 timeframes is a notification stream, not a signal.

What is RSI divergence?

Divergence is a disagreement between price and momentum. Price reaches a new extreme; RSI does not confirm it. The interpretation is that the move is being made with less force than the previous one.

Regular divergence — exhaustion

Bullish: price makes a lower low, RSI makes a higher low. The selling that produced the new low was weaker than the selling that produced the previous one.

Bearish: price makes a higher high, RSI makes a lower high. Buying pressure is fading into the new high.

Regular divergence anticipates a potential reversal.

Hidden divergence — continuation

Bullish: price makes a higher low, RSI makes a lower low. The pullback went deeper in momentum than in price — typically a healthy dip inside an uptrend.

Bearish: price makes a lower high, RSI makes a higher high — a healthy bounce inside a downtrend.

Hidden divergence suggests the existing trend resumes. It is less discussed and, in trending markets, often the more useful of the two — precisely because it trades with the trend rather than against it.

The four settings that decide whether alerts are usable

1. Closed candles only — non-negotiable

An RSI value computed on a forming candle is provisional. A divergence visible at minute 4 of a 15-minute bar frequently does not exist at minute 15, because the remaining eleven minutes of order flow changed the reading.

Most retail platforms recompute on every tick because it looks responsive. The result is alerts that fire early on a market state that will not survive. Any alert you intend to act on must evaluate on closed bars. There is a concrete worked example in why your divergence alert fires at the worst possible time.

2. Pivot confirmation

A swing point is only a swing point once enough bars on both sides confirm it. If your detector requires two bars either side, the pivot is confirmed two bars late — and so is every divergence built on it.

That lateness is not a defect to engineer away. An alert that fires without confirmed pivots is guessing about a pivot that has not formed. The honest framing: choose your confirmation width deliberately (2–3 bars is typical), then accept the lag it buys you.

3. Maximum pivot distance

Two swings sixty bars apart connected by a line are not a meaningful relationship. Set a maximum lookback — commonly 20–60 bars depending on timeframe — beyond which two pivots are simply not compared. Without this constraint a detector will find "divergence" almost anywhere, because on enough bars some pair of pivots always diverges.

4. Minimum RSI displacement

Requiring the two RSI pivots to differ by a minimum amount — say two or three points — removes the large class of signals where RSI is essentially flat and the "divergence" is rounding noise.

RSI period and timeframe for divergence

The 14-period default works well for divergence and there is little reason to change it: shorter periods produce more pivots and therefore more spurious divergences, longer periods smooth away the pivot structure the signal depends on.

Timeframe matters more than period. Divergence on the 15-minute chart is frequent and mostly noise. On the 4-hour and daily it is rare and carries far more information. If you are running divergence alerts across many pairs, restricting to higher timeframes is the single most effective noise reduction available.

Turning an alert into a trade

Divergence tells you momentum is fading. It does not tell you the move is over, and acting on divergence alone is how traders end up short in the middle of a parabolic advance — divergence can persist through several higher highs before anything reverses.

A workable structure:

1. Trend filter. Regular divergence against a strong higher-timeframe trend is the lowest-quality version of this signal. Either require trend agreement, or restrict counter-trend divergence to exhaustion contexts.
2. Location. Divergence at a structural support or resistance level is materially better than divergence in open space. Momentum supplies timing; structure supplies the level.
3. Trigger. Wait for confirmation — a break of the swing that formed the divergence, or a candle close back through a level. The divergence is the setup, not the entry.
4. Invalidation. The stop belongs beyond the extreme that formed the divergence. If price exceeds it, the premise is gone. This also gives you the stop distance you need to derive size; see risk management.

Alert hygiene: why most people mute their alerts

Thirty pairs × four timeframes × two divergence types is 240 alert channels. Unfiltered, that produces a stream nobody reads, and the one that mattered scrolls past with the rest.

• Restrict the universe. Pairs you would actually trade, at sizes their books support.
• Restrict the timeframes. 1-hour and above for most people.
• Pre-filter by trend so you are not alerted on divergences you have already decided not to take.
• Deduplicate. A divergence that persists across several bars is one signal, not one per bar.

From alerts to automated execution

An alert still requires you to be present, awake and disciplined at the moment it fires — which reintroduces every problem automation exists to solve. The alternative is to encode the filters directly as an execution rule.

On TradeFloor, divergence is available as a trigger condition rather than only a notification, and can be combined with up to five AND/OR conditions — higher-timeframe trend agreement, a structural level, an order-flow confirmation, a news veto. Detection runs on closed candles only, with the forming bar stripped before any indicator is computed. When conditions pass, the entry and its protective bracket are placed in the same operation.

// questions or corrections · [email protected] · more essays · /blog