Hidden and regular divergence: four variants step by step

Regular and hidden divergence compare two selected price points with corresponding oscillator readings. Regular geometry is commonly studied as a reversal hypothesis; hidden geometry as a continuation hypothesis in a defined context. Neither label includes a probability of success. The worked example below provides a complete RSI calculation and its recognition time rather than a claimed historical trade.

Two flavours of divergence — and one fundamental question

Interpretations and terminology vary between authors. RSI discussions also describe opposite price/oscillator geometry as Cardwell positive or negative reversals. Classify the selected readings explicitly before interpreting them; a book attribution does not prove that one variant has a universal statistical advantage or that professional traders all use the same rule.

For bullish variants compare lows; for bearish variants compare highs. “New” here means higher or lower than the selected preceding point, not a record for the whole instrument history or every wider move. Equal extrema are excluded in our strict comparison; allowing ties or a tolerance would be a separate rule. RSI crossing 30 or 70 is not what determines these labels.

The four divergence types and what each one means

Each of the four combinations has a geometric definition. A future reversal, continuation or net trade result is a different observation that needs a defined horizon and method. The names do not supply a success rate. A numerical comparison needs a dataset, a precise outcome definition and complete rules after costs.

Keep the exact two indices and their readings in the record. A later point can create a new comparison without changing the old one, as the diagram shows. “Bullish” does not mean RSI must rise on every subsequent bar, and opposite geometry does not require opposite future outcomes.

Hidden bullish geometry: a complete RSI example

Use hypothetical closed prices in abstract units at bars 0–9: 100, 110, 105, 100, 110, 120, 115, 104, 106, 108. These are invented teaching inputs, not a historical trade or efficacy study. Calculate RSI(3) from these closes: positive change contributes gain, the absolute negative change contributes loss, and the other component is zero.

Seed each component with its average over the first three changes: both are 10/3. Then update each average as (2 × previous average + current component)/3. For positive loss use RSI = 100 − 100/(1 + average gain/average loss). At bar 3, gain/loss = 3.333333/3.333333 and RSI = 50.0000. At bar 7 they are about 3.127572/5.436214 and RSI ≈ 36.5209. This states the initialization and recurrence; it does not infer instantaneous intentions or volume from RSI.

The close lows rise 100 → 104, a 4-unit or 4% increase, while RSI falls about 13.4791 scale points. Under our exact-price-pivot convention this is hidden bullish geometry, even though both RSI readings exceed 30. The detector confirms bar 3 after bar 5 and bar 7 only after bar 9 closes at 108. An entry supposedly known at 104 on bar 7 cannot use its later confirmation. Geometry does not show fewer buyers, weaker demand or lighter volume.

Geometry and a trend filter are separate records

Two lows compare their relative heights, not the whole trend. A continuation hypothesis must say which move would continue and how its context is recognized. Hidden geometry can also occur within a lateral range; the label does not create a trend there. There is no compulsory H4/D1 pairing or statistical priority implied by the word hidden.

In a separate hypothetical comparison of already recognized points, fix price lows 100 → 104 and supplied RSI readings 50 → 40. Both variants have hidden bullish geometry. Our own context filter additionally requires the last two recognized highs and lows to rise. Highs 115 → 120 pass alongside those lows; highs 125 → 120 fail. Do not change the divergence label merely because the filter rejects the observation. These RSI states are supplied separately, not recomputed from the earlier ten-close sequence.

Embedding divergence in the structure of a trend

Store geometry and context in separate fields. Then a comparison can test the contribution of a particular trend condition without deleting cases that did not fit its intended interpretation. A market-phase label is another model to define, not proof that a type of divergence always works best at that stage.

• Selected points. Record price and oscillator indices, values, formula and parameters.
• Availability. Record the last required bar and whether observations were closed.
• Context. Preserve the defined trend or phase condition separately from geometry and a later outcome.

Choosing only one type for a test is a declared strategy restriction, not automatically a handicap. A comparison of regular and hidden rules needs the same stated instrument, timeframe, horizon, data and costs. Neither an attractive selected example nor mastery of terminology establishes an edge.

The same label does not determine a trade result

The geometry, a later condition and a supplied entry/target distance are different objects. An oscillator reading cannot prove that a pullback is only temporary, that a crowd sold or that a named trader achieved the target. Record all subsequent outcomes under the same horizon instead of present a selected invented success as evidence.

Practical takeaway: Regular and hidden divergence compare different pairs of price and oscillator extremes; neither label proves a higher hit rate.

Errors in classification and timing

These checks address reproducibility. Write down the exact data, point rule and recognition time before interpreting a pattern.

• Using an absolute RSI threshold as the label. Classification depends on point relations, not simply a 30/70 crossing.
• Moving an observation after the fact. Exact-price-bar pairing and independent oscillator pivots are different methods.
• Using confirmation too early. A marker drawn on bar 7 after bar 9 closes was not known at bar 7.
• Changing the input without recalculating. Replacing a price can change RSI at that bar and later values. A stored old reading is no longer valid input for the changed model.

Summary and what to do tomorrow morning

In the complete ten-close example, replace bar 7’s close 104 with 99 and recompute the components. RSI at bar 7 becomes about 30.5712, instead of 36.5209. Price 100 → 99 and RSI 50 → 30.5712 both fall, so this pairing does not have regular bullish geometry. Check the low detector again too. Do not keep the old RSI reading after changing its input.

Regular and hidden divergence describe different relationships between price and oscillator pivots. Neither has a universal success rate or an established reliability ranking here. Compare fully specified rules on the same instrument, timeframe and sample, including confirmation, exits and costs, before claiming that either offers a statistical advantage.

Recalculate the complete series, record each pivot’s recognition time and reproduce the separate trend-filter comparison. If testing a strategy, define entries, exits, fills, costs and evaluation before looking forward and use unused data for a fixed rule. A candle condition, ATR-based buffer or stop-to-entry move does not guarantee continuation, a maximum loss or a cost-free exit.

Related reading: RSI and MACD Divergence — Regular, Hidden and What to Do With It — a numerical comparison of selected RSI and MACD components; RSI — how to read and when it fails — fundamentals of the oscillator behind most divergence strategies; Japanese candlesticks — basics — the OHLC inputs and limits of candle interpretation.