Divergence trading: from swing comparison to trade rules

Regular and hidden divergence answer different questions

Read each row from the older swing to the newer swing. The hidden patterns below have the same price/oscillator relationships that TradingView describes as Cardwell positive and negative reversals for RSI. The labels describe a setup, not a measured probability that the next move will follow it.

Choose one swing-pairing method

This illustrative rulebook anchors every comparison to price pivots and samples the oscillator on those same bars. It does not independently locate oscillator pivots. Requiring separate pivots on both series is another valid research design, but it needs its own maximum alignment distance and tie-breaking rule; it will produce different signals.

• Use standard candles on one symbol and timeframe. Select one oscillator series for each experiment, such as RSI of close with length 14. Do not switch from MACD line to histogram after seeing which agrees with price.
• Define a price low as strictly lower than the lows of the three preceding and three following bars; reverse the comparison for highs. Reject ties. The pivot becomes eligible only when the third following bar closes.
• Pair only consecutive confirmed price lows, or consecutive confirmed highs, with pivot bars 5 to 60 bars apart. If that pair fails, do not search farther back for a better-looking match.
• At the two pivot bars, compare the saved price extremes and oscillator values using strict inequalities. Equal oscillator values do not qualify. The oscillator samples need not be their own local extrema.
• Freeze each qualifying pair at confirmation and identify it by its two bar timestamps. Process it once. All numbers here are research settings, not optimized defaults.

Define context, trigger, and expiry before testing

Run each of the four setup types as a separate experiment. For hidden bullish setups, this example requires the confirmation close above a 50-bar EMA and that EMA above its value five bars earlier. Reverse both conditions for hidden bearish setups. Regular setups use no EMA filter in this baseline. These are explicit context proxies, not proof of a trend or reversal.

• Check invalidation before the trigger on each bar. If the five-bar window ends without a trigger, discard the setup.
• Allow only one active setup and one position in each experiment. Ignore new candidates while either is active; do not silently replace a pending pair.
• Freeze the context decision at confirmation. Changing that decision on each later bar would be a different strategy.

Specify exits and size independently of the signal

For the illustrative baseline, place the initial stop one minimum tick below the second low for a long, or above the second high for a short. Define R as the absolute distance from the actual entry fill to that stop. Set the target at two times R from the fill in the trade direction, and keep both prices fixed. This is a price-distance definition of R before fees, not a promise about realized loss.

• Define how the stop is submitted and when the target becomes active in the implementation. If a market entry gaps through the intended stop, do not erase the fill from the test: close it at the first executable opportunity and record the gap case.
• Use a fixed quantity of one tradable unit for the baseline, with sufficient capital and margin for that instrument. This is a comparison convention, not a position-size recommendation. Record point value and minimum lot size.
• If neither exit fills, submit a market close at the close of the twentieth chart bar after the entry bar. The fill follows the selected execution model. Do not add an opposite-signal exit midway through a comparison.
• When both stop and target fall inside one bar, inspect order activation and intrabar sequencing. Document the broker-emulator assumptions and use finer historical detail where available.
• Include the applicable commission and a stated slippage scenario. Stops can fill worse than their price, so realized loss can exceed the planned distance.

A hypothetical bullish setup with an honest timeline

Suppose consecutive price lows at bars 80 and 100 are 100 and 98, while the oscillator samples are 24 and 31. The pair has regular bullish divergence under this price-anchored method. Assume the intervening high is 103, the tick size is 0.01, and no early break or invalidation occurs. These are invented numbers for explaining the rules, not observed trades.

• If price had already closed above 103 on bar 102, the example rules would reject the setup at confirmation instead of claiming an earlier trade.
• A chart may draw the divergence line back to bar 100. Keep a separate marker at bar 103 for detection and bar 104 for the entry decision.

When to skip or reject a divergence setup

A divergence can coexist with continued price movement against the proposed trade. Additional confirmation changes the entry time and price; it does not remove that possibility. Record rejected setups as well as accepted trades so the rules can be audited.

• Skip missing oscillator values, tied pivots, insufficient warm-up history, pairs outside the spacing limit, and hidden setups that fail the frozen trend filter.
• Reject stale, invalidated, or expired candidates. A later attractive move does not justify reopening the same pair.
• Do not treat several oscillators derived from the same prices as independent votes without testing that assumption. Keep a single-series baseline for comparison.
• Reject a backtest that enters on the original pivot bar while using right-side bars to recognize that pivot. Future confirmation cannot authorize a past order.
• Keep SMT divergence separate: it compares the structure of related assets rather than price and an oscillator on this one instrument.

Test whether the rulebook survives another sample

Save the symbol, exchange, session, timeframe, date ranges, script version, pivot parameters, oscillator, costs, and fill settings with each run. Reserve a later period before choosing parameters. A parameter search on the full dataset leaves no untouched period for evaluating the selected rules.

• Pineify Strategy Optimizer can test exposed parameter combinations. Select a small research range and retain the untouched evaluation period separately.
• Backtest Deep Report can analyze exported trade records. Keep the signal log and script alongside the CSV: completed trades alone cannot prove when a pivot was first known.
• No performance experiment is reported on this page. Fewer trades after a filter are a change in sample, not evidence of better signals.