Advanced Fibonacci: confluence, anchors and testing rules

Advanced retracement work starts by limiting drawing freedom. If the swing, interval, coefficient and zone width can change after seeing a reaction, a convincing match is easy to select retrospectively. Define what qualifies as confluence, preserve when anchors became available and test whether an added condition improves results after costs.

Confluence does not prove independent evidence

Confluence means selected levels are near one another or several analysis conditions occur in the same region. Daily and four-hour retracements and an average may derive from overlapping prices. Three tool names do not automatically provide three independent observations. Copying one grid to another interval adds no new market data, and a drawing does not identify how many participants placed orders there.

Ratios of Fibonacci numbers approach the reciprocal of the golden ratio: 21/34 is about 0.6176 and 144/233 about 0.6180. The limit is 1/φ ≈ 0.6180339, where φ = (1 + √5)/2 ≈ 1.6180339. These are mathematical reference levels. Their identity alone does not show where market orders are placed or establish a probability of a price reversal.

Freeze definitions and information availability

Record feed, price type, timezone, intervals and the rule selecting extremes. TradingView permits exact coordinates, Reverse and optional log-based calculation. This lesson uses linear prices with 0% at the move’s end B and 100% at its start A, so P(r) = B + r(A − B). Identical anchor prices without identical settings need not give identical lines.

Separate an extreme’s timestamp from its recognition time. A current daily close is not final at midday; a confirmed pivot can be drawn into the past despite requiring later bars. Preserve the values available at the decision time instead of import final higher-interval data into an earlier signal.

What confluence of retracements is and why it matters

The calculations are 1.12000 − 0.618 × 0.04000 = 1.09528 and 1.10080 − 0.5 × 0.01080 = 1.09540. For several levels, measure the group’s maximum minus minimum, not only neighboring gaps. Adding 1.09570 makes the full span 4.2 pips, failing the same 2-pip convention. Grouping tolerance, zone width and stop distance remain different quantities.

In an independent variation, raise only the larger move’s endpoint from 1.12000 to 1.12100. Its 61.8% line becomes 1.095662. The 10-pip anchor change moves the line 3.82 pips; its distance from the unchanged smaller-move level is now 2.62 pips and fails the chosen test. Rounded to five decimals, 1.09566 still gives 2.6 pips. A tiny-looking anchor choice can change classification without supplying new trend information.

If two predefined anchor-selection methods yield different qualifications, keep both records. Do not choose the one matching a later bounce and discard the other. Coincidence with an average or a drawn supply/demand zone is another hypothesis, without automatic independence or evidence that orders cluster there.

Harmonic definitions require their own complete structures. In Scott Carney’s method, Gartley uses B near 61.8% and D near 78.6% of XA; Bat uses an 88.6% XA retracement, whereas ideal Butterfly and Crab use XA projections around 127% and 161.8%. The 78.6% label therefore does not describe every pattern’s D point, and each method has additional leg conditions. These promoted definitions identify geometry, not a safe boundary or an independent efficacy study.

Search freedom can produce more near matches

Define the exact number x = (√5 − 1)/2. It is the positive solution of x² + x − 1 = 0, so 1/x = x + 1. Its decimal expansion starts 0.6180339. The rounded value 0.618 is only an approximation; substituting it does not preserve the exact identity. Separately, the regular-pentagon diagonal-to-side ratio is φ ≈ 1.6180339.

Consider an own hypothetical control model, not market observations. Freeze one reference price inside a 100-pip interval, at least 2 pips from either endpoint. Draw each extra line independently and uniformly from that interval. A match within 2 pips occupies a 4-pip band, so one line matches with probability 4/100 = 0.04. For m lines, at least one match has probability 1 − 0.96^m: 4.00% for one, about 33.52% for ten and 87.01% for fifty.

Real Fibonacci grids share anchors and dependent levels, so those probabilities do not describe Fibonacci patterns or future bounces. The model illustrates selection freedom: record every searched grid and coefficient, rather than only the pair ultimately chosen. Apply the same search rule to favorable and unfavorable observations.

Separate the observed zone from fills and costs

“Price confirmed the level” needs a measurable definition. An illustrative research rule requires a closed H1 candle’s Bid low inside 1.09528–1.09540 and close above 1.09540. Low 1.09532 and close 1.09550 pass that description. They do not establish a purchase at the candle low; the actual entry remains unknown until execution.

1. Assumed entry. In a separate hypothetical settlement, buy EUR10,000 at Ask 1.09565.
2. Alternative exits. Selling at Bid 1.09365 gives −USD20 before additional costs; selling at 1.09965 gives +USD40. After assumed USD4 costs, these become −USD24 and +USD36: ratio 1.5 although the price distances give 2. Spread is already in the assumed fills.
3. Worse stop execution. If an ordinary stop instead fills at Bid 1.09265, the same model gives −USD30 before costs and −USD34 after them. This is not the maximum possible loss; specify expiry, cancellation and exits separately.

Practical takeaway: A filter may raise win percentage while reducing total results by omitting trades. Compare the same observation window and costs.

Preserve the complete test specification

Compare the same entry rule with and without the confluence filter, retaining identical execution, costs and exits. Include transaction count, net total, drawdown and the individual-result distribution; fewer entries or a higher win rate alone do not establish a better system.

• Anchor rule. State how each swing becomes eligible without a universal H4 or minimum-candle requirement.
• Availability. Record recognition time and whether higher-interval data are final.
• Search history. Preserve every tested anchor, coefficient and tolerance, including failed variants.
• Execution scope. A cluster width is not a stop buffer or loss cap; no universal 15–20-pip distance follows from a ratio.
• Unused data. Select parameters on one period and assess another without changing rules. Compare filter versions under the same observation and cost assumptions.

A worked example — how to calculate a confluence zone

These calculations describe reference levels, not a forecast for either pair. Reproduce a claimed second grid from its own anchors and settings before evaluating proximity. Keep the geometric check separate from the observed reaction and from a strategy’s executed result.

What to do tomorrow

The diagram’s predefined mask passes trades 1, 3 and 8 from eight fixed net results. Win rate rises from 4/8 to 2/3 and mean from +0.5 R to +1 R, while total falls from +4 R to +3 R. These are different metrics over one hypothetical window, without a Fibonacci backtest or probability estimate. Do not redesign the mask after seeing the four winners.

1. Recalculate the cluster. Verify 1.09528 and 1.09540, the 1.2-pip separation and the chosen 2-pip tolerance; then check the added third line.
2. Test one anchor change. Raise the larger endpoint to 1.12100, retain the smaller grid, and reproduce the 2.62-pip failure without retuning the tolerance.
3. Separate price and result. Recompute the EUR10,000 alternative exits and USD4 costs. Do not add the spread twice or confuse the chosen zone with a maximum loss.
4. Compare a fixed filter. Use identical entry/exit/execution rules with and without the extra condition, retaining failed variants and unused data. A positive comparison supports further checks rather than promise future success.

Related reading: Fibonacci retracement basics explains formulas and orientation; the confluence system discusses combining specified conditions; multi-timeframe analysis separates the information available on different intervals.