Indicator confluence appears on random data too
When 8 moving averages "all agree" on a chart, is that because the market shows structure, or because moving averages of the same price are mathematically redundant by construction? This page answers that question with statistics — not with predictions.
Method
- 200 deterministic random walks (seeded Brownian + small drift, 250 bars each).
- Three indicator setups applied to each walk: (A) 5 moving averages (5/10/20/50/100), (B) 8 Fibonacci-period MAs (3/5/8/13/21/34/55/89), (C) 5 diverse indicators (momentum / volatility / 24-bar cycle / long-term divergence / independent noise).
- For each walk, compute the effective independent count N_eff = K² / Σᵢⱼ ρᵢⱼ² over the K indicators (Cheverud-style correction).
- Plot the histogram of N_eff over 200 trials. Mark one trend-bearing observed series as a red vertical-style marker for percentile comparison.
Formula: N_eff = K² / Σᵢⱼ ρᵢⱼ² — K = number of indicators. ρᵢⱼ = Pearson correlation between indicator i and j. Fully independent → N_eff = K. Fully redundant → N_eff = 1.
A. Five moving averages (5 / 10 / 20 / 50 / 100)
The classic FX "5-indicator confluence" setup: five MAs of different periods applied to the same close price.
Reading: Even on random walks, this setup concentrates N_eff near 1.0–1.7 — far below the apparent K=5. The visible "5-indicator agreement" carries only ~1–2 independent votes' worth of information. The market is not telling you anything here that random data does not also "tell".
B. Eight Fibonacci-period moving averages — the high-redundancy trap
MAs at periods 3, 5, 8, 13, 21, 34, 55, 89. These appear extensively in retracement / Elliott-style commentary.
★ Reading: Random walks produce essentially the same N_eff distribution as the observed trend-bearing series. The observed point lands in the 90+ percentile of the random distribution — statistically indistinguishable. "Fibonacci MA confluence" is a property of the indicator design, not evidence of structure in the price.
C. Five diverse indicators (momentum / volatility / cycle / long-term divergence / noise)
Indicators built from genuinely different signals on the same price series.
Reading: N_eff stays close to K=5 on random walks — independence is preserved by design. When the observed series has real periodic structure, it moves N_eff modestly out of the random band. This is what an indicator panel looks like when redundancy has been minimized.
Takeaway
How much agreement an indicator panel shows tells you mostly about how the panel was designed, not about the market. The same Fibonacci MA confluence pattern that looks impressive on a real chart appears equally often on pure noise. The honest counter-measure is not "more indicators" — it is choosing indicators that measure genuinely different things, and accepting that when a panel is structurally redundant the "agreement" is uninformative.
Research details
A more detailed presentation of the same analysis — with 8-state classification labels (the Rei research framework) and richer interaction — lives on the operator's research site: rei-aios.pages.dev/#/neither-flat-eval ↗
What this page does NOT do
- It does NOT predict price direction. The whole point is the opposite: showing where indicator "agreement" is uninformative.
- It does NOT use real market data. Synthetic random walks only — the redundancy structure of indicators is the same regardless.
- It does NOT recommend any indicator setup. Even "diverse" panels (Scenario C) carry no guarantee of predictive value.
- It does NOT provide investment advice. This is a statistical demonstration, not a trading method. (FIEA / 金商法 line-safe.)