YPHER

Bitcoin Candlestick Patterns Are Lying to You

| Debonnaire
research patterns

The Pattern Recognition Problem

Every trading book teaches the same patterns. Hammer. Engulfing. Morning Star. Doji. The implication is always the same: if you see this shape, the market will do that thing.

But has anyone actually tested these claims on Bitcoin? Rigorously? With proper statistical controls?

We did.

8,000 Hypotheses, One Correction

CYPHER’s pattern research pipeline doesn’t just test the classic 40-50 candlestick patterns. It tests every meaningful combination of candle properties — body size relative to range, wick ratios, volume confirmation, position within recent price structure, and time-of-day effects.

This generated approximately 8,000 individual pattern hypotheses across multiple timeframes.

Here is the problem: when you test 8,000 hypotheses at a 5% significance level, you expect 400 false positives by pure chance. This is the multiple comparisons problem, and it is the reason most published pattern research is worthless.

We applied the Benjamini-Hochberg (BH) correction — a standard statistical procedure that controls the false discovery rate across all tests simultaneously. It is more permissive than the conservative Bonferroni correction, but it still eliminates the vast majority of spurious results.

What Survived

Out of 8,000 hypotheses, 598 patterns survived BH correction with a controlled false discovery rate of 5%.

These 598 edges are not theoretical. They are:

  • Statistically significant after multiple comparison correction
  • Economically meaningful — the expected move after the pattern is large enough to overcome spread and fees
  • Regime-tagged — each pattern is labeled with the market conditions where it holds (trending, ranging, volatile)
  • Integrated into CYPHER’s signal pipeline as one of the voter inputs

The other 7,402 patterns? They looked real in a backtest. They were noise.

Why This Matters

Pattern recognition without statistical rigor is astrology. It feels meaningful. It tells a story. But it does not survive contact with live markets.

The 598 edges in CYPHER’s pattern library are the ones that did survive. Every one of them earned its place through data, not dogma.

Back to Blog