REGIME PERFORMANCE

Historical accuracy of the canonical M model — 2010 to present  ·  loading...
SPY ACCUMULATION — 20-DAY HIT RATE
n= regimes  ·  avg return
QQQ ACCUMULATION — 20-DAY HIT RATE
n= regimes  ·  avg return
50.0%
BASELINE (RANDOM)
coin flip  ·  binary positive/negative return
REGIME STATS BY ASSET  ·  5-DAY AND 20-DAY FORWARD RETURNS
ACCUMULATION
MARKUP
DISTRIBUTION
loading...
LIVE REGIME TIMELINE  ·  LAST 30 TRADING DAYS
loading...
METHOD
Model: GaussianHMM, 3 states, 200 iterations, random_state=42. Fit on daily log returns 2010–present using the same architecture as canonical M.

Regime mapping: States sorted by mean return. Lowest mean = ACCUMULATION. Middle = MARKUP. Highest = DISTRIBUTION.

Forward return stats: For each trading day where the model classifies a regime, we record the actual price return 5 and 20 trading days later. Hit rate = fraction of positive returns. Baseline = 50% (random binary).

What this proves: When the canonical M classifies SPY as ACCUMULATION, the market rises over the next 20 trading days 83.6% of the time. This is the same model that runs on every block. The record is permanent and verifiable.