Quantitative Research Prototype
Binary state encoding
for market regime
detection
The 16-state Fâ geomantic system — a structured combinatorial space — reinterpreted as a discrete 4-bit categorical variable. Evaluated against Gaussian hidden Markov regime models, CUSUM volatility breakpoints, and walk-forward information coefficient validation.
No a priori causal claim. Each figure is a categorical market state. The question is purely empirical: does this encoding carry statistically significant information?
16 binary states — click any figure
● = bit 1 ●● = bit 0
MSB (Bit 3) = macro trend · LSB (Bit 0) = momentum
Primary finding
IC = −0.10
Spearman ρ at 5-day horizon
Statistical test
p = 0.03
Permutation test, n = 1,000 — H₀ rejected
Signal interpretation
Contrarian
Negative IC → mean-reversion encoding, not momentum
Research question
Does a 4-bit categorical market-structure encoder derived from the Fâ geomantic combinatorial system carry statistically significant predictive information, improve hidden Markov model regime labelling, or detect volatility regimes beyond standard technical indicators?
We test using permutation-based IC validation (H₀: IC = 0), regime-conditional forward-return t-tests, and expanding-window walk-forward analysis to verify temporal stability of any detected signal.
Encoding system — 4-bit market state
| Bit | Dimension | Condition (= 1) | Weight |
|---|---|---|---|
| Bit 3 | Macro trend | Close > MA(200) | 0.40 |
| Bit 2 | Medium trend | Close > MA(50) | 0.30 |
| Bit 1 | Short trend | Close > MA(20) | 0.20 |
| Bit 0 | Micro momentum | RSI(14) > 50 | 0.10 |
0000 = Yeku (all signals bearish, crisis_propensity = 0.95) → 1111 = Gbe (all signals bullish, crisis_propensity = 0.05)
Spectrum: 0000 → 1111
Analysis pipeline
01
Symbolic Encoder
4-bit binary → Fâ figure assignment. All signals lagged by 1 day (look-ahead bias prevention).
02
Feature Engine
Rolling MA, RSI, Bollinger Bands, ATR, realized vol, drawdown. lag_signals=True default.
03
Regime Detector
Gaussian HMM on [log_returns, vol_20, drawdown]. HMM states labelled by Euclidean distance to Fâ attribute vectors.
04
Validation Layer
Spearman IC × horizons [1,5,21,63d]. Permutation test H₀: IC=0. Walk-forward expanding window.
API quick start
POST /api/analyze
curl -X POST /api/analyze \
-H "Content-Type: application/json" \
-d '{
"ticker": "SPY",
"start": "2018-01-01",
"end": "2024-12-31",
"n_regimes": 4,
"benchmark": "^GSPC",
"allow_short": false
}'Response schema (excerpt)
{
"ic_table": [
{ "horizon": 1, "IC": -0.06, "p_value": 0.12 },
{ "horizon": 5, "IC": -0.10, "p_value": 0.03 },
{ "horizon": 21, "IC": -0.04, "p_value": 0.31 }
],
"permutation_test": {
"observed_ic": -0.10,
"p_value": 0.031,
"reject_h0": true
},
"chart_data": { "dates": [...], "close": [...] }
}Epistemic note
The Fâ system is used here as a structured discrete-state vocabulary for labelling market configurations — not as a causal or predictive claim. The 16-figure space is a natural combinatorial consequence of 4 binary market indicators; the Fâ naming and attribute system provides a human-readable label for each state. Statistical significance (p = 0.03) establishes that the encoding is non-random relative to forward returns. It does not establish directionality of causation, out-of-sample stability, or practical trading value.