How 8Z compression principles detect market regimes and control trading aggression — the technical bridge from DNA analysis to financial markets.
Multiple mathematical generators compete to describe data. MDL picks the one with the shortest total description. DCC adapts search depth based on signal complexity.
DNA: CA rules, L-systems, logistic maps → Z-scores 28-74
Audio: LPC, periodic, harmonic models → beats FLAC
Multiple market regime models compete to describe price action. MDL picks the one with the lowest prediction error. DCC adapts trading aggression based on regime stability.
Crypto: BTC/ETH sync + MDL regime → captures momentum
Stocks: SPY/QQQ sync + MDL regime → same framework
The principle is identical: find the mathematical generator that compresses the data best (shortest description length), then use that generator's predictions. In DNA, the generator reveals biological structure. In audio, it enables better compression. In trading, it reveals the current market regime and optimal trade direction.
Each bar, 4 generators make predictions about the next price. The one whose total description is shortest (model complexity + prediction errors) wins = current regime.
// For each generator G: total_cost_G = params_G × model_penalty + log₂(prediction_error_variance_G) ↑ complexity cost ↑ data cost (how well it fits) // Winner = generator with LOWEST total cost regime = argmin(total_cost_TREND, total_cost_REVERSION, total_cost_MOMENTUM, total_cost_RANGE)
Why this works: MOMENTUM has only 1 parameter, so its model cost is lowest. But if the market is ranging, MOMENTUM's prediction errors will be huge → high data cost → it loses. RANGE has higher model cost but low prediction errors in sideways markets → it wins. MDL automatically selects the regime without any human judgment.
| Regime | Direction | Longs? | Shorts? | Logic |
|---|---|---|---|---|
| TREND ▲ | Up | ✓ Yes | ✗ No | Trade with the trend, not against it |
| TREND ▼ | Down | ✗ No | ✓ Yes | Short with the trend |
| REVERT ▲ | Below mean | ✓ Yes | ✗ No | Buy: price should revert up to mean |
| REVERT ▼ | Above mean | ✗ No | ✓ Yes | Sell: price should revert down |
| MOMENTUM ▲ | Up | ✓ Yes | ✗ No | Ride the velocity |
| MOMENTUM ▼ | Down | ✗ No | ✓ Yes | Ride downward velocity |
| RANGE ◆ | None | ✗ No | ✗ No | No clear direction → sit out |
Ported from 8Z-Audio and the TSP solver. DCC monitors the complexity of the regime-change stream and adjusts trading parameters in real-time.
Same regime winning repeatedly. Market has clear structure.
→ Tighter SL (×0.8–1.0)
→ Full position size (100%)
→ Lower impulse bar (easy entry)
→ EXPLOIT the stable regime
Moderate regime changes. Normal market conditions.
→ Standard SL (×1.0)
→ Normal size (75%)
→ Standard impulse bar
→ BALANCED approach
Regimes flipping chaotically. No stable structure to exploit.
→ Wider SL (×1.2)
→ Half position size (50%)
→ Higher impulse bar (strict entry)
→ EXPLORE / reduce exposure
| Concept | 8Z Audio Codec | Trading Strategy |
|---|---|---|
| Signal | Audio frame (4096 samples) | Price bar (1s / 1m / 1h) |
| Generators | LPC, Periodic, Harmonic, Delta | Trend, Reversion, Momentum, Range |
| MDL decides | Which predictor compresses best | Which regime fits price action best |
| DCC monitors | Predictor-change complexity | Regime-change complexity |
| Seizure = | Same LPC order winning → narrow search | Same regime winning → trade aggressively |
| Noise = | Winners jumping → widen search | Regimes chaotic → reduce size |
| u controls | Search depth (candidates tested) | Position size + SL tightness |
Each gate is independent. A signal must pass ALL gates to trigger an entry. This is the same principle as the DNA Scanner: multiple null models must fail before declaring structure is real.
The strategy auto-detects asset class from syminfo.type and configures appropriate sync pairs and parameter scales.
| Asset Class | Sync Leader 1 | Sync Leader 2 | Typical Correlation | Default Leverage |
|---|---|---|---|---|
| Crypto (BTC) | ETH/USDT | SOL/USDT | 0.82–0.90 | 100x (MEXC) |
| Crypto (ETH) | BTC/USDT | ETH/USDT | 0.82–0.90 | 100x |
| US Stocks | SPY | QQQ | 0.90–0.96 | 1x–4x |
| Forex | DXY | EUR/USD | -0.85–0.95 | 30x–50x |
| Commodities | SPY (risk) | DXY (dollar) | varies | 1x–10x |
Cross-asset considerations: Crypto correlations break during idiosyncratic events (ETH merge, regulatory news). Stock correlations are highest during crashes (correlation→1). Forex correlation structure is fundamentally different (currency pairs share components). The sync concept works across all classes, but threshold calibration must differ. The MDL regime detection handles this automatically — if the market structure changes, the winning generator changes, and the DCC adjusts.
| Feature | V2.2 | V3.0 | Impact |
|---|---|---|---|
| Pine version | v5 | v6 | Strict bools, dynamic requests, future-proof |
| Sync feeds | Lookahead ON (default) | Lookahead OFF (strict) | Removes potential bias, measures real edge |
| MDL regime | None | 4 generators | Structural market awareness |
| DCC adaptive | None | Coupling u ∈ [0,1] | Auto-scales aggression to market clarity |
| VWAP tilt | None | z-score filter | Bridges manual reversal edge |
| Cooldown | None | 8 bars | Breaks whipsaw loss chains |
| Circuit breaker | None | 5 losses → pause | Prevents drawdown spirals |
| Stop-loss | 0.033% | 0.045% (DCC-adjusted) | Fewer noise stops |
| Asset classes | Crypto only | Crypto, stocks, forex, commodities | Universal framework |
| New presets | S1-S6, TF1-TF5 | + STK1-2, FX1-2 | Stock and forex presets |
Extend generator library: Fourier harmonics, CA patterns, chaotic attractors. Refine MDL cost calibration. DCC threshold adaptation per asset class.
Monte Carlo validation framework. Permutation tests. Walk-forward optimization. Decompose the 70K% return into its real components.
Order book imbalance signals. Microstructure regime classifier. Options/funding rate as MDL generators. Continuous volume-weighted sync score.
News as MDL generator. Social sentiment regime detector. X/Twitter volume spikes. Narrative momentum indicator.
Sync Momentum v3.0 — MDL/DCC Universal Engine
Bojan Dobrečevič · AIM³ Institute · March 2026
Part of the 8Z-LO cross-domain compression research program.
Not financial advice. Past performance does not guarantee future results.