Research Note v2.0 — April 2026

From Commodities to Crypto:
A Regime-Adaptive Trading Framework

A systematic discretionary approach to crypto — applying 12 years of commodity trading intuition to a multi-layer, walk-forward validated decision framework.

Onesun Lee · USC Marshall MBA Class of 2026 · Ex-Samsung C&T · Ex-Triptik Trading SA

HEADLINE RESULTS — TRUE WALK-FORWARD OOS (2019–2025)
+638%
OOS Compounded
vs B&H +416%
0.79
Avg Sharpe
walk-forward OOS
-26%
Avg Max DD
vs B&H -65%
52%
Avg Exposure
position utilisation
1.16
Sharpe (Continuous)
full-period simulation
7yr
OOS Validation
2018 train → 2019 test
Core claim: Trained on 2018 on-chain data (earliest BGeometrics / CoinMetrics availability), the model achieves +638% compounded OOS return (2019–2025) versus Bitcoin Buy & Hold +416% over the same comparable period — with half the drawdown (MaxDD -26% vs -65%) and Sharpe 0.79 vs ~0.4.

01 — THE PROBLEM

The Old Playbook Is Broken

Most crypto trading analysis falls into one of two traps: single-indicator dependence (MVRV says buy, so buy) or narrative-driven positioning (ETF inflows = bullish, therefore long). Both approaches worked in previous cycles but have shown diminishing reliability in the post-ETF institutional era.

Since the January 2024 spot ETF approval, Bitcoin's market structure has fundamentally changed. On-chain metrics have decoupled from price action. Whale execution has fragmented across exchanges. Cycle indicators like Pi Cycle and 200W MA Heatmap failed to produce their historical overheat signals. The old playbook is broken.

This framework was built to address that structural shift — embedding on-chain analysis within a multi-layer decision architecture that adapts to the regime it operates in, and continuously updated as market structure evolves.

02 — ARCHITECTURE

Four Decision Layers

The model makes four decisions in sequence, each constraining the next. This mirrors how institutional commodity desks operate — you don't trade the spread without knowing the macro, and you don't size the position without knowing the spread.

Layer 1 — Macro Regime

DXY trend, M2 growth, VIX level, and HY credit spread classify the market into four regimes: EXPANSION, STABLE, TIGHTENING, BLACK_SWAN. All downstream decisions are regime-conditional — position floors, score thresholds, and signal activation vary by regime.

Layer 2 — Cycle Phase

MVRV Z-Score and NUPL determine where we are in the BTC market cycle: BEAR, BULL_EARLY, BULL_MID, BULL_LATE. Controls trailing stop width and position cap. BULL_LATE tightens stops; BEAR disables long entries regardless of score.

Layer 3 — Composite Score

Eight variables — MVRV Z-Score, NUPL, Fear & Greed, SOPR, VIX, DXY, 4-week and 12-week momentum — combined with regime-conditional weights. Score 0–100: high = fear (buy zone), low = greed (reduce zone). Contrarian framing: 92% accuracy at historical cycle inflection points.

Layer 4 — Position Sizing

Entry/exit thresholds, exposure floors and caps, trailing stop at 10%, and ETF-era structural signals (whale_dist, early_warning, trending_up, etf_bull) determine the final position. Score below 50 → 0% exposure. No leverage on the long side.

EXPANSION

Liquidity expanding, dollar weakening. Core long 90–98%. Wide trailing stops. ~31% of historical days.

STABLE

Mixed signals. Position scales with 200d MA + momentum. trending_up filter prevents false bull signals. ~60% of days.

TIGHTENING

Liquidity contracting, dollar strengthening. Minimal or short exposure. ~5% of historical days.

BLACK SWAN

VIX spike, credit stress. Immediate risk reduction. ~4% of historical days.

In a structural bull move, the biggest risk is not being in the market.

03 — V2 SIGNAL UPGRADES

Post-ETF Structural Adaptations

V2 introduces four structural signals developed in response to the January 2024 ETF approval — each addressing a specific failure mode identified in V1 OOS analysis.

whale_dist

Price/MVRV-Z Divergence. When BTC is near its 90-day high but MVRV-Z has rolled below 75% of its 90-day peak, institutional distribution is underway. Applied post-2024 only — the ETF era structurally changed how whale behavior manifests. Captured the 2025 drawdown that traditional on-chain metrics missed entirely.

early_warning

VIX + MVRV-Z Macro Stress. Fires when VIX > 25 AND MVRV-Z has fallen >50% over 20 days AND BTC is below the 200-day MA (outside BLACK_SWAN). Fired 76 days in 2022, zero days in 2019/2021/2023/2024/2025. Reduces position to 30% when active — directly addressing the 2022 January regime lag problem.

trending_up

STABLE Regime Momentum Filter. Position scales to 75% in STABLE only when BTC is above the 200-day MA AND composite score ≥ 35. Distinguishes genuine bull-market STABLE periods (2023 H2, +57.8% OOS) from sideways consolidation — directly addressing the "chopsolidation" regime classification problem.

etf_bull

ETF Flow Confirmation. 20-day cumulative ETF flow (Farside data, updated weekly) exceeds $1B. Provides additional position conviction in STABLE regime when institutional demand is confirmed. Fires zero days pre-2024 — structurally new signal for the institutional era.

V1 — In-Sample Issues

❌ Score weights fitted on full dataset
❌ OOS always starts at position = 0%
❌ No ETF-era signal adaptation
❌ STABLE regime too broad (chopsolidation)
❌ 2022 regime lag unaddressed

Result: +906% in-sample (look-ahead bias present)

V2 — True OOS Fixes

✅ Score recomputed per period, no leakage
✅ Initial position from actual regime state
✅ whale_dist + etf_bull for ETF era
✅ trending_up filter in STABLE
✅ early_warning for macro stress

Result: +638% compounded True OOS


04 — TRUE WALK-FORWARD OOS (2019–2025)

Year-by-Year Validation

Trained on 2018 data (earliest on-chain data availability). Each test year uses only information available at that date — no look-ahead, no parameter re-fitting.

Period Strategy B&H CAGR Max DD Sharpe Avg Exp
2019+40.5%+82.4%+40.6%-46.3%0.9361%
2020+211.6%+315.2%+211.6%-14.5%3.1461%
2021+15.5%+44.1%+15.5%-29.1%0.4649%
2022-36.9%-65.0%-37.0%-37.5%-2.2422%
2023+57.8%+153.3%+58.0%-18.3%1.8361%
2024+67.9%+107.8%+67.9%-14.4%1.9346%
2025-12.7%-9.7%-12.8%-24.3%-0.4964%
AVERAGE-26.4%0.7952%
OOS Compounded (2019–2025): +638%
vs Bitcoin Buy & Hold over the same period: +416%
Strategy outperforms B&H by +223 percentage points with half the drawdown.
Honest limitation: 6 of 7 years underperform B&H on annual excess return basis. The outperformance comes from superior risk management in down years (2022: -37% vs B&H -65%) and compounding effect. This is a risk-adjusted framework, not a pure return maximizer. Sharpe 0.79 vs B&H ~0.4 reflects the genuine edge.

05 — KNOWN LIMITATIONS & ACTIVE DEVELOPMENT

Where the Framework Breaks Down

2022 January regime lag: Macro indicators lagged BTC price by ~4 months. EXPANSION classification persisted while BTC was already falling. early_warning partially mitigates but cannot fully solve macro lag.
2024 vs 2025 separation: Both years had strong ETF inflows. The structural difference — retail leverage via funding rates (7x higher in 2024) — requires a funding rate momentum filter currently in development.
Structural long bias: BTC appreciated significantly during the test period. The framework's real test will come in a prolonged sideways or structurally declining market.
Active development: Funding rate momentum signal (Deribit) · LTH/STH SOPR integration · Continuous walk-forward validation as live data accumulates · Options overlay for BLACK_SWAN tail risk

06 — BACKGROUND

A Commodity Trader's Perspective

This framework was built by a commodity trader, not a quant. Over 12 years trading crude oil, naphtha, and petrochemicals across physical and derivatives markets at Samsung C&T (benzene specialist, ~2M tons/year, ~₩3T revenue, ₩10B+ annual P&L) and Triptik Trading SA (Geneva, Asia petrochemicals) taught me that markets are regime-dependent.

The benzene/naphtha spread behaves differently when Chinese inventory is at 200K tons vs. 30K tons. Contango vs. backwardation changes everything about position sizing and entry timing. You can't trade a single number; you need to know which regime you're in.

Crypto is no different. MVRV Z-Score at 2.5 in an EXPANSION regime (M2 growing, DXY falling) is a hold signal. The same reading in a TIGHTENING regime is an exit signal. The framework encodes this context-dependency explicitly — which is why adding more signals hurt performance, but adding better regime classification consistently improved it.

The transition from discretionary to systematic trading is not about replacing intuition with algorithms. It's about encoding intuition into a repeatable, testable framework — and then being honest about where it breaks down.