Methodology

Data sources

• CoinGecko: prices, volumes, and market caps for the top-100 assets and the USDT/USDC/DAI stablecoin supply.
• FRED (St. Louis Fed): WALCL (Fed balance sheet, weekly), WTREGEN (Treasury General Account, weekly), RRPONTSYD (overnight reverse repo, daily).
• alternative.me: the crypto Fear & Greed index (daily).
• Binance / OKX: BTC perpetual funding rates.

Ingestion runs server-side every 5 minutes. When a source fails, the affected value is stored as null and shown as “—”. We do not substitute estimates.

Net liquidity

net_liquidity = WALCL − WTREGEN − RRPONTSYD

The Fed's balance sheet minus the two big “parking lots” that drain usable dollars from the system. Rising net liquidity has historically coincided with easier financial conditions.

WD Liquidity Index (0–100)

stablecoin component  = min(30, stablecoin_mcap / $200B × 30)
price component       = min(25, btc_price / $150k × 25)
sentiment component   = min(20, fear_greed / 100 × 20)
liquidity component   = min(25, net_liquidity / $7T × 25)

liquidity_index = round(sum of components), clamped to 0–100

WD Risk Score (0–100)

risk_score = fear_greed                     # base

# funding adjustment (skipped when the feed is down)
funding > 0.05%  → −10      crowded longs
funding > 0.02%  → −5
funding < −0.01% → +5       shorts paying

# 24h momentum adjustment
change > +5% → +8      change > +2% → +4
change < −5% → −8      change < −2% → −4

clamped to 0–100

Regime classification

risk_score ≥ 55 → RISK-ON
risk_score ≤ 40 → RISK-OFF
otherwise       → NEUTRAL

hysteresis: a regime flip is committed only after 6 consecutive
5-minute runs (30 minutes) agree on the new classification.

Historical statistics

Episode durations, BTC returns, and transition frequencies are computed from a daily regime series produced by running the exact formulas above over historical data (CoinGecko daily closes, the full Fear & Greed archive, FRED observations since 2022, and Binance funding history), then extended forward by live ingestion. Every aggregate displays its sample size; below n=5 completed episodes we display “insufficient data” instead of a number.

Limitations, read this part

• The regime thresholds (40/55) and index component weights are design choices, not fitted parameters. They have not been optimized against out-of-sample data.
• Fear & Greed dominates the Risk Score by construction. The score is best read as “sentiment, sanity-checked by positioning and momentum.”
• The “volume trend” cluster is a 24h volume delta, a rough proxy. It is not on-chain exchange flow data.
• Transition frequencies are descriptive day-to-day counts, not a predictive model.
• Backfilled history uses daily closes; the live series uses 5-minute data with hysteresis. Boundary timing can differ by up to a day between the two, which is why episodes are labeled “modeled” vs “live”.
• Nothing here is investment advice.