September 11, 2026 · BTC at $76,774 · forward liquidity tilt, ~11-week lead
Liquidity is draining on a forward lead. A tilt against risk, not a forecast.
Liquidity leads BTC by roughly 6-16 weeks (peak ~11), and the IC survives when the lead isn't cherry-picked. The right test isn't whether it beats price momentum standalone — momentum already embeds liquidity — but whether it ADDS to it. Orthogonalized to 12-week momentum, liquidity's residual IC is ~+0.20, and a combined momentum + liquidity signal earned a higher out-of-sample Sharpe (1.31 vs 1.21) than momentum alone. A real incremental edge, validated across three crypto cycles (2014-26) — not standalone alpha.
Three layers — US net liquidity (WALCL − TGA − RRP), FX-adjusted global central-bank balance sheets, and a financial-conditions overlay (NFCI + 10y real + HY OAS) — each differenced into a 13/52-week rate of change, z-scored against three years of its own history, then averaged and mapped to 0-100 through the normal CDF (50 = neutral). The forward read shifts that signal by each asset's empirically-measured lead.
Why change, not level: a balance-sheet level is non-stationary and correlates with any trending asset by construction — the classic spurious-fit trap that wrecks retail net-liquidity charts. Momentum is what actually leads. Every series is lagged by its real publication delay in the backtest so it isn't fiction. Deliberately down-weighted: RRP (its 2022-24 channel is spent — successor via SOFR−IORB is a TODO); the global FX conversion (a mechanical dollar artifact); and it can't see the AI-capex factor that drives ~half of QQQ. Reproduce it with scripts/validate_liquidity.py.
Liquidity here is a coincident-to-modestly-leading read on a ~6-16 week lead, small but real and validated across three crypto cycles. The score is a probability tilt, not a forecast: read a 70 as “odds favor risk,” not “risk will rise.” It doesn't replace price momentum — it adds to it: orthogonalized to 12-week momentum, its residual edge is ~+0.20 for BTC, and a combined momentum + liquidity signal beat momentum alone out of sample (Sharpe 1.31 vs 1.21). Use it alongside trend, not instead of it.
What would make this wrong: it is blind to shadow money — dealer repo, collateral and rehypothecation chains — which is roughly 70% of the real edge and isn't reachable from public balance-sheet data. And in a genuine stress flush, liquidity and risk can fall together — though in this sample the lead actually held in stress, which is why the guard warns rather than trusting it blindly. No dashboard proves itself in real time; the honest bar is beating 12-week momentum out of sample.