Macro Liquidity & Credit Stress Simulator V3for JcbNext

How scoring works. Each indicator's level is mapped to a 0–100 stress score by piecewise-linear interpolation between calibrated thresholds. Scores are then aggregated convexly (a generalised/power mean — p=2 within a category, p=3 across categories), not by a plain average. This is deliberate: a crisis is defined by something breaking, so a calm market must not be allowed to average away an extreme one. When readings are similar this behaves like a normal weighted average; it only diverges when stress is concentrated — exactly when it matters. Missing inputs are excluded and weights re-normalise automatically.

Regime bands are backtest-calibrated. The thresholds (Crisis ≥75, High Stress ≥58, Tightening ≥40) were set so the composite lands in the right band across historical crises — see the Crisis Backtest: 2008 → Crisis, COVID → High Stress, 2022/SVB → Tightening.

Why these weights — the framework is funding-led, because liquidity crises start in the plumbing and propagate outward:

• Interbank funding — 20%. FRA-OIS & SOFR-OIS are the earliest, first-order gauge of bank-to-bank funding stress. When this breaks, everything follows.
• USD funding / cross-currency basis — 20%. Measures how scarce dollars are globally. A parallel first-order signal to interbank.
• Credit spreads — 20%. HY & IG OAS are the real-economy stress signal and the most-watched recession gauge.
• Treasury volatility — 15%. MOVE & the 2s10s curve drive collateral and hedging costs; second-order to direct funding.
• Equity risk appetite — 10%. VIX is a sentiment meter — useful, but a confirmation rather than a cause.
• Banking liquidity — 10%. Reserves, RRP and discount-window borrowing move slowly but define the system's buffer.
• Repo market — 5%. The canary in the coal mine: rare, but a high-signal early warning when it spikes.

Credit de-duplication. The credit score uses cash OAS only (HY + IG). CDX IG/HY measure the same underlying credit risk via CDS, so including them would double-count credit — they stay visible in the table for context but are excluded from the composite.

Historical context. Each FRED indicator shows a percentile and z-score vs its own trailing 5 years (the "vs 5Y" column) — answering "is today unusual?", not just "what is the level?".

Weights are a transparent house view, not a fitted model. The Crisis Backtest shows how the composite behaved in past stress events.