Markets overview

Units and statistics

Each series has a kind that sets its units everywhere. Prices (ES, DXY, USD/JPY…) show % changes and annualized volatility of daily log returns. Rates and spreads show changes in basis points and annualized normal (bp) volatility of daily changes, the convention on rates desks. Levels (VIX, NFCI…) show point changes. Volatility is both a 21-day sample standard deviation and RiskMetrics EWMA, σ²_t = λσ²_t−1 + (1−λ)r²_t−1, λ = 0.94, annualized by √252 (√365 for Bitcoin, √52 for weekly series). Range position is the mid-rank percentile of today's value against the prior year, plus a robust z-score, (x − median) / (1.4826 · MAD), which is not distorted by the fat tails and spikes that dominate stress series.

Risk Appetite Index

A composite in the Credit Suisse / Goldman Sachs risk-appetite tradition (Kumar & Persaud, IMF 2002), with transforms matched to the statistics of each input. Trending assets (ES, Bitcoin, USD/JPY, inverse DXY) enter as volatility-scaled momentum: the 21-day log return divided by EWMA vol × √21, a t-statistic-like score. Mean-reverting stress gauges (VIX, VIX/VIX3M, MOVE, high-yield OAS) enter as robust z-scores of their level against the trailing year, sign-flipped. Scores are clipped to ±3 and averaged with equal weight, so no single gauge can dominate and the result needs no unstable PCA re-weighting. The label uses a 3-day mean: ≥ +0.75σ risk-on, ≤ −0.75σ risk-off. Every point in the history uses only data available on that date.

Absorption ratio

Kritzman, Li, Page & Rigobon (2011, Journal of Portfolio Management): the fraction of total variance explained by the top fifth of eigenvectors of the correlation matrix of daily returns (yield changes for the 10Y) across ten assets, over a 63-day window. A high ratio means markets are tightly coupled and shocks propagate broadly. The standardized shift, ΔAR = (15-day mean − 1-year mean) / 1-year std, is the signal the authors found preceded drawdowns; above +1σ is flagged.

Recession probability

The New York Fed model (Estrella & Mishkin 1996; Estrella & Trubin 2006): P(recession in 12 months) = Φ(α + β·S), where S is the one-month average of the 10-year minus 3-month Treasury spread in percentage points, with the published coefficients α = −0.5333, β = −0.6330. Both legs are FRED H.15 constant-maturity closes. The NY Fed uses monthly averages and the bond-equivalent 3-month yield, so values can differ from its published table by a point or two.

Other series

Curve spreads use H.15 closes on both legs, never a live quote against a stale one. Credit is the ICE BofA option-adjusted spreads (Gilchrist & Zakrajšek 2012). The term premium is the Kim–Wright estimate from the Fed Board. NFCI (Chicago Fed) and STLFSI4 (St. Louis Fed) are weekly composites centred on zero. Yen positioning is CFTC Legacy non-commercial net futures, with a robust z-score against three years. Carry-to-vol divides the Fed–BOJ gap by EWMA-annualized USD/JPY vol, a free proxy for the implied vol used in practice. This is decision support, not a trading signal.