Personality clusters

Every market has two personalities. Its structural identity is the signature it carries across decades: how trending or mean-reverting it is on average, how fat its tails usually are, whether its vol regime is stable or coiled. Its current state is where it sits today — which can drift a long way from the structural norm when the macro environment shifts.

We measure both. The two maps below show the same 18-market universe viewed through those two lenses. Markets that cluster together on the structural map share long-run personality. Markets that cluster together on the current map are in similar regime state right now — even if their structural identities are nothing alike.

Reader shortcut: structural answers “what kind of market is this?”; current answers “what’s it doing today?” When the two disagree, the market is unusually far from itself.

The structural identity map uses each market’s 37-year lifetime-mean Big-6 features. This is what the market looks like across decades — its baseline personality, not this week’s pose. Hover any dot for its market name and PC coordinates.

The current state map uses the most recent reading of each Big-6 feature. Markets drift around this map as regimes change. A market that lands far from its usual neighbors is doing something unusual. Hover any dot for its market name and PC coordinates.

The two maps disagree on about 80% of top-3 neighbor rankings. That’s not noise — it’s the whole point of measuring both.

A market’s structural identity is slow-moving: ES has been a trending, normal-vol, clean-tail asset for decades. But its current reading can drift anywhere on the map as the macro environment pushes it around. The gap between a market’s structural position and its current position is itself a signal: it tells you how far this market is from itself right now.

That gap is the raw material for our next product — a continuous stress detector that flags markets as they drift away from their structural norm, coming in Phase 4.

If the scatter maps show the big picture, the distance heatmaps show every pairwise comparison in one view. Rows and columns are sorted by cluster; dark cells are market pairs that share personality. Hover any cell for the exact distance and the market pair.

Both maps use the same recipe:

1. Features: the Big-6 — Hurst exponent, annualized vol, vol-of-vol, kurtosis, skewness, sample entropy.
2. Snapshot: structural uses each feature’s 37-year lifetime mean; current uses the most recent daily reading.
3. Normalization: z-score across the 18-market universe (standardized per feature), then winsorized at ±2.5 so one extreme market can’t dominate the geometry.
4. Distance: Euclidean in z-space.
5. Clustering: K-means at k=4. (Silhouette-picked k=2 on both snapshots but the pick is dominated by a single outlier — VX structurally, NG today — so we force k=4 for interpretable groupings.)
6. Visualization: PCA to 2D for the scatter; sorted raw-distance matrix for the heatmap.

Last rebuild: structural 2026-11-07T03:03:24+00:00, current 2026-11-07T03:03:24+00:00. Pipeline: build_personality_similarity.py.

• All markets — tile grid grouped by archetype. Pick any market to see its DNA radar and current-state readings.
• Feature glossary — plain-English definitions for each Big-6 feature used above.
• Continuous stress detector (Phase 4) — the current-minus-structural delta, per market, as a live stress score.