ML Market Map — Hong Kong Equity Clusters for 2026-10-15

A daily unsupervised machine-learning read of the Hong Kong Equity market: 2,487 stocks grouped into 7 consensus clusters (KMeans + Gaussian-mixture + hierarchical, over robust-scaled PCA features) for 2026-10-15. Descriptive, not predictive — there is no buy or sell signal. Research, not investment advice.

Markets: US Equity · China A-Share · Japan Equity · Hong Kong Equity

Today's clusters

large cap · shallow drawdown · high margin — 887 names; mostly Industrials; drivers: r2 252 d (+0.61), log market cap (+0.53), max drawdown 252 (+0.46)
low margin · expensive (low E/P) · high downside vol — 520 names; mostly Industrials; drivers: profit margin (−0.66), val ep z (−0.56), vol of vol 63 (+0.53)
expensive (low E/P) · high leverage · low margin — 251 names; mostly Real Estate; drivers: val ep z (−3.85), leverage debt to mcap (+2.94), profit margin (−1.33)
wide upside excursions (21d) · rising (3m) · favorable excursion edge — 217 names; mostly Industrials; drivers: pa mfe 21d (+2.95), ret 20d (+2.04), rv 20 (+1.68)
high turnover · attention spent · attention flaring — 214 names; mostly Technology; drivers: turnover to mcap (+3.93), pa sir recovered (+3.06), pa sir infectious (+1.15)
high leverage · high cash yield · cheap (high B/P) — 207 names; mostly Industrials; drivers: leverage debt to mcap (+2.91), val cfop z (+2.65), val bp z (+1.62)
strong 1y momentum · wide 1y upside range · favorable 1y edge — 191 names; mostly Industrials; drivers: ret 252d (+3.33), pa mfe 52w (+2.80), ret 126d (+2.12)

Machine-readable data (free, read-only JSON)

The full map, per-ticker cluster assignments with confidence and anomaly scores, and PCA structure are published as open JSON for automated and AI-analyst consumption:

/api/ml/latest?market=hk — clusters, names, sector mix, representative tickers, top features
/api/ml/tickers?market=hk — per-ticker cluster, confidence, anomaly score, day-over-day migration, tenure
/api/ml/pca?market=hk — PCA components, explained variance, feature loadings

Descriptive market-structure research only. Unsupervised clustering finds structure, not direction; a tight cluster or an anomaly is a starting point for research, never a trade signal.