Correlation compression
If the rolling market mode rises above its upper quartile, the islands are likely to contract into one market-wide trade. Group diversification should then be treated more cautiously.
Current three-year window
What rose, what moved together, and what never quite belonged
Loading the current pipeline snapshot…
01 · Group portfolio returns
These are not official Hang Seng subindices. Each is a synthetic, equal-weighted portfolio of the companies the correlation analysis placed together, rebalanced daily and rebased to 100 at the start of the window. They answer a simple question: how did each discovered group actually trade?
02 · Strongest and weakest relationships
Pearson is the observed daily-return correlation. “Filtered” is the noise-reduced value used to draw the atlas. The ranking uses Pearson so the numbers remain directly interpretable; the second column shows what survives the statistical filter.
Highest correlations
| Companies | Pearson | Filtered |
|---|
Lowest correlations
| Companies | Pearson | Filtered |
|---|
03 · Which groups can be trusted?
Group loyalty comes from rebuilding the full map 200 times on block-resampled returns. A 100% group repeatedly re-forms with the same members; a lower score marks a looser, more sample-dependent grouping. This is also what controls marker color in the 2D and 3D atlas.
Average loyalty by group
Companies most between groups
04 · What the structure suggests next
This is a monitoring outlook, not a target-price model. It combines the latest rolling network regime, recent group direction and bootstrap loyalty to show what appears durable, what is unsettled, and which change would invalidate the current reading.
Current network regime
Calculating…
Signals that would change the reading
Boundary migration
A sustained change in loyalty among the palest companies would be early evidence that exposures are being repriced. Redraw the map before treating a new membership as durable.
Momentum confirmation
A one-month rebound only becomes a broader signal if it persists across the three-month window and spreads to several groups. A single winning line is not a market forecast.
05 · Reading the numbers
Return rankings are historical and highly sensitive to the selected window. Correlation is not causation, and low correlation does not mean that two companies will hedge one another in a future shock. Hong Kong's Investor and Financial Education Council likewise cautions that past performance is not indicative of future performance. The stress scenario is consistent with published evidence that equity correlations can rise when markets come under pressure (Preis et al., 2012). This is an educational view of past co-movement, not investment advice.