Live tool / NIFTY-50 correlation network
← BackNetwork Graph
Node size is eigenvector centrality; an edge is a kNN link on |ρ| over the trailing window; communities are Louvain. Move the window dial and the plate re-forms — it does not re-draw. Nodes hold their position in proportion to how little their structural role changed, so whatever moves is telling you its centrality was an artefact of the window.
What these numbers mean
- N
- How many stocks are in the graph. The NIFTY-50, minus any name without enough price history in the window.
- Edges
- How many links survived. Two stocks are linked when their returns move together strongly enough — here, each stock keeps its k=4 strongest partners. Not every pair is joined; that is the point.
- K
- How many communities the algorithm found — clusters of stocks that move as a bloc. Nobody tells it “banks” or “IT”; it finds them from the price data alone.
- Q
- Modularity, 0 to 1. How cleanly the market splits into those blocs. Q ≈ 0 means the grouping is no better than chance; Q > 0.3 means real structure; 0.4–0.7 is a strongly clustered market. Ours usually sits near 0.44.
- Asof
- The trading day the graph was built from. It is rebuilt for every session.
- Node size
- Eigenvector centrality, 0 to 1 — how central a stock is, counting not just how many things it moves with but how central those are. A big dot is a hub: the market’s weather reaches it first.
- Node colour
- Which community it belongs to. Colours are for telling groups apart, nothing more — they carry no ranking.
- Bars
- The estimation window: how many trailing days of returns the correlations are measured over. Drag it and the graph re-solves. This is the honest part — a structure that only exists at one window setting was never really there.
- The table
- Below the graph, every stock scored five ways (see the legend under it), with its rank on each. Click any row — or any node — to open that stock on its own.
- Strongest edges
- Inside a selected stock, its correlations (|ρ|, 0 to 1) with its nearest neighbours. 0.82 is a very tight co-movement; 0.30 is loose.
knn k=4 · pearson · LouvainCOMPUTING…
COMPUTING GRAPH…
Estimation window60 BARS
Centrality — every stock, five ways0 names · ranked by eigenvector
| # | Stock | Cmty | Eigenvector | PageRank | Degree | Betweenness | Closeness |
|---|
Click any row to inspect that stock. Or click a node in the graph above.
The five centrality measures
- Eigenvector
- Influence via influential peers. 0 to 1. Being linked to many stocks scores well; being linked to central ones scores far better. The top score is the market’s main hub — a shock there travels furthest. This is also what sizes the nodes in the graph.
- PageRank
- Where a random walk settles. Set something loose on the network and let it wander the links; PageRank is the share of time spent on each stock. Kinder than Eigenvector to well-connected names outside the core. Sums to 1 across all stocks, so values are small — ~0.02 is average here.
- Degree
- Raw connectedness. The summed correlation strength of every link a stock has. Blunt but honest: it counts the crowd, and doesn’t weigh whether the crowd matters.
- Betweenness
- The bridge. How often a stock sits on the shortest path between two others. High = it connects otherwise separate parts of the market; it may be small, yet removing it would split the network. Most names score near 0 — that skew is the signal.
- Closeness
- Reach. The inverse of the average distance to every other stock. High = news reaches it quickly from anywhere; low = it sits out on the rim.
- # of 49
- Each stock’s rank on that measure across the whole universe, 1 = most central. A name can top one measure and sit mid-table on another — that disagreement is often the interesting part.
Communities— detected