> Financial assets don't move independently — they form a web of relationships. This page builds that web from raw price data and lets you explore its structure.
The process works in three steps:
- Rolling correlations — For each trading day, we compute pairwise Pearson correlations over a sliding window of returns (default: 60 days). This produces a correlation matrix that evolves over time.
- Thresholding — We convert the correlation matrix into a graph by drawing an edge between two stocks whenever |ρ| exceeds the threshold (default: 0.3). Pairs that are uncorrelated get no edge.
- Network snapshots — Each date yields a separate graph. By stepping through dates you can watch the network rewire itself as market regimes shift — edges appear and disappear as correlations strengthen or weaken.
This network is the input to the curvature computation. Ollivier-Ricci curvature is calculated on each edge, measuring whether a connection is "well-supported" by neighboring structure (positive curvature) or acts as a fragile bridge (negative curvature).
Parameter guidance: A 60-day window with a 0.3 threshold is a good default — it captures medium-term co-movement while filtering out noise. Shorter windows (20–30 days) react faster but produce noisier networks. Higher thresholds (0.5+) keep only the strongest relationships, giving a sparser graph. The Minimum Spanning Tree method guarantees a connected graph with exactly N-1 edges, useful if you want to see the backbone of the correlation structure.
Use the arrow buttons or click Play to animate through dates and watch the network rewire. During calm markets you'll see loose sector clusters; during stress, edges multiply as correlations spike and the graph densifies.
> Loading network...
What changes over time?
Each date shows a different network snapshot built from that day's 60-day rolling correlations. As you step through dates, watch for these patterns:
- Edges appearing/disappearing — correlations cross the threshold as market regimes shift. More edges = more assets moving in lockstep.
- Clusters merging — during stress, previously independent groups (e.g. tech vs. financials) become connected as sell-offs hit all sectors.
- Density rising — check the statistics panel below. A jump in density from ~20% to ~40%+ typically accompanies elevated fragility.
- Hub nodes growing — some stocks gain many connections, becoming central to the network. These hubs transmit shocks broadly.
These structural shifts are what Ollivier-Ricci curvature quantifies: bridge edges between clusters carry negative curvature, and when bridges proliferate the network's overall curvature drops — signaling rising systemic fragility.
> Statistics will appear here
What to watch: Nodes is the number of assets in the graph (should match your universe). Edges counts how many pairs exceed the correlation threshold — a sudden jump in edges signals a correlation regime shift. Density is the fraction of all possible edges that actually exist; values above 50% suggest the market is moving as a block, which historically coincides with elevated fragility.