Frequently Asked Questions

Rolling correlation measures how two assets move together over a sliding time window. A value of +1 means they move in perfect lockstep, -1 means they move in opposite directions, and 0 means no linear relationship. By "rolling" the window forward day by day, you can see how the relationship between two assets changes over time — revealing regime shifts, contagion events, and diversification opportunities that a single static number would miss.

Select a built-in watchlist (like TWSE Top 50 or S&P 500 Top 50) from the landing page, or build your own by entering ticker symbols. The tool fetches daily price data, computes pairwise rolling correlations, and generates interactive charts. You can filter by sector, search for specific pairs, play back the correlation surface over time, and explore the correlation network graph.

The window is the number of trailing trading days used to compute each correlation value. A 30-day window captures roughly six weeks of market activity. Shorter windows (20d) react faster to regime changes but are noisier; longer windows (60d, 90d) are smoother but lag behind structural breaks. The default 30-day window balances responsiveness and stability.

Beta adjustment removes the common market factor from each stock's returns before computing correlations. Without adjustment, most stocks in the same market show high correlation simply because they all move with the index. Beta-adjusted (residual) correlations reveal the idiosyncratic co-movement between two stocks — the relationship that persists after stripping out the broad market influence. This is the "Defensive" mode in the Cluster Finder.

The Cluster Finder searches all possible combinations of k stocks (default 3) to find the group with the highest average pairwise correlation. It has three modes: Off (raw correlations), Defensive (beta-adjusted residual correlations — finds stocks correlated beyond market exposure), and Offensive (raw correlation weighted by mean beta — finds highly correlated stocks that also amplify market moves).

The Correlation Network is a force-directed graph where each node is a stock and edges represent strong correlations. Thicker edges mean higher correlation. Nodes are colored by sector and sized by degree centrality (how many strong connections a stock has). Community detection (Louvain algorithm) groups tightly connected clusters, making it easy to spot structural relationships across the market.

The 3D surface visualizes how the cross-sectional distribution of all pairwise correlations evolves over time. The x-axis is time, the y-axis is the correlation value, and the z-axis (height/color) is the density of pairs at that correlation level on that date. Peaks indicate where most pairs cluster. The surface makes regime shifts visible — you can see the entire market's correlation structure compress during crises or fan out during calm periods.

When building a custom watchlist, you can opt to share it with the community by checking "Share to community" and entering an author name. Shared watchlists appear in a collapsible section on the landing page so other users can explore them. Community watchlists expire after 180 days of inactivity; private (unshared) custom watchlists expire after 30 days.

All price data is sourced from Yahoo Finance (yfinance). Built-in watchlist constituents are either scraped from Wikipedia or maintained as curated static lists updated periodically. Sector classifications come from the GICS (Global Industry Classification Standard) as reported by Yahoo Finance. The tool fetches up to 10 years of daily price history per ticker.