Regression and correlation

REGR, regression analysis

The REGR panel runs OLS, Ridge, or Lasso regression of a dependent series on one or more independent series. Outputs:

• Coefficients with confidence intervals.
• R², variance explained.
• Scatter with the fit line overlaid.
• Rolling beta, the coefficient on a single independent variable computed across rolling windows. Reveals when the relationship is stable vs when it's shifted.
• Diagnostics, residual plot, residual ACF, Q-Q plot for the standard regression sanity checks.

CORR, pairwise correlation

The CORR panel computes the rolling correlation between two series and plots it over time. Simpler than REGR, no fit, just the correlation coefficient. Use when you want to read the strength of a relationship without modelling it.

CORRMAT, correlation matrix

The CORRMAT panel renders the full pairwise correlation matrix across the top-N tracked coins, computed on log returns over a configurable window. Visualised as a heatmap.

Reads to look for:

• Tight cluster, most coins highly correlated. The market is moving as one block. Beta-driven tape; alt selection matters less than direction.
• Decoupling, most coins moderately correlated, one or two outliers. The outliers are where idiosyncratic flow is happening.
• Sector-level structure, meme coins correlating tightly with each other, L1 chains in their own cluster, etc. Useful for thinking about basket trades.

PCA, principal components

The PCA panel runs principal component analysis on the universe's log returns. The first principal component (PC1) is "market beta", almost every coin loads on it positively. PC2 and PC3 are second-order factors: typically "risk-on / risk-off", "altcoin season vs BTC dominance", "DeFi vs L1".

Outputs:

• Explained variance per component, how much of the universe's variation each PC accounts for.
• Loadings, each coin's exposure to each PC. The factor structure of the universe.
• PC time-series, each PC plotted over time. PC1 looks like the market; PC2/PC3 are where the interesting research questions live.

Used most often as a research tool, "is meme-coin rotation showing up as a distinct factor today?", rather than a daily read.