regime_features¶

Defined in fynance.features.regime

regime_features(X, w=21, period=252)[source]

Build the causal regime feature matrix: trailing vol and mean return.

Both columns use trailing windows (past only), so the matrix is causal — the value at t depends on X[:t+1] only.

Parameters:

Xnp.ndarray: One-dimensional price/level series.
wint: Rolling window.
periodint: Annualization factor for the volatility column.

Returns:

np.ndarray: Shape (len(X), 2) — [realized_volatility, rolling_mean_log_return].

Notes

Row 0 is a deterministic warmup point: there is no prior observation, so both the realized volatility and the mean log-return are 0 (an artificially “ultra-calm” point). It is kept so the matrix stays aligned with X; callers fitting a clustering on the matrix may wish to drop or mask this row.