rolling_lasso

rolling_lasso(y, *factors, window, alpha=1.0, max_iter=1000, tolerance=1e-08, name=None, metadata=None)

Return prior-window lasso-regression predictions.

Parameters

y : Panel | Graph

Dependent-variable Panel or single-output Graph. factors : Panel | Graph

One or more factor Panel or single-output Graph inputs. window : int

Positive trailing-window length in rows. alpha : float, default 1.0

Smoothing or regularization parameter, depending on the operation. max_iter : int, default 1000

Maximum coordinate-descent iterations. tolerance : float, default 1e-08

Convergence tolerance for coordinate descent. name : str | None, default None

Optional graph-node name. A generated name is used when omitted. metadata : Mapping[str, Any] | None, default None

Optional metadata stored on the graph node.

Returns

Graph

Lazy single-output graph. Call .compute() to materialize a Panel.

Examples

import pandas as pd

from bagelquant_core import Domain, Panel
from bagelquant_core.composer import rolling_lasso

domain = Domain(calendar=pd.to_datetime(["2024-01-02", "2024-01-03", "2024-01-04"]), universe=["a", "b"])
left = Panel.from_domain(pd.DataFrame({"a": [1.0, 2.0, 4.0], "b": [2.0, 3.0, 8.0]}, index=domain.sessions), domain)
right = Panel.from_domain(pd.DataFrame({"a": [1.0, 1.0, 2.0], "b": [1.0, 2.0, 4.0]}, index=domain.sessions), domain)

result = rolling_lasso(left, right, window=2).compute().data
print(result)

Notes

Inputs are aligned by index and columns before the operation runs.

Rolling calculations run independently down each asset column.

The model is fit on prior rows only and predicts the current row.