rols 0.2.0

Vectorized rolling and expanding regression for multi-target, multi-factor time series.

rOLS

Vectorized rolling and expanding regression for multi-target, multi-factor time series.

Built for performance at scale: hundreds of regressors against hundreds of targets, over thousands of time steps, without loops.

Adapted for applications where dynamic relationships matter most: estimating rolling betas in finance to isolate idiosyncratic sensitivity to narrative factors; tracking time-varying price elasticities in economics to capture structural shifts; attributing regional temperature anomalies in climate science by orthogonalizing forcing factors; and adaptively filtering signals in real time. Designed for speed and scalability, it enables precise, loop-free analysis across domains where traditional methods fall short.

Metric	Value
PyPI Version
Python Versions
License
Downloads
GitHub Stars

---

Overview

rOLS estimates time-varying sensitivities (betas) between a set of factors and a set of targets using a rolling or expanding window. At each time step, it fits a fresh regression on the most recent observations, giving you a full time series of betas, signals, and R² for each (factor, target) pair.

It supports:

• OLS and Ridge regression (controlled by a single lambda_ parameter)
• Multiple controls — partialled out via Frisch-Waugh-Lovell, keeping the per-factor math univariate regardless of how many controls you add
• Rolling Gram-Schmidt orthogonalization — factors and/or controls can be orthogonalized within their group before estimation, so each beta reflects incremental explanatory power
• HAC standard errors — Newey-West robust SEs, computed on demand
• Expanding windows as an alternative to fixed rolling windows
• Lagged signals to avoid look-ahead bias

---

Installation

pip install rols

Requires Python 3.10+ and numpy / pandas.

---

Quick start

import pandas as pd
import pandas_datareader as pdr
import pandas_datareader.data as web
from rols import RollingOLS

# Loading some factors
factor_df = web.DataReader(['CPIAUCSL', 'CPILFESL'], 'fred', start=start, ).pct_change().dropna()
factors = factor_df.columns.tolist()

# Loading some targets
asset_df = web.DataReader('12_Industry_Portfolios', 'famafrench', start=start, )[1]
asset_df.index = asset_df.index.to_timestamp()
assets = asset_df.columns.tolist()

# Loading some controls
control_df = pdr.get_data_famafrench("F-F_Research_Data_Factors", start=start, )[0].div(100.0).drop(columns=["RF"])
control_df.index = control_df.index.to_timestamp()
controls = control_df.columns.tolist()

# Merge data into one dataframe aligned by date
df = pd.merge(factor_df, asset_df, left_index=True, right_index=True, how='left').ffill()
df = pd.merge(df, control_df, left_index=True, right_index=True, how='left').ffill()

# Running the roling regression
ols = RollingOLS(window=12, expanding=False, lambda_=0.0)
ols.fit(factors=df[factors], controls=df[controls], orthogonalize_controls=True, orthogonalize_factors=True)
result = ols.transform(assets=df[assets])

# Plot some results
for f in factors:
    result.get_beta(f).plot(title=f)

---

API

RollingOLS(...) — constructor

Parameter	Default	Description
window	252	Rolling window length
min_periods	window	Minimum observations to produce a result
expanding	False	Use expanding window instead of rolling
lambda_	0.0	Ridge regularization. 0 = standard OLS
ewma_halflife	None	Exponentially weight observations within each window (half-life in periods). None = equal weighting. Not compatible with expanding=True
adj_r2	False	Compute adjusted R²
lag_signal	False	Use beta_{t-1} * factor_t instead of beta_t * factor_t
hac_lags	None	Newey-West lags for HAC SE. None disables HAC
dtype	"float32"	DataFrame storage dtype (see note below). Use "float64" for higher precision
asset_chunk_size	100	Controls peak memory during residualization
warn_singular	True	Warn (RuntimeWarning) on singular windows. Set False to suppress

---

.fit(factors, controls=None, orthogonalize_factors=False, orthogonalize_controls=False)

Fits the model on the regressors side. Residualizes factors against controls (Frisch-Waugh step 1) if controls are provided.

# No controls
ols.fit(df[["f1", "f2", "f3"]])

# With controls
ols.fit(df[["f1", "f2"]], controls=df[["ctrl1", "ctrl2"]])

# With rolling orthogonalization
# f1 is untouched, f2 is orthogonalized against f1, f3 against f1 and f2
ols.fit(df[["f1", "f2", "f3"]], orthogonalize_factors=True)

Column order matters for orthogonalization — place higher-priority factors first. Each factor's beta will then reflect its incremental contribution beyond all preceding ones.

---

.transform(targets)

Projects targets onto the fitted factor structure and returns a RollingOLSResult.

result = ols.transform(df[["y1", "y2", "y3"]])

The fitted model can be reused on different target sets without re-fitting:

ols.fit(df[["f1", "f2"]], controls=df[["ctrl1"]])
result_a = ols.transform(df[group_a])
result_b = ols.transform(df[group_b])

---

.fit_transform(factors, targets, controls=None, ...)

Convenience one-liner when you don't need to reuse the fitted model.

result = RollingOLS(window=60).fit_transform(
    df[["f1", "f2"]],
    df[["y1", "y2"]],
    controls=df[["ctrl1"]],
    orthogonalize_factors=True,
)

---

RollingOLSResult — getters

All results are indexed by time (rows) and target (columns).

result.get_beta("f1")          # DataFrame (T x N_targets)
result.get_signal("f1")        # beta_t * factor_t (or lagged)
result.get_r2("f1")            # rolling R²
result.get_residuals("f1")     # regression residuals (FWL step 3)
result.get_factor_adjusted_returns()    # controls removed only (FWL step 2)

result.get_se("f1")            # Newey-West SE — requires hac_lags
result.get_tstat("f1")         # beta / SE

result.get_control_beta("f1", "ctrl1")  # requires return_control_betas=True

get_factor_adjusted_returns() returns asset returns with only the controls partialled out (e_it = r_it - B_t' * ctrl_t, FWL step 2) — not specific to any factor, so it takes no argument. This differs from get_residuals(factor), which additionally removes the narrative factor (FWL step 3). If no controls were provided at fit(), it returns the original asset returns.

get_control_beta(factor, control) returns the control's joint rolling beta — its coefficient from the full regression, recovered via Frisch-Waugh-Lovell (each control partialled out against all other controls), not a univariate marginal beta. The value does not depend on factor (control betas are shared across factors). Requires return_control_betas=True on transform()/fit_transform().

Long format — useful for downstream analysis, filtering, or plotting:

result.to_long("f1")                    # date, target, beta, signal, r2
result.to_long("f1", include_se=True)   # + se, t_stat
result.to_long_all()                    # all factors stacked

---

Examples

Ridge regression

# lambda_ > 0 adds λI to X'X before solving
# stabilizes estimation when factors are correlated
ols = RollingOLS(window=120, lambda_=1e-3)
result = ols.fit(df[["f1", "f2", "f3"]]).transform(df[targets])

HAC standard errors

import numpy as np

# Common rule of thumb for lag selection: floor(T^(1/3))
hac_lags = int(np.floor(len(df) ** (1/3)))

ols = RollingOLS(window=120, hac_lags=hac_lags)
result = ols.fit(df[["f1", "f2"]]).transform(df[targets])

se    = result.get_se("f1")      # Newey-West SE
tstat = result.get_tstat("f1")  # t-statistics

EWMA observation weighting

By default every observation in a window counts equally. When recent data should carry more weight — e.g. narrative-beta estimation in finance, where the latest behaviour matters most — set ewma_halflife to weight observations exponentially. An observation ewma_halflife periods in the past gets half the weight of the most recent one.

# ~3-month half-life inside a 1-year window
ols = RollingOLS(window=252, ewma_halflife=63)
result = ols.fit(df[["f1", "f2"]]).transform(df[targets])

The weighting flows through the betas, R², and the Frisch-Waugh residualization (weighted least squares per window). NaN rows are dropped per window and the surviving weights renormalized to sum to 1, so missing data does not distort the scheme. ewma_halflife cannot be combined with expanding=True (an expanding window has no fixed length to precompute weights over), and HAC standard errors are still computed with equal weights.

Orthogonalization with importance ordering

When factors have a natural priority order, orthogonalization ensures each beta measures incremental contribution beyond higher-priority factors.

# f1 is the primary factor — left untouched
# f2 is orthogonalized against f1
# f3 is orthogonalized against f1 and f2
ols = RollingOLS(window=120)
ols.fit(
    df[["f1", "f2", "f3"]],
    orthogonalize_factors=True,
)
result = ols.transform(df[targets])

Expanding window

ols = RollingOLS(window=30, min_periods=30, expanding=True)
result = ols.fit(df[["f1"]]).transform(df[targets])

Lagged signal (avoiding look-ahead)

# beta estimated at t-1, multiplied by factor at t
ols = RollingOLS(window=60, lag_signal=True)
result = ols.fit(df[["f1"]]).transform(df[targets])
signal = result.get_signal("f1")

---

Design notes

Frisch-Waugh-Lovell — when controls are provided, rOLS residualizes both factors and targets against controls before running the per-factor regression. This is mathematically equivalent to the full joint regression but keeps the inner loop univariate, making it fast regardless of how many controls are added.

Stride tricks — the rolling window matrix operations use numpy.lib.stride_tricks.as_strided to build zero-copy sliding window views, avoiding explicit loops over time for the fixed-window case.

Memory — asset residualization is chunked (asset_chunk_size) to bound peak memory when the number of targets is large. Reduce this value if you hit memory limits.

Precision (dtype) — dtype controls the storage precision of the input and intermediate pandas DataFrames only. Internal matrix operations (gram matrix accumulation and the linear solve) always run in float64 regardless of this setting, because np.linalg.solve loses accuracy in float32 for ill-conditioned windows. So float32 reduces DataFrame memory but does not change the numerical precision of the regression itself.

HAC caching — standard errors are computed lazily and cached on first call to get_se(). Calling it multiple times for the same factor incurs no extra cost.