gs_quant.timeseries.statistics.RollingLinearRegression

class RollingLinearRegression(X, y, w, fit_intercept=True)[source]

Fit a rolling ordinary least squares (OLS) linear regression model.

Parameters:

  • X (Union[Series, List[Series]]) – observations of the explanatory variable(s)

  • y (Series) – observations of the dependant variable

  • w (int) – number of observations in each rolling window. Must be larger than the number of observations or explanatory variables

  • fit_intercept (bool) – whether to calculate intercept in the model

Usage

Fit OLS Model based on observations of the explanatory variables(s) X and the dependant variable y across a rolling window with fixed number of observations. The parameters of each rolling window are stored at the end of each window. If X and y are not aligned, only use the intersection of dates/times.

Examples

Run linear regressions on y vs. x1 and x2 in a rolling window of 22 observations and compute the R Squared:

>>> x1 = generate_series(100)
>>> x2 = generate_series(100)
>>> y = generate_series(100)
>>> r = RollingLinearRegression([x1, x2], y, 22)
>>> r.r_squared()

Methods

coefficient(i)

Estimated coefficients.

fitted_values()

Fitted values at the end of each rolling window.

r_squared()

Coefficients of determination (R Squared) of rolling regressions.

standard_deviation_of_errors()

Standard deviations of the error terms.