ar1_residual_sd#
- impulso.ar1_residual_sd(y)[source]#
Per-variable residual standard deviation of a univariate AR(1)-with-constant fit.
For each column of y independently, fits y[t] = c + phi * y[t-1] + e[t] by ordinary least squares over the T - 1 observations t = 1, …, T - 1, then returns the residual standard deviation sqrt(sum(e**2) / dof) with dof = max(T - 1 - 2, 1) — the T - 1 fitted observations less the 2 estimated parameters (intercept and own-lag coefficient), floored at 1 so the computation stays well-defined for very short series. This is the sigma scale consumed by minnesota_dummies, and the scale in which Impulso’s other data-dependent priors (e.g. VAR.exog_prior_scale) are defined; see docs/adr/0012-exog-prior-scales-with-data.md.