Home › Stats-TimeSeries

Vector autoregression — Lodestar.Stats.TimeSeries

One entry point, VectorAutoregression.Fit, which estimates a VAR(p): several series that move together, each explained by every series' own past, with the table statsmodels' VAR(y).fit(p) prints — the coefficients per equation with their standard errors, t statistics and p-values, the residual covariances, the log-likelihood, and the four information criteria.

Why this model and not the others. Decision 0004 read ARIMA, SARIMAX, VAR and state space. The likelihood-fitted three do not reproduce their own answer at the tolerance every corpus here is held to; the VAR is least squares on the stacked lags, which numpy.linalg.lstsq matches at a relative gap of 0.0. Forecasting stays delegated to Microsoft.ML.TimeSeries (decision 0004).

Types

Type	What it is
VectorAutoregression	Fits the model and builds the table.
VarSummary	The coefficients per equation, their inference, and the whole-model numbers.
VarOptions	Whether each equation carries a constant.

See also

• Stationarity tests — what to run before fitting one.
• Serial correlation — the diagnostics that lead to a lag order.
• statsmodels → .NET, Python → C# equivalence.