Home › Stats-TimeSeries › Stationarity tests

LagSelection

How the augmented Dickey-Fuller test chooses its lag order.

public enum LagSelection

Fields — Akaike takes the smallest Akaike criterion, the reference's "AIC" and the default. Schwarz takes the smallest Schwarz criterion, "BIC". TStatistic walks down from the maximum and stops at the first lag whose own t statistic reaches 1.645, "t-stat". Fixed searches nothing and uses the maximum lag as the lag, autolag=None.

Example — a fixed lag reports no criterion.

using Lodestar.Stats.TimeSeries;

double[] walk = [0.0, 1.2, 0.7, 2.1, 3.0, 2.4, 3.9, 5.1, 4.6, 6.0, 7.3, 6.8,
                 8.2, 9.5, 9.1, 10.4, 11.8, 11.2, 12.7, 14.0, 13.5, 14.9, 16.2, 15.8];

DickeyFullerResult fixedLag = Stationarity.AugmentedDickeyFuller(
    walk, new DickeyFullerOptions { LagSelection = LagSelection.Fixed, MaxLag = 1 });

int usedLag = fixedLag.UsedLag;                                     // => 1
bool noCriterion = double.IsNaN(fixedLag.InformationCriterion);     // => True

Remarks — a tie between two criteria keeps the shorter lag, as the reference's min over (criterion, lag) pairs does.

Applies to — net10.0, netstandard2.0.

See also — DickeyFullerOptions, DickeyFullerResult.