Home › Stats-TimeSeries › Stationarity tests

DickeyFullerOptions

What an augmented Dickey-Fuller test may be told.

public sealed record DickeyFullerOptions

Properties — Regression is the regression's deterministic terms, a TrendTerms; Constant by default. LagSelection is how the lag order is chosen, a LagSelection; Akaike by default. MaxLag is the largest lag the search considers, or the lag itself under LagSelection.Fixed; null by default, which takes Schwert's ceil(12·(n/100)^¼) capped at n/2 − terms − 1.

Exceptions — ArgumentOutOfRangeException when MaxLag is negative, or Regression or LagSelection is a value its enum does not declare, checked where the value is set.

Example — Schwarz's criterion picks the same lag as Akaike's on this series, and reports its own value.

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 schwarz = Stationarity.AugmentedDickeyFuller(
    walk, new DickeyFullerOptions { LagSelection = LagSelection.Schwarz });

int usedLag = schwarz.UsedLag;                                  // => 3
double criterion = Math.Round(schwarz.InformationCriterion, 4);  // => -34.6446

Remarks — without a constant the table changes too: TrendTerms.None reads MacKinnon's no-constant surface, which has no upper cut-off, so a large positive statistic reads a p-value close to 1 rather than exactly 1.

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 bare = Stationarity.AugmentedDickeyFuller(
    walk,
    new DickeyFullerOptions { Regression = TrendTerms.None, LagSelection = LagSelection.Fixed, MaxLag = 2 });

double statistic = Math.Round(bare.Statistic, 4);            // => 3.9128
double p = Math.Round(bare.PValue, 4);                       // => 1
double onePercent = Math.Round(bare.CriticalValues[0], 4);   // => -2.6804

Being a record of value types, two option sets with the same three values are equal.

Applies to — net10.0, netstandard2.0.

See also — Stationarity.AugmentedDickeyFuller, DickeyFullerResult, the Python equivalence table.