Home › Stats-TimeSeries › Stationarity tests

DickeyFullerResult

An augmented Dickey-Fuller test: the statistic, its MacKinnon p-value and what the regression used.

public sealed class DickeyFullerResult

Properties — Statistic is the lagged level's t statistic in the chosen regression. PValue is MacKinnon's (1994) approximate p-value against the null of a unit root. UsedLag is the lag order the regression was fitted at. ObservationCount is how many rows that regression fitted: the series length less the lag, less one. CriticalValues holds MacKinnon's (2010) critical values at 1 %, 5 % and 10 %, for ObservationCount. InformationCriterion is the winning criterion of the lag search, NaN under LagSelection.Fixed.

Example — the statistic sits above every critical value, so no level rejects a unit root.

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 result = Stationarity.AugmentedDickeyFuller(walk);

double tenPercent = Math.Round(result.CriticalValues[2], 4);     // => -2.6507
bool rejectsAtTen = result.Statistic < result.CriticalValues[2];  // => False

Remarks — there is no public constructor. A result is what Stationarity.AugmentedDickeyFuller returns. Under LagSelection.TStatistic, InformationCriterion is the absolute t statistic of the last lag tried, as the reference's icbest is.

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 result = Stationarity.AugmentedDickeyFuller(
    walk, new DickeyFullerOptions { LagSelection = LagSelection.TStatistic });

double lastLagT = Math.Round(result.InformationCriterion, 4);  // => 2.442

A class rather than a record, for the reason AutocorrelationResult gives: a record's equality would compare CriticalValues by reference.

Applies to — net10.0, netstandard2.0.

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