Stats TimeSeries trendterms - CyrilB1531/lodestar GitHub Wiki

HomeStats-TimeSeriesStationarity tests

TrendTerms

The deterministic terms a unit-root or stationarity regression carries.

public enum TrendTerms

FieldsNone is no constant and no trend, the reference's "n", for the augmented Dickey-Fuller test only. Constant is a constant, "c", the default of both tests. ConstantAndTrend adds a linear trend, "ct". ConstantAndQuadraticTrend adds a linear and a quadratic trend, "ctt", for the augmented Dickey-Fuller test only.

Example — the same series against a quadratic trend, where the unit root is rejected outright.

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 quadratic = Stationarity.AugmentedDickeyFuller(
    walk, new DickeyFullerOptions { Regression = TrendTerms.ConstantAndQuadraticTrend });

double statistic = Math.Round(quadratic.Statistic, 4);  // => -15.9543
int usedLag = quadratic.UsedLag;                        // => 1

Remarks — each choice reads its own MacKinnon table, so the same statistic means different things under different terms. KpssOptions.Regression refuses the two the KPSS test does not define.

Applies to — net10.0, netstandard2.0.

See alsoDickeyFullerOptions, KpssOptions.