Home › Stats-TimeSeries › Serial correlation

AutocorrelationResult

An autocorrelation sequence and the band around it, indexed by lag.

public sealed class AutocorrelationResult

Properties — Values is the correlation at each lag, from 0 — where it is always 1 — upwards. ConfidenceLower is the lower end of the band, centred on Values rather than on zero. ConfidenceUpper is the upper end.

Example — a correlogram: the value at each lag, bracketed by its own band.

using Lodestar.Stats.TimeSeries;

double[] series = [1.0, 3.0, 2.0, 5.0, 4.0, 7.0, 6.0, 9.0, 8.0, 11.0, 10.0, 13.0];

AutocorrelationResult result = SerialCorrelation.Autocorrelation(series, lagCount: 4);

int lagCount = result.Values.Count;                        // => 5
double lag2 = Math.Round(result.Values[2], 4);             // => 0.568
double lag2Lower = Math.Round(result.ConfidenceLower[2], 4);  // => -0.1701

Remarks — there is no public constructor. A result is what SerialCorrelation.Autocorrelation and SerialCorrelation.PartialAutocorrelation return, never something a caller assembles by hand — the three lists are what a real computation produces, not a promise a constructor could leave broken.

A class rather than a record: a record's equality would compare these three lists by reference, so two results holding the same numbers would compare unequal (#668). Nobody compares two correlograms, so this promises no equality at all rather than a broken one.

Applies to — net10.0, netstandard2.0.

See also — SerialCorrelation.Autocorrelation, SerialCorrelation.PartialAutocorrelation, AutocorrelationOptions, the Python equivalence table.