top of page

Testing for White Noise: A Comprehensive Guide

Jan 23, 2024

White noise, a term widely used in the fields of mathematics, statistics and engineering, refers to a random signal or process consisting of equal intensity at varying frequencies, ultimately creating a constant power spectral density. Given its vast application, from time series analysis to sound masking and relaxation, a pertinent question arises: ‘How to test for white noise?’ This article outlines various methods of testing.

Visual Inspection

The simplest method of initial analysis, a visual inspection of the time plot and the autocorrelation function (ACF) plot, can help determine whether the data set is white noise. A random and non-repeating time plot indicates a white noise process, while ACF plot values should be evenly distributed around zero.

Statistical Tests

To provide a more rigorous and definitive conclusion, employing statistical tests is necessary. The most commonly applied tests are the Ljung-Box (LB) test, Bartlett’s test, and the Turning Point (TP) test.

  1. Ljung-Box (LB) Test: This test examines whether there is significant autocorrelation in the data set by comparing the observed ACF values with those expected under white noise conditions. A high p-value (typically >0.05) signifies the null hypothesis (i.e., the data is white noise) cannot be rejected.

  2. Bartlett’s Test: Utilized for testing the homogeneity of variances, Bartlett’s test ensures the data is randomly sampled. It tests the null hypothesis that all input data originates from populations with equal variances. A high p-value indicates white noise.

  3. Turning Point (TP) Test: The TP test focuses on the randomness of sign changes in successive values of the data set. If the number of turning points is within the expected range for a random process, the data may be treated as white noise.

Model Fitting

An alternative approach entails fitting an Autoregressive Integrated Moving Average (ARIMA) model to the data set. If the model displays no significant AR and MA terms (i.e., ARIMA (0, 0, 0)), the data can be considered white noise.

In conclusion, exploring a combination of visual inspection, statistical tests, and model fitting techniques is vital to effectively test for white noise. Ultimately, this allows for appropriate decision-making when applying white noise concepts in various real-world applications.

bottom of page