NICEFA Logo
nicefa.
Library
←Module

Derivation of the Autocorrelation Function (ACF) for a White Noise Process

Derive the Autocorrelation Function (ACF) for white noise. Understand its theoretical properties and intuitive meaning in Time Series Analysis.

The Formal Theorem

Let {Xt}t∈Z \{X_t\}_{t \in \mathbb{Z}} be a discrete-time white noise process, defined by E[Xt]=μ E[X_t] = \mu for all t∈Z t \in \mathbb{Z} , Var(Xt)=σ2>0 \text{Var}(X_t) = \sigma^2 > 0 for all t∈Z t \in \mathbb{Z} , and Cov(Xt,Xs)=0 \text{Cov}(X_t, X_s) = 0 for all t≠s t \neq s . The autocorrelation function (ACF) of Xt X_t , denoted by ρX(k) \rho_X(k) , is formally defined as:
ρX(k)=Cov(Xt,Xt−k)Var(Xt)Var(Xt−k) \rho_X(k) = \frac{\text{Cov}(X_t, X_{t-k})}{\sqrt{\text{Var}(X_t)\text{Var}(X_{t-k})}}
for any integer lag k k . For a white noise process, this simplifies to:
ρX(k)={1if k=00if k≠0 \rho_X(k) = \begin{cases} 1 & \text{if } k=0 \\ 0 & \text{if } k \neq 0 \end{cases}

Analytical Intuition.

Imagine a symphony orchestra where each musician plays their instrument perfectly, but their performances are completely independent. If we measure the sound level at any given moment, we get a value. Now, consider how correlated this measurement is with a measurement taken at a *different* moment. For white noise, because each 'note' is random and uninfluenced by any other, the correlation between measurements at distinct times is precisely zero. Only when we compare a measurement to itself (lag k=0 k=0 ) do we see perfect correlation, a value of 1.
CAUTION

Institutional Warning.

Students often struggle with why the covariance is zero for t≠s t \neq s but variance is non-zero. Remember, white noise implies independence between *distinct* time points, not that the process is inherently zero.

Institutional Deep Dive.

01
The essence of a white noise process lies in its absolute lack of temporal dependence. Think of it as a perfect, unbiased die being rolled at every discrete point in time. Each roll (or observation Xt X_t ) is a brand new, independent random event. The autocorrelation function (ACF) is our tool to quantify how much a time series at one point in time is related to its past or future values. It's essentially a measure of linear dependence between Xt X_t and Xt−k X_{t-k} , normalized by the variance to make it scale-invariant and comparable across different series.
02
Core Logic: The definition of the ACF is ρX(k)=Cov(Xt,Xt−k)Var(Xt)Var(Xt−k) \rho_X(k) = \frac{\text{Cov}(X_t, X_{t-k})}{\sqrt{\text{Var}(X_t)\text{Var}(X_{t-k})}} . For a white noise process, the defining characteristic is that Cov(Xt,Xt−k)=0 \text{Cov}(X_t, X_{t-k}) = 0 for all t≠t−k t \neq t-k , which simplifies to t≠s t \neq s for any pair of distinct time points. This holds true for any non-zero lag k k (i.e., k≠0 k \neq 0 ). When k=0 k=0 , we are comparing Xt X_t with itself. In this case, Cov(Xt,Xt−0)=Cov(Xt,Xt) \text{Cov}(X_t, X_{t-0}) = \text{Cov}(X_t, X_t) , which is the variance of Xt X_t , i.e., σ2 \sigma^2 . The denominator becomes Var(Xt)Var(Xt)=σ2⋅σ2=σ2 \sqrt{\text{Var}(X_t)\text{Var}(X_t)} = \sqrt{\sigma^2 \cdot \sigma^2} = \sigma^2 . Therefore, for k=0 k=0 , ρX(0)=σ2σ2=1 \rho_X(0) = \frac{\sigma^2}{\sigma^2} = 1 .
03
Geometric Mechanics: From a geometric perspective, covariance measures the degree to which two random variables move together. In a high-dimensional space where each dimension represents a time point, a white noise process generates points such that the vector representing Xt X_t is orthogonal to the vector representing Xt−k X_{t-k} for k≠0 k \neq 0 . Orthogonality implies zero correlation (and hence zero covariance for scaled variables). The ACF graph for white noise is a single spike at lag 0 (value 1) and zeros everywhere else, visually depicting this absolute independence across time.
04
Institutional Pitfalls: A common misunderstanding is confusing white noise with *random walks*. A random walk's innovations are white noise, but the random walk itself is highly autocorrelated. Another pitfall is assuming a time series *is* white noise just because its sample ACF plot looks 'flat' or shows no significant spikes beyond lag 0. Rigorous statistical tests (like Ljung-Box) are crucial for formal verification, as visual inspection can be misleading, especially with short series or subtle dependence structures.

Academic Inquiries.

01

What is the difference between white noise and a random walk?

White noise is a sequence of independent and identically distributed random variables with zero mean and finite non-zero variance. A random walk is a process whose increments are white noise, leading to strong autocorrelation in the random walk itself.

02

Does white noise imply stationarity?

Yes, a white noise process is strictly stationary (all moments exist and are constant) and weakly stationary (mean, variance, and autocovariance are constant over time). The defining properties (constant mean, variance, and zero covariance for different times) directly imply stationarity.

03

Can a time series with a non-zero mean be white noise?

The definition often assumes a zero mean for simplicity, but technically, a process Yt Y_t is white noise if Yt=Xt+μ Y_t = X_t + \mu , where Xt X_t is a zero-mean white noise process and μ \mu is a constant. The autocorrelation function is unaffected by the mean.

04

How is the ACF of white noise used in practice?

The ACF of white noise serves as a benchmark. When analyzing a time series, if its sample ACF resembles that of white noise, it suggests the series might be IID (independent and identically distributed), and no further modeling for temporal dependence might be needed. It's also crucial for identifying the order of ARMA models.

Standardized References.

  • Definitive Institutional SourceBrockwell, Peter J., and Richard A. Davis. Introduction to Time Series and Forecasting.

Related Proofs Cluster.

Intermediate

Proof that Autocovariance Depends Only on Lag for Weakly Stationary Processes

Explore the rigorous proof that autocovariance for weakly stationary processes depends only on lag, understanding its deep implications for time series analysis.

Intermediate

Proof of the Stationarity Condition for an AR(1) Process (|φ| < 1)

Unravel the stationarity condition for AR(1) processes. Rigorous proof, cinematic intuition, and essential insights for time series analysis.

Intermediate

Proof of the Invertibility Condition for an MA(1) Process (|θ| < 1)

Exploring the cinematic intuition of Proof of the Invertibility Condition for an MA(1) Process (|θ| < 1).

Intermediate

Derivation of the Mean for a Stationary AR(1) Process

Exploring the cinematic intuition of Derivation of the Mean for a Stationary AR(1) Process.

Institutional Citation

Reference this proof in your academic research or publications.

NICEFA Visual Mathematics. (2026). Derivation of the Autocorrelation Function (ACF) for a White Noise Process: Visual Proof & Intuition. Retrieved from https://www.nicefa.org/library/time-series-analysis/derivation-of-the-autocorrelation-function--acf--for-a-white-noise-process

Dominate the Logic.

"Abstract theory is just a movement we haven't seen yet."

Master the ProofEarly Access