The University of Hong Kong · FACULTY OF STATISTICS

STAT8003 Chap.3 Stationarity, White Noise and Differencing

- one subject, every graph, every model, every mark
6 Chapters3-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 3 of 8 · STAT8003

Stationarity, White Noise and Differencing

Compare distributional and second-order stability

This chapter compares strict stationarity with the weak definition used in later calculations. Strict stationarity preserves every finite joint distribution after shifting all time indices. Weak stationarity requires a constant mean and an autocovariance determined only by lag.

Neither should be inferred from a smooth-looking plot.

Use moments only when they exist

Finite second moments connect the two definitions. A strictly stationary Cauchy sequence lacks the required moments and therefore does not qualify as weakly stationary. White noise with finite variance supplies the reference process: constant zero mean, constant variance and zero autocorrelation at every nonzero lag.

The shorter gamma-k and rho-k notation is used only after weak stationarity has been established.

Transform by showing what remains

A random walk is non-stationary in level because its variance grows over time and its covariance depends on more than lag. First differencing subtracts the previous level and leaves the current innovation. The conclusion is therefore algebraic, not visual.

The transformed mean and covariance should be stated explicitly so the reason for stationarity can be checked.

In this chapter

What this chapter covers

  • 01

    Strict stationarity

  • 02

    Weak stationarity

  • 03

    Finite-moment condition

  • 04

    White noise

  • 05

    Non-stationary random walk

  • 06

    First differencing

Worked example · free

Classify a random constant process

Q [7 marks]. AskSia-authored practice. The marks shown are an AskSia study allocation, not a University marking scheme. Let Yt=X for every t, where X has finite mean and variance. Classify the process and find its autocovariance.
Stationarity concerns probability laws, not a tidy plot.Stationarity concerns probability laws, not a tidy plot. A stable mean and lag covariance can coexist with noisy movement, while smooth drift can violate both.constant momentschanging level or spreadstationarynon-stationary
Stationarity concerns probability laws, not a tidy plot. A stable mean and lag covariance can coexist with noisy movement, while smooth drift can violate both.
  • 3Check shifted joint distributions.
  • 2Check constant first and second moments.
  • 2State the autocovariance at every lag.
Every finite vector equals repeated copies of X, so shifting indices changes no joint distribution and the process is strictly stationary. Its mean is constant and Cov(Yt,Yt-k)=Var(X) for every k, so it is weakly stationary as well.
Sia tip — A flat realised path can come from perfect dependence; it does not mean the model contains no randomness.
Glossary

Key terms

Strict Stationarity
Invariance of every finite joint distribution under a common time shift. This distributional property does not itself guarantee the finite means and variances required for second-order analysis.
Weak Stationarity
A constant mean and an autocovariance depending only on lag. It permits the shorter gamma-k and rho-k notation and supports stable estimation of second-order dependence from one record.
First Difference
The transformed value Zt minus its immediately preceding value. It removes a random-walk accumulation to expose the innovation and converts a deterministic linear trend into a constant contribution.
FAQ

Stationarity, White Noise and Differencing FAQ

Can a process be strictly stationary but not weakly stationary?

Yes. Strict stationarity does not guarantee that the mean and variance exist. An independent Cauchy sequence preserves its joint laws under shifts but lacks the finite moments required by weak stationarity.

What is the ACF of white noise?

Under the course definition, white noise has correlation one at lag zero and zero at every nonzero lag. This pattern becomes the target for residuals after a fitted model has removed systematic dependence.

Why does differencing a random walk work?

The random-walk equation adds a new innovation to the previous level. Subtracting that previous level leaves exactly the current innovation, whose mean, variance and nonzero-lag autocovariances are stable across time.

Can a time plot establish stationarity by itself?

No. A plot can suggest drift, changing variance or seasonal behaviour, but stationarity is a property of probability laws or moments. A short non-stationary record may look calm, and a stationary process may generate an unusual run.

Use the stated model, justified transformations and formal evidence to support the verdict, then write the constant-mean and lag-covariance conditions explicitly before adopting stationary notation.

When may gamma-k notation be used?

Use gamma-k only after weak stationarity has made autocovariance a function of lag rather than the two calendar times separately. Before that point, keep gamma(t,s). The notation change is evidence that the stationary assumption has been invoked.

Study strategy

Exam move

For each classification, write the definition first and check every clause. Do not infer stationarity from the appearance of a plot. When a transformation is proposed, derive the transformed process symbolically and state its mean and covariance properties. Keep strict and weak conclusions separate unless finite moments or a multivariate normal family justify the connection. Make a two-column comparison sheet.

Under strict stationarity, write shifted finite joint distributions; under weak stationarity, write constant mean and lag-only covariance. Add a warning that strict stationarity needs finite second moments before it implies weak stationarity. Test the sheet on independent Cauchy noise, finite-variance white noise, a random constant process and a random walk.

For each, give a one-sentence verdict and point to the exact condition that passes or fails. Then practise transformations algebraically. Difference a deterministic trend plus noise and a random walk, and calculate the transformed mean, variance and lag-one covariance rather than saying only that the data look more stable.

When a plot is supplied, use it to suggest a problem but reserve the mathematical verdict for the stated process or formal evidence. A stationary process can produce an unusual path, and a non-stationary record can look calm over a short window. During the final check, inspect the notation.

The symbols gamma k and rho k already assume weak stationarity; if the conditions have not been established, retain the two-time notation gamma(t,s) and rho(t,s). That discipline prevents the answer from assuming the property it is meant to test.

Working through Stationarity, White Noise and Differencing in STAT8003? Sia is AskSia’s AI Statistics tutor — ask any STAT8003 Stationarity, White Noise and Differencing question and get a clear, step-by-step explanation grounded in how STAT8003 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 52 of your subjects at The University of Hong Kong, plus 1,000+ more Bibles across the AskSia library.
Sia - your STAT8003 tutor, unlimited, worked the way the exam marks it
The full 3-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works