The University of Manchester · FACULTY OF ECONOMICS

ECON31031 Chap.6 Time Series Foundations and Stationarity

- one subject, every graph, every model, every mark
7 Chapters3-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 6 of 9 · ECON31031

Time Series Foundations and Stationarity

Time Series Foundations and Stationarity opens Part II of the module, which follows on from the time series material in the prerequisite econometrics module. A time series is one realisation of an unknown process observed at a fixed frequency, with a single value at each date.

The lag operator shifts a series back in time and the difference operator, equal to one minus the lag operator, measures change, including seasonal differences such as annual changes in quarterly data. Autocovariances and autocorrelations measure dependence across dates, and the correlogram plots their sample versions.

White noise, with zero mean, constant variance and no autocorrelation, is the basic building block, and weak stationarity, constant finite mean, variance and lag-dependent autocovariances, is the assumption that lets one sample path stand for the process.

Several visual tools come before any model: time plots of levels, logs, differences or percentage changes, correlograms of sample autocorrelations and partial autocorrelations, and histograms of residuals. In R a single display command produces the time plot, ACF and PACF together, which makes a quick first check of stationarity routine.

In this chapter

What this chapter covers

  • 01

    Univariate time series and sampling frequency

  • 02

    Lag operators and lag polynomials

  • 03

    First, seasonal and second differences

  • 04

    Autocovariance, autocorrelation and the sample ACF

  • 05

    Time plots, correlograms and the PACF

  • 06

    White noise

  • 07

    Weak (covariance) stationarity

Worked example · free

Compute a first-order sample autocorrelation

Q [4 marks]. A short series takes the values 2, 4, 3, 5 and 1. Compute its first differences and its sample autocorrelation at lag 1. This practice weighting is assigned by AskSia and is not an official assessment scheme.
Sixty draws of a white noise process with mean zero and unit variance, drawn here from a normal distribution.Sixty draws of a white noise process with mean zero and unit variance, drawn here from a normal distribution. The level never drifts, the spread stays constant, and 59 of the 60 values sit inside the shaded band two standard deviations either side of zero.mean 0+2σ−2σt = 1t = 60time index
Sixty draws of a white noise process with mean zero and unit variance, drawn here from a normal distribution. The level never drifts, the spread stays constant, and 59 of the 60 values sit inside the shaded band two standard deviations either side of zero.
  • 2First differences are 2, -1, 2 and -4. The mean of the series is 3, so the deviations are -1, 1, 0, 2 and -2 with squares summing to 10.
  • 2Products of neighbouring deviations are -1, 0, 0 and -4, summing to -5, so the lag-1 autocorrelation is -5/10 = -0.5.
The first differences are 2, -1, 2 and -4, and the lag-1 sample autocorrelation is -0.5.
Sia tip — Divide by the sum of all squared deviations, not by the number of products; the denominator is the same at every lag.
Glossary

Key terms

Lag operator
The operator that shifts a time series back by one period, so that L applied to y at t gives y at t minus 1.
Correlogram
A bar chart of sample autocorrelations plotted against the lag.
Seasonal difference
The change between a value and the value from the same season one year earlier, such as four quarters back.
FAQ

Time Series Foundations and Stationarity FAQ

What makes a process weakly stationary?

Three conditions must hold at every date: the mean is finite and constant, the variance is finite and constant, and the covariance between two values depends only on how many periods separate them. White noise is stationary with every autocorrelation beyond lag zero equal to zero.

Does white noise have to be normally distributed?

No. The definition only requires a zero mean, a constant variance and no correlation between different dates. Normal draws are a convenient example, and the ARMA likelihood assumes normality, but the white noise definition itself says nothing about the distribution.

Why is stationarity such a useful assumption?

If the distribution is the same at every date, the T observations can be treated as T draws from one distribution, which makes estimation possible from a single sample path. The supplementary reading warns that the assumption can fail, for example when an average level changes across decades.

What is the difference between the lag and difference operators?

The lag operator shifts a series back one period, so it returns the previous value. The difference operator subtracts that previous value from the current one, so it equals one minus the lag operator; seasonal differences subtract the value from the same season a year earlier.

How is a sample autocorrelation calculated?

Subtract the mean from every value, multiply each deviation by the deviation k periods earlier, and add those products. Divide the total by the sum of all squared deviations. The same denominator is used at every lag, so longer lags have fewer products but no smaller divisor.

What is a correlogram used for?

It plots sample autocorrelations against the lag, showing at a glance how strongly a series is related to its own past. Slow decay suggests strong persistence or nonstationarity, while values close to zero beyond lag zero are what white noise should produce.

Study strategy

Exam move

Learn the operator notation by applying it to short numerical series: lags, first differences, seasonal differences and second differences. Compute sample autocorrelations by hand for an eight-value series until the denominator rule is automatic.

Write the stationarity definition from memory with all three conditions and the word finite, then practise judging plotted series and correlograms for wandering levels, widening spread or slowly decaying autocorrelations. Draw two or three series yourself, a white noise sequence, a trending series and a seasonal one, and sketch the correlogram you would expect for each before checking with R.

Learning to predict the plot from the process is the fastest preparation for the identification work in the next chapter. Keep the three stationarity conditions on your notes sheet in exactly the form you would write them in an answer.

Working through Time Series Foundations and Stationarity in ECON31031? Sia is AskSia’s AI Economics tutor — ask any ECON31031 Time Series Foundations and Stationarity question and get a clear, step-by-step explanation grounded in how ECON31031 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 41 of your subjects at The University of Manchester, plus 1,000+ more Bibles across the AskSia library.
Sia - your ECON31031 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