STAT 443:

Forecasting

Paul Marriott

Preamble

Introduction

Examples
Observed data
examples
STAT 443: Forecasting

Paul Marriott

[email protected]
M3 4204
 STAT 443:
Forecasting

Paul Marriott
Overview
Preamble

Introduction
• This course is an introduction to the area of statistical
Examples
Observed data
examples
forecasting and control.
• empirical information and modelling can be used to
make forecasts and evaluate the associated
uncertainty.
• Time series analysis and modelling, building on the
student’s knowledge of regression.
• Linear filtering, the Box-Jenkin’s approach to
forecasting, Bayesian and state space methods and
frequency analysis.
• Each forecasting approach is illustrated with real world
examples.
 STAT 443:
Forecasting

Paul Marriott
Learning Materials
Preamble

Introduction

Examples
Observed data
examples

• Note can be downloaded from Desire2Learn (D2L) site

(https://learn.uwaterloo.ca)
Slides are just a subset of notes
• Optional Reference Text: see bibliography on D2L
• Notes from class
• Office hours 2:00-3:00 Tuesday and Thursday, or by
appointment.
 STAT 443:
Forecasting

Paul Marriott
Course Objectives
By the conclusion of this course, students should have
Preamble
achieved the following objectives:
Introduction

Examples
• Be able to give a qualitative description of the
Observed data
examples
structures of a wide range of types of time series, and
discuss the following topics: time scales, trends,
variability, seasonality, and other deterministic aspects,
outliers, change points, stationarity and data generation
mechanism.
• Be able to use R to generate appropriate plots and
summary statistics to communicate to others important
aspects to a wide range of types of time series.
• Be able to discuss, with a variety of examples, the
importance of forecasting, prediction and control in
real-world applications and how to quantify the
effectiveness of differing methods. They will be able to
discuss fundamental differences between short,
medium and long term prediction.
 STAT 443:
Forecasting

Paul Marriott
Course Objectives
Preamble

Introduction
• Be able to select and use appropriate forecasting
Examples methodologies studied in the course to make forecasts
Observed data
examples based on data and appropriate subject matter
information.
• Be able to quantify the uncertainties and assumptions
associated with each of the forecasting and filtering
methodologies studied in the class.
• Be fluent in a number of R based tools which can be
used to compute and summarise forecasts.
• They should be able to describe the importance of
statistical modelling in forecasting, prediction and
control problems and master the mathematical
techniques needed to exploit such models. They
should be able to select, fit and critique such models.
 STAT 443:
Forecasting

Paul Marriott
Course Outline
Preamble

Introduction

Examples 1 Introduction to forecasting, control and time series

Observed data
examples
2 Regression methods and model building principles
3 The theory of stationary processes
4 The Box-Jenkins approach to forecasting and control
5 Bayesian and state space methods
6 Other topics in time series modelling. Depending on
time topics selected from (a) ARCH and GARCH
modelling, ( b) The Kalman filter (c) Non-linear
dynamical systems, (d) Frequency analysis, (e)
Threshold models, and (f) Multivariate time series
models
 STAT 443:
Forecasting

Paul Marriott
Tentative Timetable
Preamble
Week Monday Wednesday
Introduction
Starting
Examples
Observed data
examples
6 May Chapter 1 Chapter 1
13 May Chapter 1 Chapter 1
20 May Victoria day Chapter 2
27 May Chapter 2 Chapter 2
3 June Chapter 3 Midterm
10 June Chapter 3 Chapter 3
17 June Chapter 4 Chapter 3
24 June Chapter 4 Chapter 4
1 July Chapter 4 (Tuesday) Chapter 5
8 July Midterm Chapter 5
15 July Chapter 5 Chapter 6
22 July Chapter 6 Chapter 6
29 July Chapter 6
 STAT 443:
Forecasting

Paul Marriott
Course Assessment
Preamble • Your final mark will be calculated according to the
Introduction
following weighting: Group project – 10%, Midterm –
Examples
Observed data
30% (15% each), Final exam – 60%.
examples

• If you have a valid and documented reason for missing

a midterm the weight will be transferred to the final.
• To get an INC in the final you would have to have a
passing grade in all of all the assessments of the term.
• In order to pass the course, a student must have a
mark of at least 50% on the final exam as well as an
overall final grade of at least 50% according to the
above grading scheme.
• The project will be in groups of 4 or 5 and will involve
an analysis of a data set and an evaluation of a method
used in the course using a simulation study. There will
be a written report.
 STAT 443:
Forecasting

Paul Marriott
Cheating and Academic
Preamble
Discipline
Introduction • Cheating on assignments and projects includes copying
Examples
Observed data
another student’s solution and submitting it as your
examples
own, allowing another student to copy your solution, or
collaborating excessively with another student.
• Plagiarism is the act of presenting the ideas, words or
other intellectual property of another as one’s own. The
use of other people’s work must be properly
acknowledged and referenced in all written material
such as take-home examinations, essays, laboratory
reports, work-term reports, design projects, statistical
data, computer programs and research results.
• The standard penalty for cheating or plagiarism on an
assignment or project is as follows: no marks for the
assignment and a deduction of 5% from the final
course grade
 STAT 443:
Forecasting

Paul Marriott
Cheating and Academic
Preamble
Discipline
Introduction

Examples • It is permissible, and indeed desirable, to discuss

Observed data
examples
assignment solution methods with classmates, TAs,
and instructors.
• You should work through the solution yourself and write
it in your own words. The only exceptions are
assignments or projects which the instructor designates
as ’group’ activities.
• In academic work, it is customary to acknowledge, in
writing, all sources of help. We require that, for each
assignment or project submitted, you write (and sign)
an acknowledgement of help received, which includes
the names of the people (if any) with whom you
discussed your solutions.
 STAT 443:
Forecasting

Paul Marriott
Practice Questions
Preamble

Introduction

Examples • I will regularly post practice questions

Observed data
examples • I will not be providing solutions to these questions.
• They are purely to help you in your study. If you have
solutions that you would like to discuss please come to
office hours or talk to TAs
• Midterm/Exam questions will be closely related to these
questions
• In some sense these questions define the course from
the students point of view
• Some will involve R and data – corresponding exam
question will involve interpretation of output
 STAT 443:
Forecasting

Paul Marriott
Students Role
Preamble

Introduction

Examples
Observed data
examples
What do you need to do to be successful in this course?
• Come to lectures and take notes
• Work – individually or in groups – through the practice
questions
This will completely prepare you for the midterms and
final
• Work in the group project
 STAT 443:
Forecasting

Paul Marriott
Chapter 1: Introduction to
Preamble
Forecasting, control and time
Introduction series
Examples
Observed data
examples

“There are two kinds of forecasters: those who

don’t know, and those who don’t know they don’t
know.” J. K. Galbraith
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction
UK Deaths and Injuries Ontario road deaths
Examples

250
Observed data
2500
examples

200
2000
Number

Number

150
1500

100
1000

50
1970 1975 1980 1985 1960 1965 1970 1975

Time Time

Figure: 1.1 Number of deaths and serious injuries in road

accidents per month
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction

Examples
Observed data
examples Definition.1.2.2 (Time series plot) A time series plot is a
scatterplot of the data with time on the x-axis in, typically,
equally spaced intervals, and the observations on the
y -axis.
For visual clarity adjacent observations are connected by
lines.
The equally spaced intervals can be in units of years,
months, days, hours etc and these units are called the
period of the time series.
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction

Examples
Observed data
examples
• Terms like level, trend and seasonality can only be
defined precisely in terms of a modelling exercise.
• Here we give informal definition for use in describing
observed patterns in time series plots.
Definition.1.2.3 (Level, trend and seasonal component)
The level is the local mean of the observations, around
which we see random noise.
When the level varies with time we say there is a trend.
A seasonal effect is a systematic and calendar related effect
which repeats with a given period.
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction

Examples Definition.1.2.4 (Change point) A change point is a time at

Observed data
examples which at least of the of the following changes: the data
generation process, the way that the data is measured, or
the way that the observation is defined.

Definition.1.2.5 (Non-stationarity) Seasonality, trends,

non-constant variance and change points are all examples
of non-stationarity.
Stationarity, defined formally in Chapter 3, informally means
that the underlying random process does not change in
time.
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble
Ontario Gas Demand
Introduction

250000
Examples
Observed data
examples

200000
Gallons

150000
100000

1960 1965 1970 1975

Time

Figure: 1.2 Monthly demand for gasoline in Ontario from 1960 to

1975
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction

Examples

600
Observed data
examples

500
AirPassengers

400
300
200
100

1950 1952 1954 1956 1958 1960

Time

Figure: 1.3 Monthly totals of international airline passengers,

1949 to 1960.
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction Internet Traffic Internet Traffic

Examples 2000000
Observed data
examples

100000
1500000
Bits

Bits

60000
1000000

20000
500000

Dec 01 Dec 15 Jan 01 Jan 15 Jan

Days Days

Figure: 1.4 Network traffic in the UK academic network daily and

hourly intervals.
 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction
FTSE index 1991 − 1998 First Difference log price
Examples
Observed data
8000

examples

0.04
6000

0.00
Difference
FTSE

4000

−0.04
2000

−0.08
1992 1994 1996 1998 1992 1994 1996 1998

Time Time

Figure: 1.5 Daily FTSE stock price.

 STAT 443:
Forecasting

Paul Marriott
Time series
Example.1.2.10 (US unemployment)
Preamble

Introduction

Examples US unemployment
Observed data

12000
examples

10000
unemployed (thousands)

8000
6000
4000

1970 1980 1990 2000

Time

Figure: 1.6 US unemployment

 STAT 443:
Forecasting

Paul Marriott
Time series
Preamble

Introduction
Monthly Births in Denmark
Examples
Observed data

4000 5000 6000 7000 8000 9000

examples
birth

1900 1920 1940 1960 1980

Time

Figure: 1.18 Monthly birth data in Denmark 1900-1992

 STAT 443:
Forecasting

Paul Marriott
Review
Preamble

Introduction

Examples
Observed data
examples • In this lecture we have seen numerous examples of real
world time-series
• For each can you define the period?
• Can you, at least informally, describe the level, trend
and seasonality of the plot?
• Can you identify the seasonal component if it exists?
• Can you define and identify any change points?
• Are any of the examples stationary?