KTEE 418

FORECASTS
IN ECONOMICS AND BUSINESS
Instructor: MSc. Nguyen Thuy Quynh
Email: [email protected]
Office: Room B208 – Faculty of International Economics
 COURSE DESCRIPTION

• Course title: Forecasts in Economics and Business

• Course code: KTEE418
➢Our class: KTEE418(GD2-HK1-2223).1
• Credit hours: 3
• Prerequite(s): Econometrics 2 (KTEE318)
• Department: Faculty of International Economics
 COURSE OBJECTIVES
1. Understand forecasting and forecasting methods
2. Understand time-series forecasting models
3. Be able to use time series models in forecasting different
economic or business time series
4. Determine appropriate forecasting methods for a certain
series
5. Master EVIEWS in forecasting
MAIN TOPICS
Chapter 1: Overview of forecasting

Chapter 2: Simple forecasting models

Chapter 3: Time series decomposition method

Chapter 4: Box – Jenkins method (ARIMA model)

Chapter 5: Vector Autoregression model

Practice with EVIEWS

 COURSE EVALUATION
• Duration: 45 hours ~ 15 in-class lectures and practices

Participation Mid-term Final Test

(10%) (30%) (60%)
• Attendance • Group • MCQs and
check Project Exercises
• Exercise • Extra point
submission
 MATERIALS AND SOFTWARE
• Hanke, J.E. & Wichern, D.W. (2014), Business Forecasting, 9th Edition, Pearson
• Hyndman, R.J., & Athanasopoulos, G. (2021) Forecasting: principles and practice, 3rd edition,
Textbooks OTexts: Melbourne, Australia. (Online version, accessible at https://otexts.com/fpp3/)

• Hoai, N.T., Binh, P.T & Duy, N.K.,(2014), Forecasting and Analysis of Data in Economics and
Finance, Statistical publisher
Additional • Nguyen Quang Dong and Nguyen Thi Minh, (2013), Basic Econometrics, Publisher of National
Readings Economics University

• EVIEWS
• EXCEL
Software
 MATERIALS

Textbooks Additional Readings

 CHAPTER 1
OVERVIEW OF FORECASTING

1. Introduction to forecasting
2. Time series data
1. INTRODUCTION TO FORECASTING
1.1 Why forecasting?
1.2 Different forecasting methods
1.3 Forecasting steps
1.4 Measuring forecast error
1.1 WHY FORECASTING?
• Forecasting is to foretell the future.
• Sophisticated software packages make forecasts of future values for
variables of interest easier.
• Judgment should be used to review and perhaps modify the forecasts
produced by quantitative procedures.
• Forecasts necessarily drive policy setting and planning
➢The Fed adjusts interest rates based on the forecast of future
economic growth and inflation.
➢Operation managers set production schedules based on forecast of
future sales.
 TYPES OF FORECASTS

Nature of
Time Horizon Scope Method
Output
• Long-term • Micro: sales, • Qualitative • Point forecast: a
forecast: year(s) number of • Quantitative single number
ahead workers needed best guess
• Short-term • Macro: GDP, • Interval
forecast: day(s)/ inflation, crude forecast: a
month(s) ahead oil price, labor range of
force, number within
unemployment which the
rate future value is
expected to fall
1.2 FORECASTING METHODS

Forecasting
methods

Qualitative Quantitative
forecasting forecasting

Regression Time series

model models
1.2 FORECASTING METHODS
• Qualitative forecasting methods (more subjective)
➢These methods are based on emotions, intuitions, judgments, personal
experiences, and opinions.
➢Interviews with relevant people to know their judgements or future
intentions, then synthesize data and make forecasts.
➢There is no math involved in qualitative forecasting methods
• Quantitative forecasting methods (more objective)
➢These methods depend wholly on mathematical or quantitative models.
➢Based on historical movements of data (historical data analysis) to make
predict about future values
QUALITATIVE FORECASTING METHODS

Collect information
Judgement of experts
from relevant people

Market Survey

Sales Force
Delphi Method
Composite

Executive Opinion
QUALITATIVE FORECASTING METHODS

Delphi Method
DELPHI METHOD
 QUANTITATIVE FORECASTING MODELS
• Causal models:
➢Linear regression model: simple regression model, multiple regression
model, time series regression model
• Time series models:
➢Univariate time series model:
✓Naïve forecast
✓Moving average
✓Exponential smoothing
✓Box – Jenkin Method (ARIMA model)
➢Multivariate time series model (VAR, VECM)
 KTEE418 COURSE CONTENTS
• Forecast using time series models
➢Simple forecast model: Naïve forecast, moving average, seasonal
adjustment, exponential smoothing, simple trend model
➢Time series decomposition method: Additive model, Multiplicative
model
➢Box – Jenkin method (ARIMA Model)
➢Multivariate time series model (VAR)
 1.3 FORECASTING STEPS
Five steps in the forecasting process (Hanke, 2014):

1. Problem 4. Model
2. Data 3. Model
formulation Implementation 5. Forecast
manipulation building and
and data (the actual evaluation
and cleaning evaluation
collection forecast)
 1.3 FORECASTING STEPS
Another time series forecasting process (Hanke, 2005):
Define forecasting Determine time
Determine variable Collect historical data
Objectives horizon

Explore data patterns

Choose appropriate
model

Predict historical
observations
Re-analyze historical
data
Good Check forecast
Evaluate Make future forecast
Not Good accuracy

Good
Evaluate

Re-analyze updated
data Not Good
1.4 MEASURING FORECAST ERROR
• Yt = actual value at time t
෠𝑡 = predicted/estimated value at time t
•𝑌
• n = number of observations that have forecast error
෠𝑡
• Forecast error/ residual: et = Yt - 𝑌
• A forecasting model is considered an appropriate/good one when the
forecasting error is relatively small.
1.4 MEASURING FORECASTING ERROR
Criterion Formula
1. Mean Absolute Deviation
MAD =
 |e | t

n
2. Mean Absolute Percentage Error
MAPE =
 |e /Y |
t t

n
3. Mean Squared Error
MSE =
t
e 2

n
4. Root Mean Squared Error
RMSE =
t
e 2

n
1.4 MEASURING FORECASTING ERROR
Criterion Formula

5. Mean Percentage Error

MPE =
 (e / Y )
t t

n
6. Theil’s U-statistic

U = 1 The naïve method is as good as

the forecasting model being evaluated. U=
t
e 2

U < 1: The forecasting model being  t t −1

(Y − Y ) 2

used is better than the naïve method.

RMSE (model)
U > 1: The naïve method produce better =
forecast. RMSE (naive)
U < 0.55 => a good forecasting model
 2. TIME SERIES DATA
2.1 What is time series data?
2.2 Characteristics of time series data
2.3 Analyzing time series data
2.1 TIME SERIES DATA
• A time series consists of data that are collected, recorded, or observed
over successive increments of time.
➢a sequence of data points collected over time intervals
➢Time periods are of equal length.
• Series notation: Yt
• For example:
- Data on the annual GDP of Vietnam from 1990 to 2018
- The daily closing price of VCB from 1/7/2019 to 31/7/2019
- GDP and money supply M2 of Vietnam from 1990 to 2018 (2 series)
2.1 TIME SERIES DATA
CPI
240

220

200

180

160

140

120

100
1996 1998 2000 2002 2004 2006 2008 2010 2012
2.1 TIME SERIES DATA
200

160

120

80

40

0
94 96 98 00 02 04 06 08 10 12 14 16

AGRI_INDEX OIL_PRICE
2.1 TIME SERIES DATA
Some sources for time series data:
• Macro-level data:
✓General Statistics Office: https://www.gso.gov.vn/en/data-and-statistics/
✓Vietnam Customs:
https://www.customs.gov.vn/Lists/EnglishStatistics/Default.aspx?language=en-US
✓International Trade Center: https://www.intracen.org/itc/market-info-tools/trade-
statistics/
✓World Bank: https://data.worldbank.org/?name_desc=true
✓Asian Development Bank: https://data.adb.org/
✓OECD: https://stats.oecd.org/
• Micro-level Data: Financial reports, Internal Documents
 2.2 CHARACTERISTICS OF TIME SERIES DATA
Frequency: indicates the regularity in which data is collected (daily, weekly,
monthly, quarterly, yearly…)
• Low frequency:
✓GDP, inflation rate, unemployment rate, M2
• High frequency:
• Stock price
• Oil price, Gold price, Exchange rate
Number of observations for time series data is usually less than those for
cross-sectional and panel data.
Minimal Data Requirement:
✓For simple forecasting models: less than 30 observations
✓For more complicated models (ARIMA, VAR,…): >30 observations
2.3 ANALYZING TIME SERIES DATA
What to analyze for a time series?
- Components of a time series
- Stationarity
2.3 ANALYZING TIME SERIES DATA
2.3.1 Components of a time series
• Trend: represents the underlying growth (or decline) over several time
periods
• Seasonality: the pattern of change that appears annually and repeats itself
year after year
• Cyclical component: a series of wavelike fluctuations or cycles of
more than one year’s duration, usually known as business cycle or
economic cycle
• Irregular component: unpredictable or random fluctuations (similar to
the disturbance in regression model)
TIME SERIES COMPONENTS
TIME SERIES COMPONENTS
 TIME SERIES COMPONENTS
Y
240

200

160

120

80

40
I II III IV I II III IV I II III IV I II III IV I II III IV
2003 2004 2005 2006 2007
 MODELS RELATING THE OBSERVED VALUE OF Yt
• Additive components model:

Yt = Tt + St + Ct + I t
• The time series being analyzed has roughly the same variability
throughout the length of the series.
• The seasonal pattern repeats year after year with almost unchanged
magnitude.
• The values of the series fall essentially within a band of constant width
centered on the trend.
 MODELS RELATING THE OBSERVED VALUE OF Yt
• Multiplicative components model:

Yt = Tt * St * Ct * I t

• The values of the series spread out (shrinks) as the trend increases
(deceases).
• The seasonal pattern repeats year after year with increasing or decreasing
magnitude.
• The graph looks like a megaphone or funnel.
TMODELS RELATING THE OBSERVED VALUE OF Yt
• How to determine whether a series is additive or multiplicative?
• => Draw the graph
 2.3 ANALYZING TIME SERIES DATA
2.3.2 Stationarity
• A stationary time series is one whose properties do not depend on the time at
which the series is observed.
• The statistical properties do not change over time
✓Mean: E(Yt) = μ
✓Variance: Var(Yt) = σ2
✓Covariance: Cov(Yt, Yt+k) = cov(Ys, Ys+k) = γ𝑘
=> A weakly stationary series
• Time series with trends, or with seasonality, are not stationary.
 STATIONARITY
SALES
1,560

1,520

1,480

1,440

1,400

1,360

1,320

1,280

1,240

1,200
5 10 15 20 25 30 35
 2.3 ANALYZING TIME SERIES DATA
Stationarity: Why do we want stationary series?
- Time series forecasting models make forecasts based on historical
movements of a series, assuming that those movements continue to
happen.
- A stationary series has stable patterns (statistical properties) => good for
time series forecasting
 2.3 ANALYZING TIME SERIES DATA
Usually, a time series may have one or more of the following patterns:
- Stationarity
- Trend
- Seasonality
- Cyclical component (less observed)
 2.4 TECHNIQUES TO
EXPLORE TIME SERIES DATA PATTERN
▪ Plotting graph of data over time
▪ Analyzing series’ correlogram
▪ Using unit root test
 PLOTTING THE SERIES OVER TIME
• Command: line [var] or plot [var]
• The graph can be used to observe the components of a series: Is there a
trend (linear or non-linear)? Seasonality? Cyclical components?
GDP
2,000

1,600

1,200

800

400

0
1955 1960 1965 1970 1975 1980 1985 1990 1995
 ANALYZING CORRELOGRAM
• View → Correlogram → view the ACF and PACF of level/ 1st difference/
2nd difference
 UNIT ROOT TEST
• Is the series stationary?
• View → Unit Root Test
 2.5. CHOOSING A FORECASTING METHOD
• The selection of appropriate forecasting models depends on:
✓Data pattern
✓Number of available historical observations
✓Forecasting horizon (long-term, short-term)
✓Forecasting error
 2.5. CHOOSING A FORECASTING METHOD
Minimal Data Requirements
Forecasting method Pattern of Data Time Horizon
Nonseasonal Seasonal
Naïve Stationary, Trending, 1 or 2 4 Short-term
Seasonal

Moving Averages Stationary 4 – 20 Short-term

Exponential Smoothing
+ Đơn Stationary 5 – 10 Short-term

+ Holt’s Trending 10 – 15 Short-term to

intermediate-term

+ Winter’s Trending, Seasonal 4×s Short-term to

intermediate-term
 2.5. CHOOSING A FORECASTING METHOD

Minimal Data Requirements

Forecasting method Pattern of Data Time Horizon
Nonseasonal Seasonal
Time series Trending, Seasonal 5×s Short-term to
decomposition intermediate-term

ARIMA model Stationary, Trending, 24 3×s Short-term to

Seasonal intermediate-term
VAR >=50 Short-term to
intermediate-term