Telecommunications demand a

review of forecasting issues

Robert Fildes, Lancaster University

R.Fildes@Lancaster.ac.uk

This paper is based on on work published in the IJF: 2002(4) as

part of a special issue on Telecoms Demand Forecasting.
The text downloadable from Science Direct

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Consumer End Uses
video
games
medical
(d) satellite
home shopping
via
internet (b) mobile
call centres

Local
carrier (a) PSDN IXC POT IXC

Business End Uses

(e) local bypass
data transfer
video conferencing
IXC2 IXC2 group work
(c) Cable CAD/CAM

Robert Fildes, IIF & Lancaster Centre for Forecasting

 A Hierarchy of Forecasts

Usage
l Overall demand for an application

l Competition within technologies to service application

Access
l Competition between suppliers

l Once investment in place

operating costs are minimal
Equipment
Forecasting Implications
e.g. for entrants, price collusion, price leadership

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Types of Telecommunications Forecasting Problems

n Stable Markets - aggregate Corporate/ regulator data

Public Switched available

n Stable Markets - disaggregate Surveys, customer behaviour

Calling patterns
Intentions Surveys
n Growth Markets Feature evaluations
Trial markets
Special Services
Analogies
New services other products
l ADSL other countries
Delphi/ Expert judgement

Statistical time series methods

n Operations
Different
Differentproblems
problems--different
differentorganisational
organisationalresponsibilities
responsibilities
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Stable Markets
Point-to-point PSTN
Local, long-distance, international
Drivers
Price, income, advertising, value-for-money
Problem-specific drivers
Call-back on international
Price elasticity declining
US international message telephone service
Depends on price, culture (IMTS): (Econometric comparisons, Madden et
al, IJF)

Econometric models with

Focus on price elasticity estimation
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Conclusions
Conclusions -- Stable
Stable Markets
Markets

n Increasing econometric sophistication Missing

Missingout
outor
orincluding
including
model specification (externalities: market inappropriate
inappropriatevariables
variables
size)
diagnostics mis-understanding
mis-understandingthe
themarket
market
multicollinearity - the interrelationship poor
poorforecasts,
forecasts,
between the explanatory variables, bad
badpolicy
policy
e.g ownership of durables and income

Does
Doesthe
themodel
modelwork
workbetter
better
But no stability or forecasting tests
than
thanalternatives?
alternatives?
Time varying coefficients? IsIsthe
themarket
marketchanging,
changing,
are
arenewnewservices
servicesgetting
gettingmore
more
acceptable,
acceptable,isisquality
qualitybecoming
becoming
more
moreimportant?
important?
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Stable
Stable Markets
Markets Disaggregate
Disaggregate

Access & Usage

Based on Cross-sectional household data
Conditional models of usage, given choice of technology
or carrier
Focus on price elasticity
By household demographics to support universal service analysis
Linked to marketing planning

Evaluation:
Evaluation:
Despite
Despitetheir
theiruse
usefor
forforecasting
forecasting
Dynamics
Dynamicsof ofchange
changenot
nottaken
takeninto
intoaccount
account
No
Noforecasting
forecastingevaluation
evaluation
Robert Fildes, IIF & Lancaster Centre for Forecasting
 New
New Products
Products && Services
Services

Market Potential Intentions Surveys

mobile Feature evaluations
really new products, e.g. 3G Choice models
Trial markets
Analogies
other products
Diffusion Path other countries
E.g mobile
But now market dominated by
switchers?

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Market Potential for M1, Market Potential for Time line
first generation, M1 not M2 second generation, M2

Adopters
at T-1

Leavers Switchers
Adopters & Adopters First generation
Users at T at T Up to T
Exit
Switchers Leavers
Adopters & Adopters &
Exit Both generations,
Users at T+1 Users at T+1 Exit
Competition

Forecasting telecommunication service subscribers in substitutive and

competitive environments, Jun et al, IJF, 2002
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Competition
Competition between
between Technologies
Technologies

160

140

120

100
1st Gen'ion

1st Gen'ion
80 2nd Gen'ion
TOTAL
60

40

20

0
0 10 20 30 40 50
Period

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Market
Market Potential
Potential -Decomposition
-Decomposition Methods
Methods II
(applies
(appliesto
tonew
newand
andestablished
establishedmarkets
markets
Segmentation Approaches to Forecasting)
All users

Business Personal
Subscribers Subscribers

Business Professional Below 21 years 21 - 50 > 50

Small Income > 30K Attending College Not attending Income < 20K Not retired Retired

Large Income < 30K

Commercial Income > 30K Income <30K

Large
Service

Forecast consumption in each segment

Project numbers in each segment }and multiply
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Decomposition
Decomposition Methods
Methods IIII

Step:
1. Identify the most important services (in terms of load in bits per
second on the network).
2. Segment the population, e.g business and consumer, high and low
intensity users.
3. Using ideas of a time budget (the number of hours a day available
for telecoms related use) estimate the services each segment might
use, by service, and the level of usage.
4. Model the changes in the segments over time.
5. Obtain total traffic by aggregating the individual forecasts of the
usage profiles.

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Market
Market Potential
Potential -- Decomposition
Decomposition Methods
Methods III
III
Alternative Models of choice
& Usage - for households with Internet Access
(see Rappaport et al, 2001)

Simultaneous Sequential
?
?

PSDN ADSL ISDN2 PSDN Broadband?

ADSL ISDN2

Usage models
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Choice
Choice Models
Models -- segmenting
segmenting consumers
consumers
and
and forecasting
forecasting each
each segment
segment
Used in stable & growth markets

n market research based

observed data
hypothetical questions for new products/ services

Problems
6 sample size in small populations
6dynamics
how do the parameters change over the planning horizon
6price dependency
measurement (quality)
projected behaviour
Robert Fildes, IIF & Lancaster Centre for Forecasting
 The
The Bass
Bass Model:
Model: basic
basic segmentation
segmentation
Two types of people:
Innovators
They adopt because of their
attitude to technology
Imitators 20
Gompertz
They adopt when exposed to 18

16
consumers who have adopted 14

already 12

10

4 Logistic
2

13

17

21

25

29

33

37

41

45

49

53

57
Red:
Red:adopters
adopters
Black:
Black:potential
potentialadopters
adopters
Blue:
Blue:lacking
lackinginformation
information
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Estimating the Diffusion Path:
limited if any sales data
S-Shaped curve used to represent adoptions
different curves and parameters Issues:
different market potential and uptake data limitations
trajectory
market potential affected by
dependent on key parameters
product/ social factors
The Effect of Increasing Imitation in the Social System
The Effect of Increasing Innovation in the Social System
25
25

20
20

Households (Millions)
Increases Slope of the Curve

15
15
Households (millions)

Pulls curve leftwards

10
10

5
5

0
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Modelling multinational
telecommunications demand with limited
data, Islam et al, IJF, 2002
Robert Fildes, IIF & Lancaster Centre for Forecasting
 The impact of networked groups
(Bass defines a simple network)
Models
Models of
of Interacting
Interacting Individuals
Individuals
Different networks of consumers
Different adoption patterns

Connected to
Connected to near Connected at Random
your two nearest
neighbours but
neighbours
with other links
Each
Eachgroup
groupofofindividuals
individualshas
hasits
itsown
ownrules
rules
of
ofbehaviour
behaviourand
andinteraction
interaction
affected
affectedby
byits
itsenvironment
environment
Robert Fildes, IIF & Lancaster Centre for Forecasting
 How
How can
can these
these Agent
Agent based
based models
models be
be used
used
(Collings,
(Collings, BTExact)
BTExact)

Aim to understand behaviour of interacting markets

Diverse individual behaviour
Asymmetric information and motivation

Examples
Financial markets
Customer relationship management

Robert Fildes, IIF & Lancaster Centre for Forecasting

 Diffusion models

EXTENSIONS and ISSUES

Can include marketing and economic variables

Estimation (analogy or numerical methods)

Meta Models
link the diffusion parameters to market characteristics
to other products, other markets
Genetic algorithms

Incorporation of effects from other products/ markets

Supply restrictions (in regulated markets)

Disaggregate models ( e.g. to industry specific uptake of fax)

Robert Fildes, IIF & Lancaster Centre for Forecasting
 Bass Model - Diffusion Parameters and
Meta Model
Describing the speed and shape of the adoption path
Mobile
Average diffusion
parameters
Imitation
NICs Japan

UK, Denmark

Innovation
Robert Fildes, IIF & Lancaster Centre for Forecasting
 Simulation
Simulation Modelling
Modelling
system
systemdynamics
dynamicsreplicate
replicatediffusion
diffusioncurves
curves
offer
offermore
moreflexibility
flexibilityfor
forincorporating
incorporating
information
informationand
andprocesses
processes
decision
decision variables
variables
Service provider

Marketing Quality influences

Price

Understanding Utility Acceptability decision factors

Adoption Potential Market

Key benefit: transparency of model

Robert Fildes, IIF & Lancaster Centre for Forecasting
 Evaluation - Models of Market Potential
Choice Models Problems
Based on intentions data current intentions
changing valuations
Models of Market Penetration
unvalidated
define potential = CM(t) where M(t) is c depends on time
determined by the economic/ social system
of the market
Based on analogous products and countries

Evaluation - Models of the Diffusion Path

Aggregate Bass-type diffusion Limited forecast validation
short term (if any)
use too much data
poor benchmarks
Simulation
No forecast validation
Robert Fildes, IIF & Lancaster Centre for Forecasting Limited parameter validation
 Final
Final comments
comments
Widespread interest in telecoms forecasting

Survey evidence suggests organisations which adopt a more ambitious

and rigorous approach do better

Primary methods nave qualitative

despite major investments (and disasters) riding on the results of a forecas
Structured use of judgement (E.g. Delphi: Knut Blind)

Too little academic research

too hard?
Too messy (Forecasting category sales and market share for wireless telephone
subscribers: a combined approach, Kumar et al, IJF, 2004)
IJF
IJFSeminar
Seminarobjectives:
objectives:
identify
identifywhere
whereprogress
progresshas
hasbeen
beenmade
made
Robert Fildes, IIF & Lancaster Centre for Forecasting
gap
gapanalysis
analysis