1-s2.0-S0169207014001794-main

International Journal of Forecasting 31 (2015) 1105–1126

Contents lists available at ScienceDirect

International Journal of Forecasting

journal homepage: www.elsevier.com/locate/ijforecast

Review

Forecasting in telecommunications and ICT—A reviewNigel Meade a,∗, Towhidul Islam b,1

a Imperial College Business School, Imperial College London, Exhibition Road, London SW7 2AZ, UKb Department of Marketing and Consumer Studies, University of Guelph, Guelph, Ontario, N1G 2W1, Canada

a r t i c l e i n f o a b s t r a c t

Keywords:Diffusion modelsTime seriesTechnology forecastingMobile telephonyInternet

Given the length of time that has elapsed since the IJF Special Issue on TelecommunicationsForecasting in 2002 and our reliance on information and communications technology (ICT),it is now appropriate to review the flow of benefits from forecasting to ICT and fromICT to forecasting. The importance of ICT is demonstrated by its accounting for 8.2% ofthe value added and for over 20% of employment in the OECD countries. The literaturereviewed is categorised by both the ICT area of application and the modelling approach.The ICT application areas are: mobile telephony, internet usage or provision, and other ICTrelated products and services. Themainmodelling and forecasting approaches are diffusionmodelling and forecasting, time series forecasting and technological forecasting, and thisreview devotes a section to each of these approaches. Most of the research activity in thefield (measured by numbers of papers) has occurred in the modelling of diffusion in ICT,particularlymobile telephony, producing beneficial cross-fertilisation between forecastingand ICT applications; examples are multi-generational modelling and choice modelling.Although call centre manpower planning has led to innovative forecasting models, otheranalyses of clusters of ICT time series data sets have been less innovative. Technologicalforecasting papers tend to be exercises based on expert opinion.© 2014 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.

Contents

1. Introduction.......................................................................................................................................................................................... 11062. Diffusion models .................................................................................................................................................................................. 1107

2.1. Single country diffusion analyses ........................................................................................................................................... 11072.1.1. Example analyses and comparisons of forecasting accuracy................................................................................. 11082.1.2. Analysis of the generality of the conclusions from the ‘Frank’ analysis................................................................ 11102.1.3. Definition of market potential ................................................................................................................................. 11102.1.4. Summary of similar studies ..................................................................................................................................... 11112.1.5. Other single-country analyses related to ICT diffusion.......................................................................................... 1112

2.2. Multi-country analyses of ICT diffusion ................................................................................................................................. 11132.2.1. Multi-country analyses with short time series....................................................................................................... 1113

∗ Corresponding author. Tel.: +44 0 20 7594 9116; fax: +44 0 20 7823 7685.E-mail addresses: [email protected] (N. Meade), [email protected] (T. Islam).

1 Tel.: +1 519 824 4120x53835; fax: +1 519 823 1964.

http://dx.doi.org/10.1016/j.ijforecast.2014.09.0030169-2070/© 2014 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.

http://dx.doi.org/10.1016/j.ijforecast.2014.09.003

http://www.elsevier.com/locate/ijforecast

http://www.elsevier.com/locate/ijforecast

http://crossmark.crossref.org/dialog/?doi=10.1016/j.ijforecast.2014.09.003&domain=pdf

mailto:[email protected]

mailto:[email protected]


1106 N. Meade, T. Islam / International Journal of Forecasting 31 (2015) 1105–1126

2.2.2. Multi-country diffusion analyses using covariates ................................................................................................ 11132.2.3. Other multi-national analyses related to ICT diffusion.......................................................................................... 1115

2.3. Choice modelling and other extensions ................................................................................................................................. 11152.4. Summary: the flow of benefits between ICT and forecasting methodologies for diffusion ............................................... 1116

3. Time series............................................................................................................................................................................................ 11173.1. Forecasting for call centres...................................................................................................................................................... 11173.2. ICT time series.......................................................................................................................................................................... 1118

3.2.1. Univariate telecommunications time series ........................................................................................................... 11183.2.2. Internet time series: forecasting usage and provision ........................................................................................... 11203.2.3. Forecasting for other ICT products .......................................................................................................................... 1121

3.3. Summary: the flow of benefits between ICT and time series forecasting methodology .................................................... 11214. Technological forecasting .................................................................................................................................................................... 1121

4.1. Mobile telephony..................................................................................................................................................................... 11214.2. The internet.............................................................................................................................................................................. 11214.3. Summary: the flow of benefits between ICT and technological forecasting methodology ................................................ 1122

5. Summary, conclusions and suggestions for further research ........................................................................................................... 1122Acknowledgments ............................................................................................................................................................................... 1123Appendix A. Supplementary data .................................................................................................................................................. 1123References............................................................................................................................................................................................. 1123

1. Introduction

In this paper, we review applications of modellingand forecasting in information and communications tech-nology (ICT). We understand the term ICT as includingtelecommunications and referring to both hardware andsoftware. To put the importance of ICT in context, an anal-ysis of 28 OECD countries showed that, in 2008, the ICTsector accounted for between 3.7% (Switzerland) and 13.9%(Finland) of the value added inmanufacturing and businessservices (the average proportion was 8.2%). For the samecountries, ICT-using occupations (including specialists, ad-vanced and basic users) account for over 20% of total em-ployment on average, ranging from10.9% (Turkey) to 35.3%(Luxembourg) (see OECD, 2011).

It is over a decade since the International Journal of Fore-casting published a Special Issue on TelecommunicationsForecasting in 2002 (see Fildes, 2002, 2003). Given the timethat has elapsed since then, and the almost total depen-dence of life in developed economies on ICT, we feel thatit is now appropriate to review the flow of benefits fromforecasting to ICT and from ICT to forecasting.

Our preliminary literature search found that studies fo-cussing on hardware (television, cellular phones, computerand network equipment) predominated. ICT software ap-plications, such as videoconferencing, distance learningand business to business communications (B2B), do notseem to have not generated forecasting studies.Within theconstraints of the published forecasting literature, our aimis to make this review as inclusive as possible. The pa-pers reviewed here are included primarily because theymake some contribution to modelling and forecasting inan ICT context. To assess the levels of activity in differentresearch areas, these papers can be categorised by both theICT area of application and themodelling approach. The ICTapplication areas can be classified broadly into three cat-egories: mobile telephony, some aspect of internet usageor provision, and other ICT products, such as PCs or televi-sion. The topic that has generated the most interest is mo-bile telephony, representing 27% of the papers reviewed.Modelling and forecasting in relation to the internet rep-resented 19%, and other ICT products were discussed in

39% of papers. Forecasting theory and other related issueswere the theme of the remaining 15%. Three main mod-elling and forecasting approaches have been identified inour review of forecasting in ICT and telecommunications.These are diffusion modelling and forecasting, time seriesforecasting, and technological forecasting. Considering themore recent papers (from 2000 onwards), we find that 55%can be classified as diffusion modelling, 34% can be clas-sified as time series modelling, and 11% focus on tech-nological forecasting. The review considers each of theseapproaches in turn: Section 2 discusses diffusion mod-elling, Section 3 covers time seriesmodelling and Section 4looks at technological forecasting. Within the discussionof each approach, we discuss the ICT categories identified,and also summarise the flow of benefits between forecast-ing and the relevant ICT applications for each approach.We have not attempted to use a common framework toanalyse the three approaches described in Sections 2–4.The problems addressed differ between sections, and thedegree of homogeneity among the applications of eachapproach differs considerably. For example, the studiesmodelling diffusion in a single country are comparativelyhomogeneous, and some general points about the mod-elling approach can be, and are,made. In contrast, the stud-ies in technological forecasting have few issues in common,and no conclusions about the usefulness or appropriate-ness of a technique can be drawn.

The consequence of conducting a review with the self-imposed broad remit of ‘forecasting in ICT’ is that a widerange of analyses is included. Several factors determinethe supply of academic analyses, of which we will iden-tify two. Firstly, where quantitative analysis is crucial tothe understanding of a topic, the availability of data at-tracts academic analysis. In the absence of easily avail-able data, academic analyses are harder to execute and lesscommon. The availability of data has facilitated the sup-ply of studies in Sections 2 and 3. In contrast, there are ar-eas of ICT activity in which forecasting studies are eitherrare or absent. For example, the probable effects of devel-opments in ICT on the relative popularity and future us-age of the different modes of delivery of music, films and

N. Meade, T. Islam / International Journal of Forecasting 31 (2015) 1105–1126 1107

books to consumers have received little attention. A sec-ond important factor is the filter on the supply of studies,namely the threshold for publication. The papers consid-ered in this review come from a large number of differ-ent academic fields of study, including forecasting, timeseries analysis, operations research, operations manage-ment, computer science, telecommunications, and infor-mation technology, to name themore important examples.The threshold for the publication of a given study varies be-tween the journals in these different academic fields. Con-sider, for example, a study which models and forecasts thediffusion of mobile telephony in Ruritania. The height ofthe threshold for publication in a forecasting or time se-ries journal will be influenced by the level of innovation inits modelling and forecast evaluation. In a telecommunica-tions journal, the height of the thresholdwill be influencedmore by the strategic implications of the forecasts for theindustry. Aswewill see in this review, this variety of publi-cation thresholds inevitably means that some studies thathave satisfied one set of criteria will be considered defi-cient according to a different set of criteria.

The distinction between ICT application areas such asinternet and mobile telephony will be increasingly diffi-cult to make in future reviews, as the convergence of thesetechnologies progresses. Internet usage is now the mainrevenue source for network operators in developed mar-kets, rather than voice calls or messaging. Furthermore,global internet access via mobile devices was forecast toexceed the access via fixed broadband by 2014 (PriceWaterhouse Coopers http://www.pwc.com/gx/en/global-entertainment-media-outlook/segment-insights/internet-access.jhtml). It is not very risky to forecast that ICT willcontinue to generate interestingmodelling and forecastingproblems for some time to come.

2. Diffusion models

Diffusion is the process by which an innovation isadopted by a population. Relevant examples of innovationsare, historically, fixed line telephony, or, currently, mobiletelephony. The diffusion process is characterised initiallyby the introduction of the innovation, followed by a slowgrowth in adoption as awareness increases. The growth ac-celerates to a point where adoptions per period peak, thenadoptions decelerate as the population becomes saturatedwith the innovation. The diffusion models used for theadoption of an innovation in a single population are nec-essarily a simplified representation of this process, becausethe data set available is typically a short time series, proba-bly containing fewer than 20 observations. In order to gainfurther insights into the determinants of the diffusion pro-cess, multi-country studies use other variables to explainthe differences in diffusion rates between countries.

Meade and Islam (2006) reviewed studies on themodelling and forecasting of the diffusion of innova-tions. They cover a large body of literature which looksat many different mathematical formulations of an S-shaped diffusion, where the cumulative adoption of aninnovation moves from launch to a saturated market.We do not intend to duplicate this work, but will use

it for a general reference, and specifically for its ap-pendix, which lists fifteen S-shaped growth curve equa-tions. Thus, if no formulation for a diffusion model isgiven, one should refer to the appendix mentioned. Inaddition to the diffusion of an innovation, researchersalso consider technological substitution, where an existingtechnology is replaced by a newer one, such as the replace-ment of rail steam locomotion with diesel locomotion, seefor example Blackman (1972), Fisher and Pry (1971) andSharif and Kabir (1976). One issue that will be discussedhere is the question of whether mobile telephony (or mo-bile internet) is a substitute for fixed line telephony (orfixed broadband). Researchers also consider multiple gen-erations of the same (or similar) technology; this issue isof particular interest in telecommunications and ICT. Ex-amples of this include personal computers (PC, XT, AT, 386,486, Pentium, etc.) and mobile telephones (analog, digital,2G, 3G, 4G), both of which have been through several gen-erations of technology. This is an area of research in whichICT applications (in computing and telephony) have stim-ulated novel developments in modelling. Norton and Bass(1987) extended the use of the Bass diffusionmodel to suc-cessive generations of a technology. Restrictions on the pa-rameters of the multiple generation model were relaxedsuccessively by Islam and Meade (1997) and Mahajan andMuller (1996).

We will discuss the modelling of the diffusion of ICTproducts and services in the following three sub-sections.The first deals with single country diffusions, the seconddeals with multi-country analyses, and the third looks atmodelling market share and choice models.

2.1. Single country diffusion analyses

Several authors have used a range of models for thediffusion of fixed line telephony, with examples rangingfrom Bewley and Fiebig (1988), Chaddha and Chitgopekar(1971), Lee and Lu (1987) and Lee, Lu, and Horng (1992)to Islam and Meade (1996). Similarly, and more recently,authors have compared subsets of models on data setsrelating to the diffusion of mobile telephony and otherICT in particular countries. The evolution of a diffusionprocess over time can be regarded as either a cumulativeprocess, represented by an S-shaped curve, or a non-cumulative process, represented by a bell-shaped curve.In the context of ICT, different actors are concerned withdifferent aspects of forecast diffusion. Service providersare concerned with satisfying the cumulative number ofadopters, whereas handset providers are concerned withsatisfying new adopters (the non-cumulative number ofadopters per time period). Diffusion models are mostappropriate for using with adoption data. For example, ifone is modelling the diffusion of television in the UnitedKingdom, the number of households with a televisionlicense describes the adoption process well (since eachhousehold is legally obliged to buy a license if it hasat least one television). In contrast, television sales dataare less useful, as they confound the adoption processwith replacement sales and purchases of multiple sets perhousehold. If the data used contain extra components, such

http://www.pwc.com/gx/en/global-entertainment-media-outlook/segment-insights/internet-access.jhtml




as multiple sales per adopter, the components need to beincluded in the modelling process.

In this section, since there are many studies with verysimilar structures and objectives, we begin by looking atan example analysis from the literature. In the first sub-section, we describe the steps in an appropriate mod-elling methodology, covering both the way in which thediffusion is modelled and the structure of the error pro-cess. Secondly, we study the importance of the estimationprocedure used in conjunction with the diffusion models.Thirdly, we discuss the implications of the inclusion ofextra components in the data set being modelled, in addi-tion to the adoption process. Fourthly, we review and sum-marise the range of studies on single country ICT/telecomsdiffusion in the context of the steps identified in the firstsubsection.

2.1.1. Example analyses and comparisons of forecastingaccuracy

In this modelling and forecasting exercise, there are aseries of choices to be made. The choices of the diffusionmodel and parameter estimation method are among themore important choices. An early study is that by Frank(2004), who models and forecasts the diffusion of wire-less subscribers (GSM) in Finland. This paper has been citedwidely and its approach has been replicated broadly by re-searchers considering various different national data sets.For this reason,weuse this paper as a case study, to demon-strate the choices made and the consequences of thesechoices. The variable being modelled and forecast is thenumber of mobile telephone subscriptions per 100 people,and the data are taken from the ITU and Statistics Finland;they are shown in Fig. 1. The data are not necessarily iden-tical to those used by Frank. Following Frank, the estima-tion period is from 1980 to 1997; forecasts are evaluatedfrom 1998 to 2004. This limit is chosen because there wasan upward surge in subscriptions in Finland after 2004, dueto the launch of 3G mobile technology.

The model chosen by Frank is the logistic. He uses thecumulative proportion of the Finnish population subscrib-ing to a mobile in year t as the dependent variable. Thisproportion is calculated as the number of subscriptionsdivided by the size of the population; the assumed iden-tity between subscribers and subscriptions is discussed inSection 2.1.3. We denote the dependent variable as F(t),where

F (t) =m

(1 + exp (− (a + bt))), (1)

and m, a and b are to be estimated from the data. The pa-rameter m is called the market potential, the maximumproportion of the populationwhowill become subscribers.Frank effectively assumes that no more than 100% of thepopulation will subscribe (i.e.,m ≤ 1). This restriction willbe discussed further in Section 2.1.3. Frank chooses to esti-mate these parameters,where there are Yt adopters at timet , using non-linear least squares, in conjunction with theassumption that

Yt = F (t) + ξt . (2)

We label this approach as Method 1. Frank introducedsome covariates into the model by parameterising b:

b = b0 + b1GDP t . (3)

The idea is that amore buoyant economy,meaning a higherGDP, will lead to a higher rate of diffusion. Other variableswere tried (the proportion of the population with fixedline telephony, and adummyvariable indicating the switchfrom analog to digital) but were not significant. We denotethis approach Method 1a.

Method 2, which was not used by Frank, considersnew subscriptions per period as the random variable to bemodelled, rather than cumulative subscriptions; that is:

yt = f (t) + εt , (4)

where yt = Yt −Yt−1 and f (t) = F (t)−F (t − 1). The ideahere is that Eq. (4) represents a more credible stochasticmodel, since εt is likely to be closer to the assumption ofindependent, identically distributed errors than the ξt thatis implicit in the minimisation of sums of squares. It haslong been known that ξt is prone to autocorrelation (seeMar-Molinero, 1980).

Method 3, which was not used by Frank either, is astraightforward application of the Bass (1969)model usingthe coefficient of innovation, p, and the coefficient of imita-tion, q, to describe the sub-processes within the adoptionprocess:

probability of adoption at time t, Pt

=

p + q

Y(t−1)

m

x (t) ; (5)

The expected number of adoptions at time t is (m − Yt−1)Pt . The function x(t), which is not used here, reflects theeffects of marketing variables; this development is due toBass, Krishnan, and Jain (1994). Here, the parameters areagain estimated using non-linear least squares. This ap-proach is similar to Method 2, as it focuses on new sub-scriptions per period. The method is included because it isa discrete representation of the logistic.

To summarise the choices that arise when modellingdiffusion: the first choice is the functional form of themodel of the cumulative proportion of adopters, F(t); thesecond choice is between F(t) and f (t) as the dependentvariable. As was discussed in Section 2.1, this choice willdepend on the decision maker’s role in the market. Frankchose Method 1 (and Method 1a), where F(t) is definedin Eq. (1) and the dependent variable is defined in Eq. (2).Using the Finnish data, described above, we use Methods1, 1a, 2 and 3 to estimate the model parameters. For eachmethod, we compute the root mean square errors (rmse)in cumulative proportions of adopters (compatible withMethods 1 and 1a) and the rmse in new adopters per period(compatible with Methods 2 and 3). This facilitates com-parisons of the goodness of fit and forecasting accuracy.

Parameter estimation in themodel is typically achievedby non-linear least squares regression. The constraint,m ≤

1, was applied because the solution was unstable withoutit. In the absence of this constraint, for Method 1, the mar-ket potentialm increased without a bound and a and b de-creased; that is, the in-sample fit improved at the expense


Fig. 1. Cumulativemobile subscriptions in Finland. The estimation region is 1980–1997, and the forecast region is 1998–2004. The estimation is viaMethod1, using a cumulative logistic model.

Table 1A summary of the different estimation methods for the diffusion of mobiles in Finland.

Method1 1a 2 3Cumulativelogistic

Cumulative logisticwith GDP covariate

Non-cumulativelogistic

Bassmodel

Fitted region 1980–1997 m 1.00 0.74 1.00 1.00a/a/a/p −7.83 −4.85 −10.26 0.00b/b0/b/q 0.44 −0.046 0.58 0.56b1 0.003

Cumulative rmse (in) 1.32 0.20 2.87 5.75forecast rmse 2.19 12.57 4.16 21.721998–2004 mape 2.70 11.72 5.15 26.11

Non-cumulative rmse (in) 0.89 0.20 0.44 0.96forecast rmse 1.45 3.64 2.58 3.211998–2004 mape 14.60 53.36 27.48 36.60

The parameter estimates are given in the top half of the table. The in-sample fit and the out-of-sample forecasting accuracyare shown in the bottom half of the table, measured in terms of both cumulative subscriptions and subscriptions per annum.The objective used by each method is in bold.

of out-of sample accuracy. For pragmatic estimation rea-sons, this constraint is applied to all methods to ensureconsistency. We discuss the interpretation of the value ofm in Section 2.1.3.

All four methods are used to estimate the relevantmodel parameters, and the results are summarised in Ta-ble 1. We report forecasting accuracies using the rmse andthe mean absolute percentage error (mape), both in cu-mulative terms, consistent with Methods 1 and 1a, and interms of annual subscriptions, consistent with Methods 2and 3. We note firstly that Method 1 produces very ac-curate forecasts and has the lowest error measures underboth the cumulative subscriptions and subscriptions perannum criteria. Frank produced forecasts usingMethod 1a.In our analysis, the in-sample fit is almost perfect for thecumulative subscriptions, but the forecasting is strikinglypoor for both cumulative and annual subscriptions. Here, itappears that the use of GDP as an explanation for the fluc-tuations in cumulative subscriptions has over-fitted thedata at the expense of estimating the long-term trend cap-tured by Method 1. Method 2 over-estimates the cumula-tive subscriptions, leading to a poorer forecasting accuracythan for Method 1. The Bass model is estimated in Model

3, giving the coefficient of innovation, p, as zero, which in-dicates that the Bass model has been reduced to a logistic.TheBassmodel produces the least accurate forecasts,whileMethod 1, the cumulative logistic, produces themost accu-rate forecasts, regardless of whether the accuracy is con-sidered for cumulative or non-cumulative data. The data,fitted and forecast cumulative subscriptions for Finland,are shown in Fig. 1.

Frank’s out-of-sample forecast (usingGDPas a covariateand assuming 2.5% growth per annum) is accompaniedby a prediction interval. Unfortunately, the interval seemsto relate only to uncertainty in parameter estimation,ignoring the uncertainty due to the error process. Thus,the interval is optimistically narrow: for 2004, the forecastis 90.6% ± 1.2%. In Fig. 1, we show a 95% predictioninterval using the approach described by Meade and Islam(1995); the forecast is 92.6% and the prediction interval is(89.0%–95.5%). In 2004, the actual cumulative proportionof the population with mobile subscriptions was 95.4%.

In our analysis of the Finnish data, our results were sim-ilar but not identical to Frank’s, due to the use of differ-ent software and slightly different data. Our most accurateforecastswere obtainedusing the cumulative logisticwith-out explanatory variables. A side benefit of this is that the


extrapolation of the cumulative logistic does not requirethe forecasting of GDP growth. We also demonstrate that,for forecasts over long horizons, prediction intervals thatreflect the uncertainty due to both the error process andparameter estimation add valuable information.

2.1.2. Analysis of the generality of the conclusions from the‘Frank’ analysis

In the previous section, for the Finnish data on mobilediffusion, we found that the most accurate forecasts weregenerated by Method 1, fitting a logistic to the cumulativeadoptiondata. Thenextmost accuratemethodwasMethod2, fitting the non-cumulative logistic to the annual adop-tions. The Bassmodel,Method 3, also fitted to annual adop-tions, was the least accurate. To test the generality of thesefindings, we carried out a comparison using the ITU datafor other countries. In order to get well behaved data, weapplied two filters: the country must have at least 10 non-zero observations up to and including 1997, and the coun-try must have a penetration of at least 10% by 1997. Thereare 25 countries which meet these criteria. The Finlandanalysis was therefore repeated using the data from thesecountries; the data available up to and including 1997wereused for model estimation, and the forecasting accuracywas measured using the data from 1998 to 2004. The re-sults are summarised in Table 2. As was done for Finland,the forecasting accuracy for each country in the data set ismeasured by the rmse and mape for both cumulative andannual subscriptions. The relative forecasting accuracy ofeach method is examined by ranking the methods accord-ing to these four error measures. The cumulative logistichas the lowest average rank under each of the four criteria.The non-cumulative logistic is second most accurate ac-cording to three criteria and of equal accuracy with thecumulative logistic according to the mape of annual sub-scriptions. The Bass model is least accurate according to allfour criteria. Depending on the accuracy criterion, the cu-mulative logistic was most accurate for 12–15 countries,the non-cumulative logistic was most accurate for 8–9countries, and the Bass model was most accurate for 2–4countries. In terms of the median mape for the cumulativedata, there is little difference between themethods, but theBass again performs poorly for the more volatile medianmape for annual subscriptions (due to the smaller divisors).

In summary, this analysis suggests that the methodol-ogy employed by Frank in fitting the cumulative logisticusing Eq. (2) is more likely to produce more accurate fore-casts than the non-cumulative logistic or the Bass modelfitted using Eq. (4) for mobile diffusion data.

If we look in more detail at the objective function asso-

ciated with Eq. (2), MinimiseYt − F̂ (t)

2, this is equiva-

lent toMinimiseξ̂t

2. If we express these deviations of the

cumulative subscription data from the cumulative logistic,ξ̂t , in terms of the deviations of the annual subscriptionsfrom the non-cumulative logistic, ε̂t , we find the following.

Minimiseξ̂t

2

≡ Minimise

t

i=1

ε̂i

2

≡ Minimise

t

i=1

(t + 1 − i) ε̂2i

+ 2t−1j=1

tk=j+1

(t + 1 − k) ε̂jε̂k

.

It is clear that minimising the cumulative errors is similarto minimising the weighted sum of the annual, period toperiod, errors, with the highest weighting on the earliesterrors and the lowest weighting on the deviation betweenthe fitted non-cumulative model and the most recentobservation. This strong anchoring of the fitted (bell-shaped) curve to the earlier data seems to be a successfulstrategy for the mobile diffusion data.

2.1.3. Definition of market potentialMarket potential, m, is defined in Section 2.1.1 as the

proportion of a population who will (eventually) becomesubscribers. Being defined as a proportion, m is clearlybounded above by unity. In these studies, the problemarises that, typically, the data available are the numbersof subscriptions rather than of subscribers. As the mobilemarket has developed, the average number of subscrip-tions per subscriber has reached two or three in someEuropean countries. Looking at the diffusion of 2G mo-biles in Greece, Michalakelis, Varoutas, and Sphicopoulos(2008) found that all of the models considered estimatedthe market potential to be in excess of 100%, which is notsurprising, as the later data points exceed 100%. As morepeople take out multiple subscriptions, the upper boundon market potential begins to exceed 100%; for example,in 2009 Q1 in Finland, penetration was 141.2% (subscrip-tions/person).

The application of an adoption model to subscriptiondata worked satisfactorily when most adopters took outonly one subscription. However, post-2007, the averagenumber of subscriptions per person exceeded unity inmany countries. This means that an upper bound on themarket potential, m, can no longer be identified. Since theestimatedmarket potential is a crucial determinant of longterm forecasts, and particularly of the forecasts of cumula-tive adopters that are of interest to service providers, theaccuracy of its estimation is key. Further research is nec-essary here, but one possible approach would be to modelthe taking out of the first subscription as an adoption pro-cess (where m is bounded by unity), and to model theacquisition of further subscriptions as a separate process.However, it is possible that the phenomenon of multiplesubscriptions per person may disappear or at least dimin-ish in future. The main motivation for multiple subscrip-tions is to ‘game’ the tariffs available. For example, tariff Amay be cheaper for domestic calls, tariff B may be cheaperfor international calls, and tariff C may be cheaper for data.If tariffs become simpler, for example with fixed monthlypayments regardless of usage, then the motivation behindmultiple subscriptions is reduced considerably.

In the context of fixed line telephony, the issue of mul-tiple subscriptions has been a minor issue; however, extra


Table 2A summary of the relative forecasting accuracies of three methods using data describing the diffusion of mobiles in 25 countries.

Method Cumulative errors Non-cumulative errorsRanking by rmse Ranking by mape Median mape Ranking by rmse Ranking by mape Medianmape

1. Cumulative logistic 1.76 1.72 29.6 1.64 1.72 78.02. Non-cumulative logistic 2.04 2.08 30.0 1.76 1.72 86.33. Bass 2.20 2.20 29.4 2.60 2.56 242.7

lines were acquired by some households during the 1990sto facilitate internet use. Duffy-Deno (2001) studied thedeterminants of an additional fixed line using US data from90 cities in 1998. Using a logistic regression, the significantvariables were price (the price elasticity was greater for anadditional line than for the initial line) and the age of thesubscriber (those aged 55–64 were most likely to acquirea further line). In a similar vein, Eisner and Waldon (2001)used a binomial probitmodelwithUS data from1995; theyfound that households that already subscribed to onlineservices were more likely to subscribe to a further line.

2.1.4. Summary of similar studiesSeveral authors have followed Frank’s example in de-

veloping single-country forecasts for the diffusion of mo-bile telephony. These analyses are summarised in Table 3 inorder of country. The technology considered and the timespan of the data used are given. The models used are mostcommonly the logistic, Gompertz and Bass models. (Un-like the logistic, the Gompertz represents asymmetric dif-fusion, with the peak adoption occurring before 50% of themarket potential is reached. Its cumulative proportion ofadoption is:

F (t) = m exp (−c (exp (−bt))) , (6)

where m is the market potential, and b and c are positiveparameters to be estimated from the data.)

In some cases, the model was estimated without a fore-cast being made, with the objective in some cases beingto discover the market potential, and in others to discoverwhether or not mobile telephony was acting as a substi-tute for fixed line telephony. In contrast to Frank (2004),Gamboa and Otero (2009) provide a plausible predictioninterval for mobile telephony diffusion in Colombia usinga bootstrap approach.

Islam and Meade (1996) explored several formulationsof the market potential for UK business telephones us-ing GDP-related variables. They found that, although moremanagerial insights are achieved using explanatory vari-ables, these additional insights do not necessarily lead tomore accurate forecasts, due to the need to forecast theexplanatory variables (an explanation of the poor perfor-mance of Method 1a described in Section 2.1.1). In anotherstudywith an emphasis on the estimation ofmarket poten-tials, Chen andWatanabe (2006) looked at diffusionwithinthe Japanese ICT market, considering both mobile and in-ternet access data sets. They linked the drop in cumulativefixed line subscriptions during 1998 to substitution bymo-bile telephony. Evidence that mobile telephony is acting asa substitute for fixed line telephony has also been found byFrank (2004) in Finland, Lee and Cho (2007) and Sung andLee (2002) in Korea, and Chu, Wu, Kao, and Yen (2009) inTaiwan.

Wu and Chu (2010) use Taiwanese data to test theirhypothesis that the most appropriate diffusion modelchanges as diffusion progresses through the stages of slowstart, take-off, post-inflection and saturation. The authorsdefine the beginning of the take-off stage as the timewhenadoption reaches 10%–20%, following Rogers (2003). Moreexplicit definitions of ‘sales take-off’ are given by Agar-wal and Bayus (2002), Lim, Choi, and Park (2003), Tellis,Stremersch, and Yin (2003), and Wiorkowski and Gylys(2006). After considering the logistic, Gompertz, Bass andARMA models, Wu and Chu find that the Gompertz givesa greater forecast accuracy in the pre-take-off stage, whilethe logistic is more accurate in the post-take-off stage. AsWu and Chu note, the appropriate choice of model is case-dependent (see also Meade & Islam, 2001). However, theidea of stage-dependence (within each case) needs fur-ther research, particularly considering the width of theprediction intervals associated with the point forecastsconsidered by Wu and Chu. In the early stages, the mostimportant finding is likely to be the scale of the uncertaintyin the forecast, whichever model is used.

Some authors discuss network effects or network exter-nalities in mobile telephony: that is, they consider the in-crease in benefits tomobile users as the size of the networkincreases. In terms of the Bassmodel, this effect is capturedby the coefficient of imitation, and there is a correspondingparameter in the logistic model. Doganoglu and Grzy-bowski (2007) find that, in Germany, the mobile subscrip-tion demand depends on lagged cumulative subscribersand service prices. In amicro-level study on the adoption ofPCs, Goolsbee andKlenow (2002) found that a high fractionof surrounding PC owners was the most important deter-minant of buying a PC. Primarily, this is evidence of a highcoefficient of imitation, although there may be some net-work effect related to internet access.

Summarising the accuracy of the analyses identified inTable 3 is not straightforward. In only six of the 14 anal-yses identified is it possible to find a measure of the out-of-sample forecasting accuracy. When found, thesemeasures may be averaged over different origins and/ordifferent horizons. Taking these reservations into account,themapes found for the most accurate forecasting methodconsidered are between 5% and 10%; these measures re-fer to horizons of three years or less. However, we see inTable 2 that, for the average mape over horizons of one toseven years, themedian value over 25 countries is 30%. Themape value obtained for Finland (2.7%) is atypically low. Ingeneral, the more typical pattern is that the forecasting ac-curacy deteriorates sharply for horizons in excess of fouror five years.


Table 3A summary of single-country analyses of ICT diffusion.

Authors Country Technology Time period Models used MarketPotential

Significantcovariates

Forecast (Y/N)

Gamboaand Otero(2009)

Colombia Mobile phone 1995 Q4–2008Q1

Logistic,Gompertz

Unconstrained NA Evaluation for 1year out-of-sample

Frank(2004)

Finland Mobile phone 1981–1998 Logistic Proportion ofpopulation

Fixed line(substitution)

12-yearextrapolation andprediction interval

Michalakeliset al. (2008)

Greece Mobile phone 1994–2005 Q3 Versions ofBass, Gompertzand logistic

Unconstrained NA Evaluation for3 yearsout-of-sample

Dergiadesand Dasilas(2010)

Greece Mobile phone 1993–2005 Gompertz,logistic andARMA

Time varyinganddeterminedexogenously

NA Prediction intervalby bootstrap

Singh(2008)

India Mobile phone 1995–2006 Logistic,Gompertz

Constrainedandunconstrained

NA Variousextrapolations

Massini(2004)

Italy Mobile phone 1990–2001 Logistic,Gompertz

Either rate ofdiffusion orsaturation levelfunction ofeconomiccovariates

Digital dummy No

Chen andWatanabe(2006)

Japan Fixed linetelephony,mobiletelephony,internet access

1991–2002(mobiles)

Versions oflogistic,choice/diffusion

Unconstrained Fixed line(substitution)

Focus on marketpotentials

Sung andLee (2002)

Korea Fixed linetelephony—connectionsanddisconnections

1991–1998 for 8regions

Log-linear NA Mobile stock hasnegative effecton connectionsand positiveeffect ondisconnections

No: looking forsubstitution bymobiles

Lee and Cho(2007)

Korea Mobile phone 1984–2003 Logistic, ARMA Proportion ofpopulation(sameapproach asFrank, 2004)

Digitalchangeover,GDP, fixed linetelephony(substitution)

Evaluation for 1year out-of-sample

Botelho andPinto (2004)

Portugal Mobile phone 1989–2000 Q1 Exponentialgrowth,logistic,Gompertz

Unconstrained NA Extrapolation only

Wu and Chu(2010)

Taiwan Mobile phone 1989–2007 Bass,Gompertz,logistic andARMA

Unconstrained NA Evaluation for3 yearsout-of-sample

Chu et al.(2009)

Taiwan Mobile phone Bass, Gompertzand logistic

Unconstrained Deregulation,fixed line(substitution)

Focus on marketpotentials

Massini(2004)

UnitedKingdom

Mobile phone 1990–2001 Logistic,Gompertz

As Italy Digital dummy,price, tariff

No

Hwang,Cho, andLong (2009)

Vietnam Mobile phone 1995–2006 Bass, Gompertzand logistic

Unconstrained NA Evaluation for 1year out-of-sample

2.1.5. Other single-country analyses related to ICT diffusionIn an early study, Jeffres and Atkin (1996) looked at

the factors affecting the diffusion of Integrated SystemsDigital Networks (ISDN), whichwas perceived to be slowerthan expected. They found that the more educated wereless likely to use ISDN for entertainment purposes, andthat a heavy news consumption was positively relatedto a desire for ISDN. In a more recent study, Robertson,Soopramaniem, and Fildes (2007) considered the diffusionof residential broadband in the United Kingdom. A postalsurvey was used to gather data about the length of a

householder’s subscription to broadband and income. Theyconsidered the market for household broadband as agroup of segments determined by income, where eachsegment was modelled as a Gompertz diffusion process.Using data up to 2005, their combined forecast, across allsegments, is for a broadband coverage of 95% by 2015. Thisforecast is looking a little optimistic, with Ofcom (the UKtelecommunications regulator) reporting in 2013 that 75%of UK adults had a broadband connection (fixed ormobile).

We conclude this section with a consideration of twostudies relating to topics that have received little cover-


age to date. In a relatively early study of the number of in-ternet providers, Rai, Ravichandran, and Samaddar (1998)considered the following diffusion models: exponentialgrowth, logistic and Gompertz. The choice of models hereis questionable, as there is no obvious upper bound on thenumber of providers. Exponential growthwas shown to bethe better predictor, as the quarterly data showed no signof an inflection. They accept that their modelling approachdoes not include any of the policy variables that may affectdiffusion. Looking ahead, one consequence of the growth inthe usage of ICT hardware, mobile telephones, desktop andlaptop PCs, and tablets is the issue of the sustainable andenvironmentally acceptable recycling or disposal of theseitems as they become obsolete. Yang and Williams (2009)use a logistic model to forecast the generation of obsoletePCs in the US. Their forecast depends on forecasts of de-mand for PCs and assumptions about the average lifetimeof the PC. They predict that there will be between 115 and144million obsolete PCs in 2050, indicating a potentialma-jor shortfall in US recycling facilities. Although their fore-cast focuses on PCs (where demand has stalled at the timeof writing, 2014), alternative technologies, e.g. tablets, willsubstitute for PCs, possibly with shorter lifetimes.

2.2. Multi-country analyses of ICT diffusion

Researchers have analysed ICT multi-country data setswith a variety of objectives. We begin by describing thecase where the objective is to provide forecasts from shorttime series by exploiting the availability of a cross-sectionof data from several countries. Secondly, we consider aset of broadly homogeneous studies. These use a multi-national data set to describe the diffusion of fixed ormobiletelephony or the internet, and evaluate the power of a setof covariates to explain the differences in country diffusionrates. Thirdly, we describe other multi-country analyseswith various different objectives that are not covered in theother sections.

2.2.1. Multi-country analyses with short time seriesIn the analyses covered in this section, the researchers

use cross-sectional data to compensate for a lack of timeseries data. Dekimpe, Parker, and Sarvary (1998) draw at-tention to theweaknesses of using the Bass or similarmod-els with short data series. They emphasise the importanceof sample-matching when using cross-sectional data, inorder to make comparisons across countries meaningful.They propose a staged estimation procedure using sam-ple matching to allow the use of multi-national data formodelling and forecasting diffusion. The covariates thatthey consider important in matching include income, pop-ulation heterogeneity, demographics and politics. Islam,Fiebig, andMeade (2002) usemultinational cross-sectionaldata to compare pooling methods in the production ofgrowth forecasts for digital cellular phones, ISDN lines andfax connections. The pooled forecasts tended to be moreaccurate than the available alternatives. Using a similarapproach, Hamoudia and Islam (2004) model and fore-cast the demand for SMS (short message service, i.e. texts).They pool annual data from seven European countries overfive years from 1998 to 2003 using a linearised Gompertz

model. Using themape as a measure of accuracy, their me-dian (over the countries considered) values are 9.1%, 20.1%and 25.5% over one, two and three years, respectively.

2.2.2. Multi-country diffusion analyses using covariatesIn order to enable more general conclusions about the

drivers of diffusion to be drawn, several authors have car-ried out analyses of mobile diffusion over many countries,where the coefficients of a diffusion model are parame-terised to relate the growth rate or market potential toeconomic or infrastructural variables.We summarise thesestudies in Table 4. This table is laid out in a similar way toTable 3, with two differences: the papers appear in chrono-logical order, and there is no column indicating whetherforecasting was carried out, as these studies generally didnot have forecasting as an objective. One of the most citedpapers is that of Gruber and Verboven (2001), who use alogistic model for the diffusion of mobile telephony in 15EU states from1984 to 1997. They parameterise the logisticusing a range of economic variables; these effects are esti-mated using non-linear least squares on the multi-countrytime series data. Their main conclusion is that the primecause of an increased diffusion was the transition fromanalog to digital; increased competition was also found toincrease the diffusion rate. Their analysis also suggests thatmobile telephony is a substitute for fixed line telephony.Other researchers have developed these themes, namelythe transition between technological generations, the sub-stitution between fixed and mobile telephony, and the ef-fects of competition and other covariates.

Several studies consider the effect of the transitionsbetween generations of mobile telephony. Grajek andKretschmer (2009) study multiple generations of mo-bile telephony and extend the information set from sub-scription data to usage data (average minutes of use permonth). They find that successive generations are substi-tutes when analysed via subscription data, but act as com-plements when considered via usage data. Bohlin, Gruber,and Koutroumpis (2010) examine the factors affecting thediffusion of newer generations ofmobile phones. They findthat income/head, urbanisation and internet penetrationhave a positive impact on diffusion across all generations.However, although the diffusion of the first generation(analog) stimulated the diffusion of the second generation(digital), the diffusion of the second generation did notaffect the diffusion of the third generation. The effect ofinter-firm competition, although important for earlier gen-erations, diminishes by the third generation. A number ofcovariates are identified as determinants of the diffusion ofmobile telephony, including national wealth (using GDP asa proxy), regulation, competition and price. Kiiski and Po-hjola (2002) and Liikanen, Stoneman, and Toivanen (2004)find that an increase in GDP per capita is associated witha faster diffusion. Comer and Wikle (2008) study the rea-sons for the differences in the rates of diffusion of mobilesbetween different countries. Using the growth rate of sub-scriptions between 1995 and 2005 in each country as thedependent variable, they find that the most significant ex-planatory variable is GNP. Using a worldwide data set, 75%of the variation is explained by this variable; using onlydata from Asia, this rises to 90%. Koski and Kretschmer


Table 4A summary of multi-country analyses of ICT diffusion.

Authors Countries Technology Time period Models used Market potential Significant covariates

Gruber andVerboven (2001)

15 EU countries Mobile phone 1984–1997 Logistic Too early toestimate accurately

Digital/analogchangeover; number ofcompeting suppliers.

Kiiski and Pohjola(2002)

23 OECDcountries

Internet 1995–2000 Gompertz Time dependentfunction of cost andlagged adoption

Diffusion associatedpositively withGDP/capita andnegatively with accesscost

Madden andCoble-Neal (2004a)

56 countries Fixed lineand mobilephone

1995–2000 Dynamicpanelregression

NA Evidence of substitutionfor fixed line, significantincome and price effects

Liikanen et al.(2004)

80 countries Mobile phone 1992–1998 Logistic Estimated asproportion ofpopulation

GDP has positive effect,evidence of fixed linesubstitution

Jang et al. (2005) 29 OECD coun-tries + Taiwan

Mobile phone 1980–2001 Logistic Unconstrained Digital/analogchangeover; number ofcompeting suppliers,fixed line (substitution)

Sundqvist, Frank,and Puumalainen(2005)

64 countries(results based on25)

Mobile phone 1981–2000 Bass Not clear Cluster analysis used torelate diffusion withcultural dimensions

Koski andKretschmer (2005)

32 industrialisedcountries

Digitalmobiles (2G)

1991–2000 Log(diffusion) asdependent

Standardisationaccelerates diffusion,price competition greaterbetween standards thanwithin

Rouvinen (2006) 78 countries Mobile phone 1991–2000 Gompertz Constrained Comparison of developedand developing countries— competition speedsdiffusion, conflictingstandards retardsdiffusion

Bagchi et al. (2008) 4 multi-countryregions + Indiaand China

Fixed lineand mobiletechnology

1975–2003and1985–2003

Priceadjustedlogistic

Price dependent Price and social/politicalcovariates

Comer and Wikle(2008)

206 countries Mobile phone 1995–2005 Regressions Not clear GDP (log and growth rate)

Grajek andKretschmer (2009)

Up to 157operators in 41countries

Multiplegenerationsof mobilephones

1998–2004 Logistic Estimated asproportion ofpopulation

Heterogeneity of adoptersmasked network effectsin usage data

Bohlin et al. (2010) 177 down to 67countries

3 generationsof mobilephones

1990–2007 Logistic Used rate ratherthan penetration

Income/head,urbanisation, internetpenetration

(2005) look at the effects of regulation and competitionon digital (2G) mobiles. They find that a greater stan-dardisation accelerates both the market entry and thesubsequent diffusion; in addition, they also find that theliberalisation of fixed line telephony accelerated the entryof digital. Jang, Dai, and Sung (2005) and Rouvinen (2006)find that the number of suppliers, a measure of compe-tition, speeds diffusion, while Rouvinen (2006) also findsthat competing technical standards retard diffusion. Chinnand Fairlie (2007) use panel data from 161 countries overthe period 1999–2001 to compare the determinants of thepenetration of the PC and of the internet. They find thatthe income differential is an important variable in explain-ing the penetration variability for both technologies. Forthe internet, the quality of regulation has an important in-fluence on penetration, followed by the telephone density.For both technologies, they find that the influence of ed-ucation is swamped by those of the variables mentioned.To discover the reasons for the differences in the rate oftechnology diffusion of 3G mobile phones across coun-tries, Islam and Meade (2012) investigate the impacts of

market factors, measured by competitive fractionalization,and economic globalization across 35 countries using amulti-country diffusion model. They use non-linear mixedmodelling with pooled multi-country data to estimate ageneralized Bass model, taking into account the unob-served heterogeneity inmarket saturation levels. They findthat increasing the competitionwithin themarket tends toincrease the market potential, and also find that countrieswith a higher economic globalisation index (e.g. a greateropenness to trade) are associated with higher rates of dif-fusion.

The effect of prices on the rate of diffusion of mobiletelephony has been approached in several different ways.Madden and Coble-Neal (2004a) use an economic agentapproach to model the substitution effect between fixedline and mobile telephony. Their data covers 56 countriesover the period 1995–2000. They find a significant sub-stitution effect, together with a significant price effect,although the network effect was greater in terms of caus-ing increased subscriptions. Bagchi, Kirs, and Lopez (2008)


study the impact of price decreases on the diffusion of mo-biles and fixed telephony in six large regions. They use theprice-adjusted logistic model of Gurbaxani andMendelson(1990), and essentially modify the potential by introduc-ing a multiplicative factor reflecting the price; thus, Eq. (1)becomes

F (t) =m exp (−α ln (Pt/P0))(1 + exp (− (a + bt)))

, (7)

where Pt reflects the price at time t . If the coefficient αis found to be significantly greater than zero, this indi-cates that the potential level of subscriptions has been in-creased due to the price decreases. They found that pricedecreases had a greater effect in lower income nations, es-pecially if the countries are politically stable, with high ed-ucation levels. Islam and Meade (2011) used a parametrichazard model and discrete-time survival mixture analysis(DTSMA) approaches to model the hazard probability ofthe time to sales takeoff for cellular analog telephony in70 countries. They determine the impacts of three marketfactors: price, the number of competitors, and the num-ber of competing standards. The authors showed therelative advantage of DTSMA in its ability to recognizeunobserved heterogeneity using latent classes. A failureto account for unobserved heterogeneity can cause anunderestimation of the hazard probabilities. They foundthat relatively falling prices, relatively greater numbers ofcompetitors, and relatively fewer competing standards areeach associatedwith relatively higher hazard probabilities.

2.2.3. Other multi-national analyses related to ICT diffusionRestrictions on supply have been an obstacle to the dif-

fusion of fixed line telephony in many countries. Islam andFiebig (2001) carried out a multinational study to estimatesaturation levels for fixed line telephony in 46 supply-restricted countries. They used pooled cross-sectional es-timates to forecast the supply-restricted demand wherelittle or no data were available. Bartels and Islam (2002)analyzed data from 28 countries to investigate the causesof supply restrictions for main telephones, and found thattechnical efficiency is the major determinant of supplyrestrictions.

Looking at a different aspect of mobile telephony, Shinand Bartolacci (2007) consider the diffusion of mobilevirtual network operators (MVNO). AMVNOprovides userswith mobile services without its own government-issuedbandwidth. The analysis is a cross-sectional regression oftheMVNOmarket share, explained by factors that describethemarket structure. They find that diffusion is retarded byvertically integrated markets, which are common in Asia,and enhanced by horizontal market structures, which arecommon in Europe.

Gruber and Koutroumpis (2013) examine the effectsof regulation on the adoption of broadband in a country.The regulation of former national telecommunicationsmo-nopolies is studied: to encouragemarket entry, incumbentoperators were instructed to share, sell or split their in-frastructure. Data for 167 countries over an11-year periodare used to relate the diffusion (using a logistic model)of broadband to regulatory interventions. They found that

competition between firms on the former monopoly’s net-work tended to accelerate the adoption of broadband,whereas competition between different access technolo-gies did not.

2.3. Choice modelling and other extensions

In circumstances where time series data are in shortsupply, or absent, as in the case of a new product, it maybe appropriate to survey consumer intentions. Survey dataare then used, with orwithout time series data, to estimatethe parameters of a diffusion model. Kumar, Nagpal, andVenkatesan (2002) combine amarket share surveymethodwith an overall Bass-type projection of market size topredict the demand for mobile phones at a firm level.

Jun and Park (1999) propose a choice-based approachwhere there is a utility associated with each availablechoice. In cases where several generations of a technologyare available simultaneously, the choice could be betweeneach of these generations, or non-adoption. At time t , theexpected utility for choice k is Vkt , and a multinomial logitis used to give the probability, Pkt , of an individual makingchoice k at time t , where Pkt =

exp(Vkt )j exp(Vjt)

. The expected

utility is parameterised, for example Vkt = ak (t − τk),where τk is the time elapsed since the launch of the tech-nology. The expected number of adoptions of choice k attime t is

mk − Yk,t−1

Pkt . The information source for the

utility function can be a survey-based approach if no dataare available, using conjoint analysis, for example. Alter-natively, if adoption data are available, the utilities may beestimated from these data. Thismodel is developed furtherby Jun, Kim, Park, Park, andWilson (2002), who extend thechoice-based diffusion model to include both substitutionand competition. Ida and Kuroda (2006) study the demandfor broadband in Japan using a discrete choice model. Theyfind that the demand for the main source of broadband,ADSL (asymmetric digital subscriber line), is price inelastic,in contrast to the demand for cable TV and fibre to homeservices, which have been found to be price elastic.

An example of forecasting for a new product is pro-vided by Jun et al. (2000), who forecast the demand fora new service: a low earth orbit mobile satellite service(LEO). They use a mixture of analogy and surveys to fore-cast demand. The analogy is the use of mobile phone pen-etration data in Korea to provide estimates for the Bassmodel describing the LEO service penetration. The surveydatawere used to estimate themarket potential of LEO ser-vice and the subscription probability using a logit choicemodel. The authors used the diffusion parameters of theexistingmobile phone service and themarket potential de-rived from the survey tomakemarket predictions using thediffusion model. The forecast is for approximately 280,000subscribers in 2005; however, it is undermined by the au-thors’ comments that the survey is probably too optimistic,given that it had been taken five years prior to publicationand during a period when the Korean economy was muchmore buoyant. Their reservations appear to have been jus-tified, as therewere only 20,000 subscribers in 2009 (KoreaIT Times, 7 April 2009).


In forecasting the demand for large screen televisions,Lee, Cho, Lee, and Lee (2006) adopt the following sequen-tial approach. They model utility using conjoint analysis;model the price of television as a decreasing function overtime; produce a dynamic random utility function; and usea Bassmodel via the estimation of p, q andm based on timeseries data from Q1 1998 to Q2 2003 for the whole largescreen market. Finally, forecasts for each television typeare derived by multiplying the total demand by the dy-namic choice probability. Kreng and Wang (2009) use theJun and Park model on shipment data to model the substi-tution of liquid crystal displays (LCD) for cathode ray tele-visions (CRT). The choice considered is between a CRT, amedium-size screen LCD and a large screen LCD. In addi-tion to a fixed market potential, they also consider a vari-able market potential that changes with exp(g × time),where g is an estimated parameter. They demonstrate thatthe introduction of the variable market potential improvesthe forecasting accuracy. This modification of the marketpotential can be considered as a proxy for the changing rel-ative prices of CRTs and LCDs, and the move from analogbroadcasting to digital broadcasting (which is inaccessiblefor CRT).

Jeon, Kim, and Sohn (2010) look at the convergence ofdigital products that incorporate the following into mobiletelephones: a PC, a camera, a music player (MP3), andmedical devices. The study used conjoint analysis with aKorean panel, and found that the only converged productfor which a premium could be charged was computingpower (essentially faster data reception/transmission).

In the last decade, there has been a high growth inmobile broadband services—3G systems and wireless lo-cal area networks (WLAN). The former typically have goodcoverage but are limited by low transmission speeds andhigh usage costs. WLAN have good transmission but lim-ited coverage. An emerging technology is portable inter-net services (PIS), which are characterised by high speedtransmission, portability, mobility and multi-media com-munication. An example of this is WiBro, which was intro-duced in Korea in 2006 (see Nam, Kim, & Lee, 2008). Namet al.’s study is in two parts. Firstly, the characteristics ofmobile broadband used by consumers, choosing between3G, WLAN and WiBro, are identified and survey data col-lected. A conjoint analysis is then carried out to predictrelative market shares. Secondly, a Bass model is used toforecast diffusion. In the absence of time series data, theBass coefficients are chosen by an expert panel.

VoIP (voice over internet protocol) telephony is a tech-nology that gained acceptance during the 1990s, initiallyas a novel use of the internet and later as a serious con-tender with other forms of telephony. Ida, Kinoshita, andSato (2008) used conjoint analysis with Japanese data, andfound that VoIPwas considered an additional option for in-ternet use rather than a substitute for fixed line telephony.However, in a later study in Korea Kwak and Lee (2011)found that the picture was evolving. Due to an increasedinvestment in VoIP by telecoms suppliers, they found thatthe call rates for VoIP and fixed lines were the factors thataffected the VoIP demand the most. Thus, at the level ofdeciding which medium to choose for making a call, VoIPis a substitute for fixed lines. We did not find any studies

more recent than that of Ida et al. (2008) that have lookedat the decision at subscription level, investigating whetherinternet users are prepared to forego a fixed line. The out-comeof these national analyseswill be affected by the levelof competition between the different service suppliers andthe effect of this on the pricing of both the internet andfixed line telephony. For example, in the United Kingdom,both services are often supplied by the same company. Asimilar argument can be used with VOIP over the mobileinternet andmobile telephony, in that differences in tariffswill tend to determine preferences.

Lee, Lee, and Kim (2008) look at the diffusion of homenetworking. This is the distribution of data between ICTappliances for providing communications, multimedia andautomation services anywhere in the home. Conjoint anal-ysis is used with survey data from both consumers andconstruction companies (who install the system in newlybuilt housing). The estimated utilities are used with a lo-gistic model, and incorporate expert judgement into theirforecast via Bayesian updating.

2.4. Summary: the flow of benefits between ICT and forecast-ing methodologies for diffusion

Here, we look at the flow of benefits between ICT appli-cations and diffusion modelling and forecasting method-ologies.

In the case of studies of a single series, their ICT applica-tions, mainly the forecasting of national mobile telephonediffusion, have led to some methodological developments,mainly in the area of parameterising market potential. Ona related topic, we note that there is an opportunity for fur-ther research in the appropriate modelling of subscriptiondata, rather than treating them as if they were adoptiondata. In general, studies have focussed on the trend repre-sented by the S-curve,with little attention being paid to theassociated error process. The analysis in Sections 2.1.1 and2.1.2 shows that a consideration of the error process has asignificant effect on the estimated trend, and consequentlyon the forecast accuracy. In addition, the error processfeeds into the construction of a prediction interval (used byonly two of the 14 studies in Table 3), giving valuable infor-mation about the uncertainty associated with the forecast.

In general, the multi-country diffusion studies focus onidentifying the determinants of diffusion, rather than ondirect forecasting. Although no particular methodologicalinnovations have been driven by the use of ICT data,the findings of these studies successfully illuminate thefactors that determine the diffusion of ICT products. Theidentitifcation of these factors allows business strategiststo choose appropriate timetables for their entry intointernational markets.

ICT has provided an important impetus to the devel-opment of choice modelling. The models developed haveprovided practicable solutions to the problem of forecast-ing with little or no data. The nature of ICT devices, such asmobile telephones or telephones, is well suited to choicemodelling. These devices are available in a small numberof basic types, and each type is available with a range ofextra capabilities. This device structure facilitates the con-struction of consumer intention surveys that will not only


provide the basis for modelling and forecasting, but alsofeed into product design. However, the Korean LEO exam-ple shows the importance of timeliness and relevance inthe survey questions; any compromise on either of theseissues is likely to undermine the forecasting accuracy. Fur-thermore, in none of the cases in which forecasts werebased on survey information and little or no time seriesinformation was any attempt made to quantify the uncer-tainty in the forecast. Further research in this area wouldadd to the value of this methodology.

3. Time series

The forecasting of time series is undoubtedly themost heavily researched area in forecasting; see De Gooi-jer and Hyndman (2006) for a review. The questionwe ask here is: is there anything distinctive about anICT/telecommunications time series that justifies specialconsideration? This section attempts to answer this ques-tion. In Section 3.1, we discuss a clearly defined and rele-vant activity, forecasting in call centres. In Section 3.2, wereview the literature on the forecasting of ICT-related timeseries. We draw our conclusions in Section 3.3.

3.1. Forecasting for call centres

Scheduling the manpower available to staff call cen-tres over short, medium and long horizons is an ongo-ing problem for many service organisations. For furtherbackground to this topic, Gans, Koole, and Mandelbaum(2003) provide a good overview of call centre operationsand their management; a bibliography of call centre re-lated studies was compiled by Koole (2008); and Mandel-baum (2003) summarises some more recent advances inmanagement science applied to call centre management.Since personnel costs account for around two-thirds of callcentre operating costs, effective manpower planning re-lies on a good forecasting accuracy. The relevant forecastsare over a range of horizons, from two or three months,to one to three weeks, to updating within-day forecasts inreal time. We will begin by discussing the different typesof call centres that have been examined, and then considerthe modelling approaches used.

The modelling and forecasting of call frequencies hasbeen studied for several different types of centre. In Ta-ble 5, we list studies on forecasting for call centres cate-gorised by call centre type, where the modelling approachused for each study is identified. The call centre forecastingstudies fall into four categories. The first is emergency callcentres, which require a zero or very short waiting time(fire, police, ambulance, and health-related enquiries). Thesecond is enquiries, and includes directory enquiry callsto telephone companies. The third category is telemarket-ing, where sales of goods or services are made followingthe receipt of a catalogue or the sight of an advertisement.The fourth category is call centres run by financial institu-tions. These categories are not mutually exclusive; for ex-ample, some financial institutions may run telemarketinginstitutions. The practical value of accuracy is clear inmanytypes of call centre forecasting: an under-estimation ofcall frequencies leads to under-staffing, which may lead to

either delays in dealing with emergencies or lost sales;over-estimation leads to over-staffing, which equates toexpensive wasted capacity. In general, the earlier data setsused were confidential. This is unfortunate, as alternativemodelling approaches to those used in the original studiescannot be evaluated or compared. However, there has beensome comparison of different forecastingmethods over thesame data sets in the more recent studies.

According to several authors, including Shen andHuang(2008a) andWeinberg, Brown, and Stroud (2007), themostcommonly used model for the number of calls arriving at acentre isNij + 1/4 = µ + αi + βj + εij, (8)

where Nij calls arrive during the jth sub-section of day i.The error term, εij, is assumed to be normally distributed,implying that the number of calls arriving in a particulartime slot is a non-central chi-square random variable. Sep-arate considerations of the inter-day and intra-day com-ponents, αi and βj respectively, are common tomost of themodelling approaches used. The differences between ap-proaches lie in the choice of the random variable and theestimation procedure.

Early analysts, such as Andrews and Cunningham(1995), Bianchi, Jarrett, and Hanumara (1993, 1998), andMabert (1985), used well-known time series methodssuch as Holt–Winters and ARIMA for forecasting daily callcounts. Later, more customised modelling was carried outby Tych, Pedregal, Young, and Davies (2002), who devel-oped an unobserved components model for calls to a fi-nancial services call centre. With the availability of moredetailed data, Avramidis, Deslauriers, and L’Ecuyer (2004)identified four properties of the arrivals of calls at call cen-tres. These are: (a) the variance of the total number of callsper day is greater than its mean; (b) call arrival rates varyduring the day; (c) there is a positive correlation betweencall counts in different periods of the day; (d) there aresignificant correlations between call counts on successivedays. To capture properties (a), (b) and (d), they develop aninhomogeneous Poisson process model of call arrivals, us-ing data from a Bell Canada call centre. Brown et al. (2005)carry out a statistical analysis of three aspects of call centremanagement, using data from an Israeli bank call centre.Forecasters’mainpreoccupation iswith the arrival process,which Brown et al. consider as an inhomogeneous Poissonprocess. Furthermore, Brown et al. also consider the othertwo components of a queuing system: the customer wait-ing time and the duration of service. Extending the workof Brown et al. (2005), Weinberg et al. (2007) develop aBayesian approach for estimating the parameters of theinhomogeneous Poisson process in order to provide den-sity forecasts for daily arrival rates; their data were froma US bank call centre. Shen and Huang (2008a) considerthe data as a vector time series, with daily observationsof the intra-day profile (represented by the frequency perquarter-hour slot over the business day). Their data weregathered from a US financial institution call centre. Theyuse singular value decomposition to reduce the dimension-ality of the vector, then use an autoregressive model toforecast the resulting set of time series. They demonstrate


Table 5A summary of forecasting-related studies of call centre activities.

Type of call centre Authors Modelling approach

Emergency services

Mabert (1985) Holt–Winters variants and ARIMATandberg (1995) Emphasis on time-of-day and day-of-week effectsKlungle and Maluchnik(1997)

Linear regression

Taylor (2012) Density forecasts based on Poisson models

EnquiriesSparrow (1991) Queuing theoryXu (1999) Trend extrapolationAvramidis et al. (2004) Poisson based stochastic models

Tele-marketing

Bianchi et al. (1993) Holt–Winters and ARIMAAndrews andCunningham (1995)

ARIMA and transfer function

Bianchi et al. (1998) Holt–Winters and ARIMASoyer and Tarimcilar(2008)

Emphasis on effect of marketing campaigns

Banking/financial services

Betts, Meadows, andWalley (2000)

Linear regression

Antipov and Meade(2002)

Emphasis on advertising and day-of-week effects

Tych et al. (2002) Unobserved components/seasonal ARIMABrown et al. (2005) Queuing theory approach to modelling arrivals, customer patience and service durationShen and Huang (2005,2008a,b)

Singular value decomposition is used to condense the dimensionality of the intra-dayprofile

Weinberg et al. (2007) Bayesian forecasting of an inhomogeneous Poisson processTaylor (2008) Seasonal ARIMA, an extended Holt–Winters approach plus some regression based

approachesTaylor (2012) See above

that their approach is clearly more accurate than the his-torical average model (see Eq. (8)) on an intra-day basis.Furthermore, using the same data set as Weinberg et al.(2007), Shen and Huang demonstrate that the accuracy oftheir method is very similar to that achieved by Weinberget al. Shen and Huang (2008b) further refine their theme ofa reduced dimensionality of the daily profile, in conjunc-tion with the assumption of an inhomogeneous process.

The question arises as to whether the conventionaltime series approaches used by earlier authors can bedeveloped usefully to offer similar accuracies to these laterapproaches. Broadly speaking, the answer appears to bein the affirmative. Taylor (2008) compares the forecastingaccuracies of several approaches over six series of intra-day call arrival data, five series from a UK bank, and onefrom an Israeli bank (as used by Brown et al., 2005),over horizons ranging from 30 min to two weeks. Modelsusing two seasonal patterns, intra-day and intra-week, areused in both an ARIMA framework and a Holt–Wintersframework. Using the mean absolute error as a criterion,the doubly seasonal ARIMAandHolt–Wintersmodelsweremore accurate for horizons of less than a week, while amoving average approach was more accurate for longerhorizons. Taylor (2012) develops the doubly seasonalHolt–Winters model further in order to provide densityforecasts for an intra-day call frequency. The forecastingaccuracy of this model was shown to be similar to thatof Shen and Huang (2008b), but the performance of theexponential smoothing model was more consistent acrossthe three series analysed.

In some centres, calls are stimulated by marketingcampaigns. Antipov and Meade (2002) model the re-sponse of call frequencies to such campaigns in a financial

institution whose marketing relied heavily on newspa-per and television advertisements. They develop a modelwith a dynamic level, multiplicative calendar effects (bothwithin-year and within-week), and amultiplicative adver-tising response. The inclusion of the response to advertis-ing is shown to almost halve the mean absolute forecasterror. Soyer and Tarimcilar (2008) pursue a modelling ap-proach similar to that of Avramidis et al. (2004), but includevariables that describe different advertising strategies andpromotion policies. Their modelling approach allows theeffectiveness of different marketing campaigns to be eval-uated.

3.2. ICT time series

Wewill discuss several sets of time series that describeseveral ICT/telecommunications activities. Section 2 wassubdivided according to the modelling approach; in Sec-tion 3.2, it makes more sense to subdivide by the applica-tion area, defined by a set of time series. The sets of ICTtime series considered do not appear to havemuch in com-mon, but do tend to behomogeneouswithin each set. Someexample time series from three data sets of telecommu-nications series are shown in Figs. 2–4; analyses of theseseries will be discussed below. We discuss telecommuni-cation series first, followed by a discussion of the work onother ICT series.

3.2.1. Univariate telecommunications time seriesAn early study of telecommunications data is that of

Grambsch and Stahel (1990); the data set is the numberof circuits in use under a range of ‘Special Services’ offeredby a Bell operating company in the USA. These 261 tele-coms time series are characterised by negative trends, of-


Fig. 2. Telecommunications data: a random selection of five series (Grambsch & Stahel, 1990). Note that the time axis is in months, and the volume axisis the number of circuits in service.

Fig. 3. Telecommunications data: a random selection of five series (Makridakis & Hibon, 2000). Note that the time axis is in months, and the volume axisis unknown.

18/02/2000

3/03/2000

17/03/2000

31/03/2000

14/04/2000

28/04/2000

12/05/2000

26/05/2000

9/06/2000

23/06/2000

7/072000

21/07/2000

4/08/2000

18/08/2000

1/09/2000

15/09/2000

29/09/2000

13/10/2000

27/10/2000

10/11/2000

24/11/2000

8/12/2000

22/12/2000

5/01/2001

19/01/2001

2/02/2001

16/02/2001

2/03/2001

16/03/2001

30/03/200/1

Time

Traf

fic In

dex

80

0

10

20

30

40

50

60

70

Fig. 4. Internet traffic data (Madden & Coble-Neal, 2005).

ten with large steps (see Fig. 2). Their paper proposes therobust trend model (see Meade, 2000) and demonstratesit using their data set. This case is the exception in Sec-tion 3.2, in which the modelling approach is developedspecifically for considering the properties of the time se-ries. The subsequent analyses demonstrate how well therobust trend model achieves its aim. Fildes, Hibon, Makri-dakis, and Meade (1998) used a variety of models on thistelecommunications data set, including ARIMA and expo-nential weighted moving average models; they found nomodel that outperformed the robust trend. Gardner andDiaz-Saiz (2008) also analysed this data set; aiming to chal-lenge the robust trend, they showed that simple exponen-tial smoothing with drift was almost as good.

In the M3 competition, a comparison of the accuracylevels of many different time series modelling approaches,

co-ordinated by Makridakis and Hibon (2000), the data setof 3003 time series included 149 telecoms series. However,on further investigation (suggested by Robert Fildes), wefound 120 of the 149 telecommunications series to be fromthe Grambsch and Stahel data set. The remaining 29 timeseries are identified as monthly. A sample of these series isshown in Fig. 3; no further information is given about theseseries. We see that four of these five plots show downwardtrending series and no obvious seasonality. Madden andTan (2007) examine the two subsets of the M3 telecom-munications data separately. For the monthly data, ARIMAtends to bemore accurate according tomost accuracymea-sures. The 120 time series are noted to be similar to theGrambsch and Stahel data set, but the analysis is unre-liable. For example, the mean absolute percentage errorsquoted (by Madden and Tan) differ both in relative rank-


ings and by several orders of magnitude from those quotedby Fildes et al. (1998), whose results were validated byGardner and Diaz-Saiz (2008).

Madden, Savage, and Coble-Neal (2002) compareeconometric and time series models for forecasting IMTS(international message telephone service) traffic. The dataare six annual time series from 1964 to 1997 describ-ing the outgoing message traffic (in min) from the US toAustralia, Hong Kong, Japan, the Philippines, South Koreaand Taiwan. The other available covariates are price, USGDP, and the relevant incoming message volume. Vectorautoregression (VAR) and error correction models (ECM)are compared with univariate models: ARIMA and a lineartrend model. A combined forecast is also computed as anunweighted average of the ARIMA, VAR and ECM forecasts.The forecasts are evaluated over the period 1994–1997; itis not clear whether the VAR and ECM use covariate val-ues post-1993. Depending on the accuracy measure used,the combined forecast was most accurate for most coun-tries (mean square error), while ECM was most accuratefor three countries (geometric root mean square error).

Two exceptions to the use of conventional modelling ofICT time serieswere found. The first study discussed here isthat of Mastorocostas and Hilas (2012). They consider timeseries of monthly call volumes from their university PBX.An earlier set of these data was considered by Hilas, Gou-dos, and Sahalos (2006), who used conventional methods.They used data from 1988 to 2002 for estimation, and fore-cast the data for 2003; of the eleven methods considered,linear extrapolation with seasonal adjustment was mostaccurate (e.g., amape of 14.0% for the national calls series),followed by a damped trend with multiplicative seasonal-ity (e.g., a mape of 19.5% for the national calls series). Us-ing data from 1998 to 2006 for model estimation and 2007for forecast evaluation, Mastorocostas and Hilas develop afuzzy neural network model for forecasting two monthlycall volume time series. A slightly different set of elevenconventional methods is also used, including two versionsof seasonal adjustmentwith linear extrapolation. The fuzzyneural network out-of-sample forecasts are more accurate(with a mape of 12.1% for the national calls series) thanthe best performing conventional method, damped trendwith multiplicative seasonality (with a mape of 19.1% forthe national calls series). However, the linear extrapola-tion with seasonal adjustment that performed well pre-viously performed comparatively poorly on the later data(mape = 24.4%), suggesting that the results are sensitiveto the period of data chosen.

3.2.2. Internet time series: forecasting usage and provisionMadden and Coble-Neal (2004b, 2005) collected a data

set comprising 59 time series of 232 daily observations ofan (internet) traffic index; the index ranges from 0 (verycongested) to 100 (no traffic). The data are close to whitenoise, with means of around 50–60 and occasional severedownward spikes; an example time series is shown inFig. 4. The authors use the following forecasting methods:ARARMA, ARMA, Holt’s, Holt’s damped trend, robust trend,a random walk and filtered trend. Filtered trend is an ex-trapolation of a regression of the series against time, afterthe removal of outliers. Using themedian RAE as ameasure

of the forecasting accuracy, the filtered trend was most ac-curate across all horizons. In the second exception to theuse of general time series models, Madden and Tan (2008)revisit the traffic data set and apply a feed-forward neuralnetwork; three versions are used, with different parame-ter selection procedures. Simple exponential smoothing isalso included in the analysis of these data for the first time.Unfortunately, due to changes in the choices of the hori-zon and accuracy measure, it is difficult to compare theresults of the 2005 and 2008 papers. However, the neu-ral network methods tended to produce results with accu-racy measures very similar to those of simple exponentialsmoothing for horizons of six days or more. Using medianabsolute percentage errors, simple exponential smoothingwas most accurate for most horizons. Unfortunately, it isnot possible to tell whether thesemethods outperform thefiltered trend. For this traffic data set, it is clear that the dataare stationary and have very little structure, so perhaps itis not surprising that simple exponential smoothing is theleast inaccurate forecasting method.

Ilow (2000) develops a model for forecasting Ethernetpacket data. The objective is to provide a workload modelso as to improve the network performance via dynamicbandwidth allocation. Fractional ARIMAmodelling is used.Although the empirical results are not reported fully, thebenefit of the long memory approach is demonstrated bythe lower mean square error for FARIMA than for the lin-ear AR model. Tzagkarakis, Papadopouli, and Tsakalides(2009) model and forecast internet traffic on wireless lo-cal area networks. They decompose the time series usingSingular Spectrumanalysis. In broad terms, the series is de-composed into signal, a non-linear deterministic trend, andnoise; the trend is extrapolated to provide the forecast.

The provision of broadband is modelled and forecastby Mack and Grubesic (2009), using a range of economet-ric models with data for Ohio, US. The dependent variableis the number of broadband providers in a ZIP code area,where annual data for 2000–2001 are used to forecast from2002–2004. They use the household density, populationgrowth, an urban/non-urban dummy, the number of busi-nesses in the area, the median income, the area in squaremiles, and a (0, 1) dummy variable for if the area is servedby large incumbent suppliers.With the exception of house-hold density, all of these variables were found to be signifi-cant; however, they found that the inclusion of a spatial lagterm improved the forecasting accuracy considerably. Theintuition behind the spatial lag is simply that the demandin neighbouring ZIP code areas will be similar, and thus, sowill the number of providers.

Thompson and Garbacz (2011) investigate the eco-nomic impact of mobile broadband (MB) and fixed broad-band (FB). While both MB and FB have a significant effecton GDP, they find that MB has a greater impact in higherincome countries. In a related study, Srinuan, Srinuan, andBohlin (2012) use survey data to investigate whetherMB isa complementary service or a substitute for FB. They con-clude that, in most areas of Sweden, MB is considered as asubstitute. However, this finding may be influenced by thefact that FB is subject to greater regulation in Sweden, andboth MB and FB are provided by the same companies.


3.2.3. Forecasting for other ICT productsThe medium-term demand for PCs is modelled by Ku-

rawarwala and Matsuo (1998). They compare a lineargrowth model, a Bass model and a seasonal trend modelwith ARIMA models (without and with seasonality). Usingmonthly data for the demand for several PC products, theyfind that the linear trend and seasonal trend models tendto producemore accurate forecasts. For new products withshort time series, they use the demand for analogous prod-ucts to provide initial parameter estimates; they empha-sise the importance of peak sales timing and total life cyclesales in the choice of analogy.

In a relatively early study, a hedonic pricing model fornotebook PCs was constructed by Rutherford andWilhelm(1999). Using regression analysis, they forecast competi-tive selling prices for a given configuration of notebook PC.In addition to finding variables such as CPU speed and harddisk capacity significant, they also found that certain man-ufacturers consistently either charged significant premiaor offered significant discounts.

3.3. Summary: the flow of benefits between ICT and timeseries forecasting methodology

In Section 3.1, we saw that call centre applications, ina similar way to the multi-country diffusion studies, haveproduced several substantial analyses which have beenused successfully for forecasting. The availability of intra-day data has led to the development of ‘doubly seasonal’versions of ARIMA and exponential smoothing models.Considering the studies of univariate time seriesmodellingand forecasting in Section 3.2, the majority are applica-tions of established methodologies. The only example wefound where the application extended the methodology isthe robust forecasting approach of Grambsch and Stahel(1990). The two applications of neural networks for fore-casting show potential with the two or three series theyare used with, but the methods need to be tried on a farlarger set of time series in order to demonstrate that theyare broadly competitive. Thus, the answer to the questionposed at the beginning of Section 3 is that (based on thestudies discussed in Section 2.2, and apart from the ex-ceptions mentioned) there is nothing about telecommuni-cations and other ICT time series that sets them apart. Inother words, the fact that a series is derived from ICT activ-ity is no guide for the choice of a forecasting methodology.

4. Technological forecasting

Technological evolution is an engine of growth that ex-pands market potentials and creates new market leaders(see Sood & Tellis, 2005). Technological forecasting is de-fined by Ascher (1978) as the attempt ‘to project technol-ogy capabilities and to predict the inventions and spreadof technological innovations’. This definition includes dif-fusion modelling, which we consider separately in Sec-tion 2; here, we look at studies using other approaches.Technological forecasting is important, as it is an input tostrategic management decisions regarding where to investfor new technologies, and when to shift investment fromthe old generation of technology to the new. Technological

forecasting also influences the design of new technologies.As was mentioned in the introduction, in the context ofICT, applications of non-diffusion technological forecastinghave received relatively little coverage compared to diffu-sion or time series modelling. Due to this relatively sparsebody of literature, we review the studies by applicationarea.

4.1. Mobile telephony

We discuss two examples of long term forecasts, thefirst focussing on market size and the second focussingon the timing of the introduction of a new generation oftechnology.

Firstly, we look at forecasting the market size. Jeon,Hyun, and Granger (2004) perform a long-term techno-logical forecast of the Chinese mobile telephony market.After a thorough discussion of alternative methods, theychose forecasting by analogy. The mobile telephony mar-kets of the USA, Japan and Korea are identified as possibleprecedents, as they are very large markets and are moredeveloped than that of China. The appropriateness of theanalogy was measured by the time taken for a particulargrowth phase (from 5% to 21%). Using this criterion, Koreawas chosen as themost appropriate precedent, andmarketforecasts for China are computed on this basis.

Secondly, we look at two approaches to forecasting thetiming of the introduction of a new technology. Meade andIslam (2003, 2010) used copulas to model the dependencebetween two relatedproducts or technologies (e.g., faxma-chines and cellular phones) using data from 132 countries.Given that a country has adopted one technology, this de-pendence relationship is then used to predict the adop-tion time of another technology. Kim, Daim, and Anderson(2010) forecast the year of implementation (specifically,the year of ‘first commercialisation’) of a new generationof mobile telephony. They use regression with the intro-duction date as the dependent variable; the independentvariables are four technical variables describing the tech-nology, chosenwith the help of an expert panel. These vari-ables are: channel bandwidth in kHz; number of channels;channel bit rate in kbps; and data capacity in kbps. Start-ing in 1975, they show that the interval between genera-tions is close to 11 years; depending on assumptions aboutdata capacity, their model predicted that 4G would appearbetween 2012 and 2015. The actual ‘first commercialisa-tion’ occurred at their lower limit, with 4G networks beingavailable in South Korea and the USA in 2012. By late 2013,80% of all 4G connections were concentrated in South Ko-rea (where 50% of connectionswere 4G), Japan and theUSA(where, for both countries, 20% of connections were 4G).

4.2. The internet

Similarly, there is little model-based forecasting in thecontext of the internet: instead, forecasts are based on ex-pert opinion. We begin by discussing studies that identifythe drivers of broadband adoption. The change in focus inthese studies from economic variables to applications il-lustrates the rapid rate of change over the twelve years be-tween the first study and the most recent.

Stordahl and Rand (1999) used a Delphi survey to iden-tify and quantify the drivers of the demand for broad-band. They questioned an international panel of experts


over two rounds in 1997, and found that households wereprepared to devote up to 2% of their disposable incometo broadband capacity and applications. Numerical fore-casts of demandwere computed using parameters derivedfrom the Delphi exercise. Using a similar approach, Stor-dahl (2004) produced long-term forecasts for the Europeanbroadband market, broken down by country and the cat-egory of broadband. Weber and Kauffman (2011) discussthe drivers of ICT adoption and identify three internet ap-plication contexts in which research should be pursued:location-based services, cloud computing, and socialnetworking. These application areas are already well es-tablished, but the authors see opportunities for furtherdevelopment.

The second set of studies is more qualitative, looking atpossible developments in internet technology. In a groupof articles on ‘internet prediction’, Elliott (2010) discussesa US National Science Foundation sponsored project, theGlobal Environment for Network Innovations, GENI. Thisongoing project (see www.geni.net) plans to carry out ex-periments to investigatewhat are perceived to be themoreimportant developments in the infrastructure of the inter-net. These developments include content distribution ser-vices (the investigation of more scaleable architectures fordistributing high-bandwidth content such as videos andvirtual worlds in gaming); mobility (the development ofnovel protocols in order to improve the support for mobiledevices in internet architecture); the robustness of inter-net architecture (disruption tolerant networks (DTN),investigating architectures that are independent of the cur-rent TCP/IP architecture); and routing (concerns about theability of the current routing architecture to scale up toan increasing traffic). Holzle and Barroso (2010) draw at-tention to other constraints on the development of inter-net services, and focus particularly on the development ofwarehouse-scale computers that are able to deal with thestorage and transmission of large amounts of data robustly.

Odlyzko (2010) highlights the growth and predomi-nance of mobile voice over the last decade. Looking atmobile broadband, he sees the growth in the demand fordata services outstripping the capacity, as he expects net-work managers to favour voice, which is far more lucra-tive than data transmission. This could lead to a numberof scenarios, such as a growth in applications that exploithigh capacity fixed cable internet, or a growth in servicesthat are designed for lowbandwidthmobile internet. How-ever, the assumptions of this study conflict with current(2013) anecdotal evidence (as discussed in the introduc-tion), where data transmission is tending to bemore lucra-tive than voice, and the development of 4G is facilitating agreater use of mobile broadband.

The routine linking of sensors to the internet could leadto high data flows that could swamp current data storage(Clark, 2010). Estrin (2010) discusses participatory sens-ing: the collection of data via mobile phones for collectionand analysis by the subscriber or others. Applications ofthis include health monitoring or the tracking of environ-mental sustainability. These studies are in linewith the de-velopment of ‘smart phone apps’ (applications) for healthmonitoring.

4.3. Summary: the flow of benefits between ICT and techno-logical forecasting methodology

The brevity of this section reflects the shortage of pub-lications in this area. The majority of the studies are exer-cises based on expert opinion,which tend not to be capableof generalisation into a broader forecasting context. An ex-ception is the study by Jeon et al. (2004), who do provide avery informative case study on the choice and applicationof a long-term technological forecast; they forecast the de-velopment of the Chinese mobile telephony market usingthe Korean market as an analogy.

5. Summary, conclusions and suggestions for furtherresearch

The literature on forecasting applied to an ICT applica-tion goes back at least forty years (e.g., Chaddha & Chit-gopekar, 1971); in those days, the applications related tofixed line telephony andmain-frame computers. The rangeof applications has increasedwith the development of con-sumer technology from PCs and basic mobile phones totablets and ‘smart’ phones. The interactions between ICTapplication areas and forecasting have led to several de-velopments in forecasting. The literature on the multi-generation diffusion model has included a range of ICTapplications; consumer technology products such as mo-bile phones have led to a wider use of choice models; andthe special properties of some ICT series have led to thedevelopment of the robust trend time series forecastingmethod.

Our review shows that themost activity has occurred inthe modelling of diffusion in ICT, particularly mobile tele-phony. The level of innovation in the single-country diffu-sion exercises has generally been low; often, the analysisis simply a comparison of the logistic and Gompertz mod-els using different national data sets. We do note that fit-ting the cumulative diffusion seems towork better than theestimation of diffusion model parameters using adoptionsper period. This experience contrasts with that in the mar-keting literature, where adoptions per period is the depen-dent variable that is used most widely in the estimationof the Bass model; however, this approach tended to beless successful in the context ofmobile telephony. The phe-nomenon of multiple subscriptions per person makes theinterpretation of market potentials problematic. The issueof using models that are designed for capturing adoptionsto model and forecast subscriptions needs further inves-tigation. In the context of modelling multiple generationsof ICT innovations, the use of the simultaneous hazard ap-proach (see Lillard, 1993) for predicting the timing of thenext generation of technology is worth exploring.

The analysis of themulti-country diffusion ofmobiles isgenerally more informative in terms of identifying the rel-ative levels of importance of several determinants of thediffusion rate. In summary, cross-sectional differences inthe diffusion rate can be attributed to the national GNP.Over time, the general consensus is that diffusion accel-erated due to the transition from analog to digital mobilephones and to decreases in prices. Other covariates thathave been cited as accelerating diffusion include the num-ber of competing suppliers and the presence of fewer com-

http://www.geni.net


peting standards. Although these analyses do not focus onforecasting directly, their identification of the importantdeterminants of diffusion facilitates better-informedman-agement decision making.

Modelling the choice between alternatives (possiblecompeting products or suppliers) has generated some newapproaches, for example that of Jun and Park (1999). Fur-ther experience in modelling and forecasting using thisapproach is needed to enable us to assess the strengthsand weaknesses of the approach. Another choice-basedmethodology that has potential for ICT applications is thegeneration of time series forecasts from cross-sectionaldiscrete choice experiments. This approach was suggestedby Islam (2014) and Islam and Meade (2013). They usedrecent developments in discrete choice experimentationto measure household-level preferences, and establisheda causal link between the attributes of the technology andadoption time intentions using discrete time survival mix-ture analysis and the Bass diffusion model.

The forecasting of time series for call centres has stim-ulated several innovations. Examples include the use ofmodels incorporating multiple seasonal patterns and thedevelopment of inhomogeneous Poisson process models.We saw that there are several clusters of ICT data sets thatare broadly homogeneous, and thus, tend to be modelledand forecast better by a single model (or class of models).An early study of one of these clusters of time series ledto an innovation in forecasting methodology, namely therobust trend of Grambsch and Stahel (1990).

Aswe pointed out in the introduction, the supply of aca-demic analyses is affected by factors such as the ease ofaccess of relevant data and the publication criteria of jour-nal editors. This review has covered the literature availableand only the literature available. The distribution of thisbody of literature has influenced both the topics discussedin this review and the emphasis placed upon them. It doesnot necessarily follow that the distribution of academic ef-fort has been optimal for the long term benefit either ofthe ICT industry or of forecasting theory and methodol-ogy. We will try to bear this in mind in our discussionof future research. Looking to the future, it is likely thatdiffusionmodellingwill continue to dominate the ICT fore-casting literature, due to the increasing number of appli-cations as a result of further developments in consumertechnology. The emphasis is likely to move to higher fre-quency (than annual) data as the rate of diffusion increases.These developments are also likely to reinforce the use ofchoice modelling for informing product planning and pro-duction. For future time series ICT applications, the avail-ability of higher frequency data in call centres is likely tolead to a greater emphasis on the real-time updating ofintra-day forecasts. More generally, it is likely that ICT willbenefit from general innovations in forecasting method-ologies, rather than ICT stimulating time series modellinginnovations. The technological developments in ICT are ofwidespread importance and have led to the waxing andwaning of huge corporations. The use of expert opinion islikely to be the main resource; however, the use of anal-ogous data (as per Jeon et al., 2004) for modelling phe-nomena such as technological convergence deservesconsideration, and research outside ICT suggests that itleads to improvements in accuracy (see Önkal, Goodwin,Thomson, Gönül, & Pollock, 2009).

Acknowledgments

We are grateful for the comments on the initial draftsof this paper by three anonymous referees, and to theSocial Science and Humanities Research Council (SSHRC),Canada, grant # 430199, for research support.Wehave alsobenefitted from many discussions with Robert Fildes.

Appendix A. Supplementary data

Supplementary material related to this article can befound online at http://dx.doi.org/10.1016/j.ijforecast.2014.09.003.

References

Agarwal, R., & Bayus, B. L. (2002). Themarket evolution and sales take-offof product innovations. Management Science, 48, 1024–1041.

Andrews, B. H., & Cunningham, S. M. (1995). L.L. Bean improves call-center forecasting. Interfaces, 25(6), 1–13.

Antipov, A., & Meade, N. (2002). Forecasting call frequency at a financialservices call centre. Journal of the Operational Research Society, 53,953–960.

Ascher,W. (1978). Forecasting: an appraisal for policymakers and planners.Baltimore, MD: The Johns Hopkins University Press.

Avramidis, A. N., Deslauriers, A., & L’Ecuyer, P. (2004). Modeling dailyarrivals to a telephone call center.Management Science, 50, 896–908.

Bagchi, K., Kirs, P., & Lopez, F. (2008). The impact of price decreases ontelephone and cell phone diffusion. Information and Management , 45,183–193.

Bartels, R., & Islam, T. (2002). Supply restricted telecommunicationsmarkets: the effect of technical efficiency on waiting times. Journalof Productivity Analysis, 18, 161–169.

Bass, F. M. (1969). A new product growth model for consumer durables.Management Science, 15, 215–227.

Bass, F. M., Krishnan, T., & Jain, D. (1994).Why the Bassmodel fits withoutdecision variables.Marketing Science, 13, 203–223.

Betts, A., Meadows, M., & Walley, P. (2000). Call centre capacitymanagement. International Journal of Service Industry Management ,11(2), 185–196.

Bewley, R., & Fiebig, D. (1988). Flexible logistic growth model with appli-cations in telecommunications. International Journal of Forecasting , 4,177–192.

Bianchi, L., Jarrett, J., & Hanumara, R. C. (1993). Forecasting incoming callsto telemarketing centers. The Journal of Business Forecasting Methodsand Systems, 12, 3–12.

Bianchi, L., Jarrett, J., & Hanumara, R. C. (1998). Improving forecastingfor telemarketing centers by ARIMA modeling with intervention.International Journal of Forecasting , 14, 497–504.

Blackman, W. A. (1972). A mathematical model for trend forecasts.Technological Forecasting and Social Change, 3, 441–452.

Bohlin, A., Gruber, H., & Koutroumpis, P. (2010). Diffusion of newtechnology generations in mobile communications. InformationEconomics and Policy, 22, 51–60.

Botelho, A., & Pinto, P. C. (2004). The diffusion of cellular phones inPortugal. Telecommunications Policy, 28, 427–437.

Brown, L., Gans, N., Mandelbaum, A., Sakov, A., Shen, H., Zeltyn, S., & Zhao,L. (2005). Statistical analysis of a telephone call center: a queuing-science perspective. Journal of the American Statistical Association, 100,36–50.

Chaddha, R. L., & Chitgopekar, S. S. (1971). A ‘generalization’ of the logisticcurve and long range forecasts (1966–1981) of residence telephones.Bell Journal of Economics and Management Science, 2, 542–560.

Chen, C., & Watanabe, C. (2006). Diffusion, substitution and competitiondynamism inside the ICT market: the case of Japan. TechnologicalForecasting and Social Change, 73, 731–759.

Chinn, M. D., & Fairlie, R. W. (2007). The determinants of the GlobalDigital Divide: a cross-country analysis of computer and Internetpenetration. Oxford Economic Papers, 59, 16–44.

Chu, W.-L., Wu, F.-S., Kao, K.-S., & Yen, D. C. (2009). Diffusion of mobiletelephony: an empirical study in Taiwan. Telecommunications Policy,33, 506–520.

Clark, D. (2010). Fighting over the future of the Internet. IEEE InternetComputing , 10, 22–23.

Comer, J. C., & Wikle, T. A. (2008). Worldwide diffusion of the cellulartelephone, 1995–2005. The Professional Geographer , 60, 252–269.












http://refhub.elsevier.com/S0169-2070(14)00179-4/sbref1
























De Gooijer, J. G., & Hyndman, R. J. (2006). 25 years of time seriesforecasting. International Journal of Forecasting , 22, 443–473.

Dekimpe, M. G., Parker, P. M., & Sarvary, M. (1998). Staged estimationto global cellular telephone adoption. Technological Forecasting andSocial Change, 57, 105–132.

Dergiades, T., & Dasilas, A. (2010). Modelling and forecasting mobiletelecommunication services: the case of Greece. Applied EconomicsLetters, 17, 1823–1828.

Doganoglu, T., & Grzybowski, L. (2007). Estimating network effects inmobile telephony in Germany. Information Economics and Policy, 19,65–79.

Duffy-Deno, K. T. (2001). Demand for additional telephone lines: anempirical note. Information Economics and Policy, 13, 283–299.

Eisner, J., & Waldon, T. (2001). The demand for bandwidth: secondtelephone lines and on-line services. Information Economics andPolicy, 13, 301–309.

Elliott, J. (2010). GENI: opening up new classes of experiments in globalnetworking. IEEE Internet Computing , 10, 39–42.

Estrin, D. (2010). Participatory sensing: applications and architecture.IEEE Internet Computing , 10, 12–14.

Fildes, R. (2002). Telecommunications demand forecasting—a review.International Journal of Forecasting , 18, 489–522.

Fildes, R. (2003). Corrigendum to ‘‘Telecommunications demandforecasting—a review’’ [International Journal of Forecasting 18(2002) 489–522]. International Journal of Forecasting , 19, 337.

Fildes, R., Hibon, M., Makridakis, S., & Meade, N. (1998). Generalisingabout univariate forecasting methods: further empirical evidence.International Journal of Forecasting , 14, 339–358.

Fisher, J. C., & Pry, R. H. (1971). Simple substitutionmodel of technologicalchange. Technological Forecasting and Social Change, 3, 75–88.

Frank, L. D. (2004). An analysis of the effect of the economic situation onmodeling and forecasting the diffusion of wireless communicationsin Finland. Technological Forecasting and Social Change, 71, 391–403.

Gamboa, L. F., & Otero, J. (2009). An estimation of the pattern of diffusionof mobile phones: the case of Colombia. Telecommunications Policy,33, 611–620.

Gans, N., Koole, G., & Mandelbaum, A. (2003). Telephone call centers:tutorial, review, and research prospects. Manufacturing and ServiceOperations Management , 5, 79–141.

Gardner, E. S., Jr., & Diaz-Saiz, J. (2008). Exponential smoothing in thetelecommunications data. International Journal of Forecasting , 24,170–174.

Goolsbee, A., & Klenow, P. J. (2002). Evidence on learning and networkexternalities in the diffusion of home computers. Journal of Law andEconomics, 45, 317–344.

Grajek, M., & Kretschmer, T. (2009). Usage and diffusion of cellulartelephony, 1998–2004. International Journal of Industrial Organization,27, 238–249.

Grambsch, P., & Stahel, W. A. (1990). Forecasting demand for specialtelephone services—a case study. International Journal of Forecasting ,6, 53–64.

Gruber, H., & Koutroumpis, P. (2013). Competition enhancing regulationand diffusion of innovation: the case of broadband networks. Journalof Regulatory Economics, 43, 168–195.

Gruber, H., & Verboven, F. (2001). The diffusion of mobile telecommuni-cation services in the European Union. European Economic Review, 45,577–588.

Gurbaxani, V., & Mendelson, H. (1990). An integrative model of ISspending growth. Information Systems Research, 1, 23–46.

Hamoudia, M., & Islam, T. (2004). Modelling and forecasting the growthof wireless messaging. Telektronikk, 100, 64–69.

Hilas, C. S., Goudos, S. K., & Sahalos, J. N. (2006). Seasonal decompositionand forecasting of telecommunication data: a comparative casestudy. Technological Forecasting and Social Change, 73, 495–509.

Holzle, U., & Barroso, L. A. (2010). Warehouse-scale computers. IEEEInternet Computing , 10, 33–35.

Hwang, J., Cho, Y., & Long, N. V. (2009). Investigation of factors affectingthe diffusion of mobile telephone services: an empirical analysis forVietnam. Telecommunications Policy, 33, 534–543.

Ida, T., Kinoshita, S., & Sato, M. (2008). Conjoint analysis of demand for IPtelephony: the case of Japan. Applied Economics, 40, 1279–1287.

Ida, T., & Kuroda, T. (2006). Discrete choice analysis of demand forbroadband in Japan. Journal of Regulatory Economics, 29, 5–22.

Ilow, J. (2000). Forecasting network traffic using FARIMA models withheavy tailed innovations. In Proceedings of the acoustics, speech, andsignal processing IEEE international conference. Vol. 6 (pp. 3814–3817).

Islam, T. (2014). Household level innovation diffusion model of photo-voltaic (PV) solar cells from stated preference data. Energy Policy, 65,340–350.

Islam, T., & Fiebig, D. (2001). Modelling the development of supplyrestricted telecommunications markets. Journal of Forecasting , 20,249–264.

Islam, T., Fiebig, D. G., & Meade, N. (2002). Modelling multinationaltelecommunications demand with limited data. International Journalof Forecasting , 18, 605–624.

Islam, T., & Meade, N. (1996). Forecasting the development of the marketfor business telephones in the UK. Journal of Operational ResearchSociety, 47, 906–918.

Islam, T., & Meade, N. (1997). The diffusion of successive generations ofa technology: a more general model. Technological Forecasting andSocial Change, 56, 49–60.

Islam, T., & Meade, N. (2011). Detecting the impact of market factors onsales take-off times of analog cellular telephones. Marketing Letters,22, 197–212.

Islam, T., & Meade, N. (2012). The impact of competition, and economicglobalization on the multinational diffusion of 3G mobile phones.Technological Forecasting and Social Change, 79, 843–850.

Islam, T., & Meade, N. (2013). The impact of attribute preferenceson adoption timing: the case of photo-voltaic (PV) solar cells forhousehold electricity generation. Energy Policy, 55, 521–530.

Jang, S. L., Dai, S.-C., & Sung, S. (2005). The pattern and externality effectof diffusion of mobile telecommunications: the case of OECD andTaiwan. Information Economics and Policy, 17, 133–148.

Jeffres, L., & Atkin, D. (1996). Predicting use of technologies forcommunication and consumer needs. Journal of Broadcasting andElectronic Media, 40, 318–330.

Jeon, Y., Hyun, K. R., & Granger, C. J. (2004). Long-term technologicalforecasting. Telektronikk, 100, 3–12.

Jeon, H. J., Kim, M. S., & Sohn, S. Y. (2010). Conjoint and WTP analysesof future mobile phones for digital convergence. TechnologicalForecasting and Social Change, 77, 457–465.

Jun, D. B., Kim, S. K., Park, M. H., Bae, M. S., Park, Y. S., & Joo, Y. J. (2000).Forecasting demand for low earth orbit mobile satellite service inKorea. Telecommunications Systems, 14, 311–319.

Jun, D. B., Kim, S. K., Park, Y. S., Park, M. H., & Wilson, A. R. (2002).Forecasting telecommunication service subscribers in substitutiveand competitive environments. International Journal of Forecasting ,18, 561–581.

Jun, D. B., & Park, Y. S. (1999). A choice-based diffusionmodel formultiplegenerations of products. Technological Forecasting and Social Change,61, 45–58.

Kiiski, S., & Pohjola, M. (2002). Cross-country diffusion of the Internet.Information Economics and Policy, 14, 287–310.

Kim, J., Daim, T., & Anderson, T. (2010). A look into the future ofwireless mobile communication technologies. Technology Analysisand Strategic Management , 22, 925–943.

Klungle, R., & Maluchnik, J. (1997). Call center forecasting at AAAMichigan. The Journal of Business Forecasting Methods and Systems,16(4), 8–13.

Koole, G. (2008). Introduction to the special issue on call centermanagement.Management Science, 54, 237.

Koski, H., & Kretschmer, T. (2005). Entry, standards and competition: firmstrategies and the diffusion of mobile telephony. Review of IndustrialOrganization, 26, 89–113.

Kreng, V. B., & Wang, H. T. (2009). A technology replacement model withvariable market potential—an empirical study of CRT and LCD TV.Technological Forecasting and Social Change, 76, 942–951.

Kumar, V., Nagpal, A., & Venkatesan, R. (2002). Forecasting category salesand market share for wireless telephone subscribers: a combinedapproach. International Journal of Forecasting , 18, 583–603.

Kurawarwala, A. A., & Matsuo, H. (1998). Product growth models formedium-term forecasting of short life cycle products. TechnologicalForecasting and Social Change, 57, 169–196.

Kwak, J. H., & Lee, B. G. (2011). Estimating demand curve in the KoreanVoIP telecommunicationsmarket. Technological Forecasting and SocialChange, 78, 713–728.

Lee, M., & Cho, Y. (2007). The diffusion of mobile telecommunicationsservices in Korea. Applied Economics Letters, 14, 477–481.

Lee, J., Cho, Y., Lee, J.-D., & Lee, C.-Y. (2006). Forecasting future demandfor large-screen television sets using conjoint analysis with diffusionmodel. Technological Forecasting and Social Change, 73, 362–376.

Lee, C.-Y., Lee, J.-D., & Kim, Y. (2008). Demand forecasting for newtechnology with a short history in a competitive environment; thecase of the home networking market in South Korea. TechnologicalForecasting and Social Change, 75, 91–106.

Lee, J. C., & Lu, K. W. (1987). On a family of data based transformedmodels useful in forecasting technological substitution. TechnologicalForecasting and Social Change, 31, 61–78.

Lee, J. C., Lu, K. W., & Horng, S. C. (1992). Technological forecasting withnon-linear models. Journal of Forecasting , 11, 195–206.

Liikanen, J., Stoneman, P., & Toivanen, O. (2004). Intergenerational effectsin the diffusion of new technology: the case of mobile phones.International Journal of Industrial Organization, 22, 1137–1154.




























































Lillard, L. A. (1993). Simultaneous equations for hazards. Journal ofEconometrics, 56, 189–217.

Lim, B.-L., Choi, M., & Park, M.-C. (2003). The late take-off phenomenonin the diffusion of telecommunications services: network effect andcritical mass. Information Economics and Policy, 15, 537–667.

Mabert, V. A. (1985). Short interval forecasting of emergency phone call(911) work loads. Journal of Operations Management , 5, 259–271.

Mack, E. A., & Grubesic, T. H. (2009). Forecasting broadband provision.Information Economics and Policy, 21, 297–311.

Madden, G., & Coble-Neal, G. (2004a). Economic determinants of globalmobile telephony growth. Information Economics and Policy, 16,519–534.

Madden, G., & Coble-Neal, G. (2004b). Internet traffic dynamics.Telektronikk, 4, 168–179.

Madden, G., & Coble-Neal, G. (2005). Forecasting international bandwidthcapability. Journal of Forecasting , 24, 299–309.

Madden, G., Savage, S. J., & Coble-Neal, G. (2002). Forecasting UnitedStates international message telephone service. International Journalof Forecasting , 18, 523–544.

Madden, G., & Tan, J. (2007). Forecasting telecommunications data withlinear models. Telecommunications Policy, 31, 31–44.

Madden, G., & Tan, J. (2008). Forecasting international bandwidthcapacity using linear and ANN methods. Applied Economics, 40,1775–1787.

Mahajan, V., & Muller, E. (1996). Timing, diffusion, and substitutionof successive generations of technological innovations: the IBMmainframe case. Technological Forecasting and Social Change, 51,109–132.

Makridakis, S., & Hibon, M. (2000). The M3-competition: results,conclusions and implications. International Journal of Forecasting , 16,451–476.

Mandelbaum, A. (2003). Call centres: research bibliography with abstracts.Technical Report. http://iew3.technion.ac.il/serveng/References/ccbib.pdf.

Mar-Molinero, C. (1980). Tractors in Spain—a logistic analysis. Journal ofthe Operational Research Society, 31, 141–152.

Massini, S. (2004). The diffusion of mobile telephony in Italy and theUK: an empirical investigation. Economics of Innovation and NewTechnology, 13, 251–277.

Mastorocostas, P., & Hilas, C. (2012). A computational intelligence-basedforecasting system for telecommunications time series. EngineeringApplications of Artificial Intelligence, 25, 200–206.

Meade, N. (2000). A note on the robust trend andARARMAmethodologiesused in the M3 competition. International Journal of Forecasting , 16,517–519.

Meade, N., & Islam, T. (1995). Prediction intervals for growth curveforecasts. Journal of Forecasting , 14, 413–430.

Meade, N., & Islam, T. (2001). Forecasting the diffusion of innovations:implications for time series extrapolation. In J. S. Armstrong (Ed.),Principles of forecasting: a handbook for researchers and practitioners(pp. 577–595). Norwell, MA: Kluwer Academic Publishers.

Meade, N., & Islam, T. (2003). Modeling the dependence betweenthe times to international adoption of two related technologies.Technological Forecasting and Social Change, 70, 759–778.

Meade, N., & Islam, T. (2006). Modelling and forecasting the diffusion ofinnovation—a 25-year review. International Journal of Forecasting , 22,519–545.

Meade, N., & Islam, T. (2010). Using copulas to model repeat purchasebehaviour—an exploratory analysis via a case study. European Journalof Operational Research, 200(3), 908–917.

Michalakelis, C., Varoutas, D., & Sphicopoulos, T. (2008). Diffusionmodels of mobile telephony in Greece. Telecommunications Policy, 32,234–245.

Nam, C., Kim, S., & Lee, H. (2008). The role of WiBro: filling the gaps inmobile broadband technologies. Technological Forecasting and SocialChange, 75, 438–448.

Norton, J. A., & Bass, F.M. (1987). A diffusion theorymodel of adoption andsubstitution for successive generations of high-technology products.Management Science, 33, 1069–1086.

Odlyzko, A. (2010). The Internet and past and future communicationsrevolutions. IEEE Internet Computing , 10, 20–21.

OECD (2011). OECD science, technology and industry scoreboard 2011.OECD Publishing.

Önkal, D., Goodwin, P., Thomson, M., Gönül, M. S., & Pollock, A. (2009).The relative influence of advice from human experts and statisticalmethods on forecast adjustments. Journal of Behavioral DecisionMaking , 22, 390–409.

Rai, A., Ravichandran, T., & Samaddar, S. (1998). How to anticipate theInternet’s global diffusion. Communications of the ACM , 41, 97–106.

Robertson, A., Soopramaniem, D., & Fildes, R. (2007). Segmental new-product diffusion of residential broadband services. Telecommunica-tions Policy, 31, 265–275.

Rogers, E. M. (2003).Diffusion of innovations (5th ed.). New York: The FreePress.

Rouvinen, P. (2006). Diffusion of digitalmobile telephony: are developingcountries different? Telecommunications Policy, 30, 46–63.

Rutherford, D. P., & Wilhelm, W. E. (1999). Forecasting notebookcomputer price as a function of constituent features. Computers andIndustrial Engineering , 37, 823–845.

Sharif, M. N., & Kabir, C. (1976). System dynamics modeling for fore-casting multilevel technological substitution. Technological Forecast-ing and Social Change, 9, 89–112.

Shen, H., & Huang, J. Z. (2005). Analysis of call centre arrival data usingsingular value decomposition. Applied Stochastic Models in Businessand Industry, 21, 251–263.

Shen, H., & Huang, J. Z. (2008a). Interday forecasting and intradayupdating of call center arrivals.Manufacturing and Service OperationsManagement , 10, 391–410.

Shen, H., &Huang, J. Z. (2008b). Forecasting time series of inhomogeneousPoisson processes with application to call center workforce manage-ment. The Annals of Applied Statistics, 2, 601–623.

Shin, D. H., & Bartolacci, M. (2007). A study ofMVNOdiffusion andmarketstructure in the EU, US, Hong Kong and Singapore. Telematics andInformatics, 24, 86–100.

Singh, S. K. (2008). The diffusion of mobile phone in India. Telecommuni-cations Policy, 32, 642–651.

Sood, A., & Tellis, G. J. (2005). Technological evolution and radicalinnovation. Journal of Marketing , 69, 152–168.

Soyer, R., & Tarimcilar, M. (2008). Modeling and analysis of call centerarrival data: a Bayesian approach.Management Science, 54, 266–278.

Sparrow, L. B. (1991). Manning the telephone enquiry bureau at BritishGas West Midland. In S. C. Littlechild (Ed.), Operations research inmanagement (pp. 167–173). New York: Prentice Hall.

Srinuan, P., Srinuan, C., & Bohlin, E. (2012). Fixed and mobile broadbandsubstitution in Sweden. Telecommunications Policy, 36, 237–251.

Stordahl, K. (2004). Long-term broadband technology forecasting.Telektronikk, 100, 13–31.

Stordahl, K., & Rand, L. (1999). Long term forecasts for broadbanddemand. Telektronikk, 95, 34–44.

Sundqvist, S., Frank, L., & Puumalainen, K. (2005). The effects of countrycharacteristics, cultural similarity and adoption timing on thediffusion ofwireless communications. Journal of Business Research, 58,107–110.

Sung, N., & Lee, Y.-H. (2002). Substitution between mobile and fixedtelephones in Korea. Review of Industrial Organization, 20, 367–374.

Tandberg, D. (1995). Time series forecasts of poison center call volume.Clinical Toxicology, 33, 11–18.

Taylor, J. W. (2008). A comparison of univariate time series methods forforecasting intraday arrivals at a call center.Management Science, 54,253–265.

Taylor, J. W. (2012). Density forecasting of intraday call center arrivalsusing models based on exponential smoothing. Management Science,58, 543–549.

Tellis, J. T., Stremersch, S., & Yin, E. (2003). The international takeoffof new products: the role of economics, culture, and countryinnovativeness.Marketing Science, 22, 188–208.

Thompson, H. G., Jr., & Garbacz, C. (2011). Economic impacts of mobileversus fixed broadband. Telecommunications Policy, 35, 999–1009.

Tych, W., Pedregal, D. J., Young, P. C., & Davies, J. (2002). An unobservedcomponent model for multi-rate forecasting of telephone calldemand: the design of a forecasting support system. InternationalJournal of Forecasting , 18, 673–695.

Tzagkarakis, G., Papadopouli, M., & Tsakalides, P. (2009). Trend forecast-ing based on singular spectrum analysis of traffic workload in a large-scale wireless LAN. Performance Evaluation, 66, 173–190.

Weber, M., & Kauffman, R. J. (2011). What drives ICT adoption?Analysis and research directions. Electronic Commerce Research andApplications, 10, 683–701.

Weinberg, J., Brown, L. D., & Stroud, J. R. (2007). Bayesian forecasting of aninhomogeneous Poisson processwith applications to call center data.Journal of the American Statistical Association, 102(480), 1185–1198.

Wiorkowski, J. J., & Gylys, V. A. (2006). An empirical real-time test fortakeoff with applications to cellular telephony. Review of MarketingScience, 4, 1–41.

Wu, F.-S., & Chu, W.-L. (2010). Diffusion models of mobile telephony.Journal of Business Research, 63, 497–501.

Xu, W. (1999). Long range planning for call centers at FedEx. The Journalof Business Forecasting Methods and Systems, 18(4), 7–11.

Yang, Y., &Williams, E. (2009). Logistic model-based forecast of sales andgeneration of obsolete computers in the US. Technological Forecastingand Social Change, 76, 1105–1114.













http://iew3.technion.ac.il/serveng/References/ccbib.pdf

























































Nigel Meade is Professor of Management Science, Imperial CollegeBusiness School, London, UK. He is an experienced statistical modelerwith a background in operational research and statistics applied toinnovation diffusion and finance. He is an associate editor of theInternational Journal of Forecasting and the European Journal of Finance. Hehas published over 50 papers and eight book chapters, and successfullysupervised more than 20 Ph.D. students.

Towhidul Islam is Professor of Marketing and Consumer Studies, Collegeof Management and Economics (CME), and CME Fellow in consumerinsights, consumerwell-being and public policy, University of Guelph. HisPh.D. is in management science from Imperial College Business School,London, UK. His main research interests are in the areas of innovationdiffusion, consumer choice and choice models. His work has appeared inthe Journal of Consumer Research,Management Science, Journal of ConsumerPsychology, International Journal of Research in Marketing, InternationalJournal of Forecasting, Journal of Forecasting, and the European Journal ofOperational Research, among other outlets.

1-s2.0-S0169207014001794-main

Documents

ict appli

importance of ict

singlecountry analyses

communications technology

example analyses

ict area of application

imperial college london

canadaa r t i c