Telektronikk 3/4.2008

Keynote Address

The paper describes the broadband evolution especially in Western Europe. The broadband penetration is separated into the business and residential markets and forecasting models are developed for both markets and also for the main broadband technologies DSL, Cable modem (Hybrid Fibre Coax), FTTx and Fixed Wireless Access. The FTTx deployment and demand in Western Europe are to a certain degree suppressed by uncertain regulatory framework for fibre accesses. The influence of mobile broadband is discussed. Finally, analyses show that the cost of Fixed Wireless Access is high, indicating that the technology only will be competitive in areas which cannot be covered by DSL and cable modem.

Although not the leader in all telecommunications areas, the US is still a large diverse market of considerable economic interest. This article provides an overview of TFI’s latest quantitative forecasts of the US telecommunications network in terms of competition, broadband data rates, Internet video, HDTV, fiber in the loop deployment, wireless broadband, and VoIP. Between now and 2016, the US is forecast to convert most of its telecommunications infrastructure to all-fiber and 4G wireless transmission and all IP-switching. However, different operators will follow different paths to get there and new players are emerging as the telephony, cable television, and Internet paradigms converge.

The focus of this paper is to outline a process that first predicts where and when a company will expand its fiber network and then forecasts the impact(s) of that expansion. Specifically, we look at Verizon’s current and future deployment of their fiber network. Forecasting a competitor’s planned investment activities is important. The ability to adequately predict where and when Verizon will add to their network provides lead time for the cable companies to plan their own triple-play deployments; to develop customer loyalty programs; and to identify customers at risk for targeted advertising and promotions. We specify a spatial model of location choice and then test this model by comparing the model’s prediction of deployment against realized deployment in California. The results suggest that the model captures key drivers of location choice.

The UK telecommunications network consists of over 5,500 exchanges that are used to deliver broadband and communication services to the majority of households. This infrastructure is rapidly becoming inadequate for supplying the data transfer needs of customers. This is partly due to population growth and migration, but mainly due to demand for bandwidth hungry internet services, for example high definition television and video conferencing. One solution that is currently being considered is to upgrade the network to use fibre-optic cables. Fibre optic cables have higher data transmission rates than the copper cables currently used and have lower attenuation. One way to minimise the costs of deploying a fibre optic network is to reduce the number of exchanges used, saving the money needed to maintain those exchanges. We attempt to determine the most profitable way to upgrade the network: optimising replacement of fibre optic cables and de-commissioning of exchanges. Since this is a combinatorial problem we use simulated annealing to find the optimal upgraded network.

In this brief paper we explore the relationship between price, wealth and broadband adoption to determine the importance of price as a driver of demand.

Convergence in the Internet, telecommunications and broadcasting industry/market is becoming a reality today. Convergence enables the provisioning of services independent of the underlying networks. For instance, end-users will be able to access broadcasting services such as television on their mobile phones or personal computers (PCs). In other words, convergence offers an end user the possibility to access services anytime and anywhere irrespective of the access network or terminal used. For a network operator convergence may provide an opportunity to have better network management, reduce the operational expenditure, expand their market reach and retain their customers. However, this ultimate goal of seamless and ubiquitous service access would require major changes in the existing network architecture, market, services and terminals. Understanding these changes is crucial for telecom and broadcasting industries in order to make a successful migration and offer higher quality services to end-users.

The paper presents a comprehensive rollout analysis of HSPA in a medium large European country with an operator who already has a GSM network. The rollout coverage will be higher in the densest areas. It is also assumed that the traffic per user is higher in the densest areas. Forecasts for mobile broadband accesses and the traffic per user in the peak hour is an important part of the analysis. The traffic forecasts are used to make a strategy for upgrading the capacity of the base stations based on throughput calculations for upgrading with additional sectors, additional carriers and HSPA 3.6, 7.2 and 14.4 Mbps respectively. Techno-economic calculations are performed for the rollout cases based on ARPU predictions, cost evolution of the network elements, evolution of OPEX and assumption of future market share and competition. The long-term access and traffic forecasts are uncertain. To examine the impact, techno-economic calculations are performed for increasing forecasts.

In a country with a supply restricted fixed line telephone network, we might expect a rapid diffusion of cellular telephony as an alternative means of communication becomes available. However, in developing countries, long take-off times are observed giving a left hand skew to the diffusion curve. The delay in take-off can be attributed to network externalities, such as difficulties in establishing links between cellular and fixed line networks. Another explanation is that the typical diffusion curve of an interactive technology is left skewed compared to a non-interactive technology, because a critical mass of users is needed before take-off can occur. To investigate the existence and causes of left skew, we use data from 70 countries covering all geographical regions and economic categories with a comprehensive set of economic, cultural and telecommunications variables to measure the evidence for left-skew. We use Bemmaor’s (1994) gamma shifted Gompertz skew parameter to investigate network externality. The power of economic and cultural covariates to explain the variation within skew measures is explored and the implications of left-skew on forecasting accuracy are discussed.

We consider the problems of explaining and forecasting the penetration and the traffic in cellular mobile networks. To this end, we create two regression models, viz. one to predict the penetration from service charges and network effects and another one to predict the traffic from service charges and diffusion and adoption effects. The results of the models can also be combined to compute the likely evolutions of essential characteristics such as Minutes of Use (MoU), Average Revenue per User (ARPU) and total revenue. Applying the models to 28 markets throughout the world we show that they perform very well. Noting the significant qualitative differences between these markets, we conclude that the model has some universality in that the results are comparable for all of them.

The ICT markets underwent a profound metamorphosis and continue to be under a high pressure. The main factors underlying this change are (a) the deregulation and liberalization of the European telecommunications markets, (b) the explosion of new services and technology, (c) the increasing and aggressive competition between incumbent Telcos, new operators and also the IT integrators, and (d) the convergence of IT and telecommunications. These factors lead to increasing complexity and difficulties in modelling and forecasting products and services on business markets. These new constraints and customers’ and players’ behaviours were not adequately integrated in the ‘traditional’ modelling (ie. multiple regression, time series with transfer function, diffusion models, …) and led to inconsistent and inaccurate forecasts, with corresponding shifts in supply and demand. Most attempts to model the Telecoms and ICT markets have concentrated on the demand side. Only a few modelling studies have been carried out on both demand and supply. Of these, the majority were at a highly aggregated level. In this paper we present: (a) a global overview of the ICT markets, (b) a description of the forecasting methodology we suggest, based on a system of simultaneous multiple regression equations, establishing multiple linkages between, on the one hand, the demand for a variety of ICT products and services and, on the other, the supply and the economic, regulatory and technology environment.

This paper proposes, describes and illustrates an alternative approach to estimate individual level models for information and communications technologies (ICT) and compares with existing established models such as mixed logit random coefficient and Hierarchical Bayes random coefficient model. In order to estimate individual level models, we have used two recent developments: (a) availability of efficient experimental designs and (b) collection of extra preference information using repeated best-worst questions. Individual level model out performed more complex models both in sample and out of sample model fit.

This paper has originally been published as: Markowitz, H. Portfolio Selection. Journal of Finance, 7 (1), 77-91, 1952. © John Wiley & Sons, Inc. Reprinted with permission from Blackwell Publishing Ltd.

The first part of the paper gives an overview of risk analysis. Important elements of the risk picture are market risk, competition risk, regulatory risk, technology risk and operational risk. The second part presents a framework for uncertainty handling and risk assessment in business case analysis. Various measures of uncertainty and risk such as volatility, value at risk (VaR), and Sharpe ratio are presented. The Markowitz portfolio theory is presented and an extension to this framework including Monte Carlo simulation together with optimization is described and illustrated with examples. Important differences between the portfolio analysis of financial securities and that of real assets or investment projects are highlighted and recommendations are given.

The growing complexity of markets calls for different approaches when searching for knowledge of market dynamics. Conjoint analysis, also called trade-off analysis is such an approach, attempting to mimic respondents’ purchase decision through ranking or choosing between levels of different attributes. The methodology allows the analyst to segment the market and feed the results into a market simulator. This simulator may be used to understand the importance of different product attributes, simulate demand elasticities, optimize product portfolios, estimate demand for a given product, make forecasts of the market dynamics, measure brand strength and measure the trade-off between attributes. We give an introduction to the ideas behind Conjoint analysis and the various steps you need to go through to conduct a Conjoint analysis. A brief overview of the main methodologies of Conjoint, ie. Conjoint Value Analysis, Adaptive Conjoint and Choice Based Conjoint is given. We further give some examples of how Telenor uses Conjoint when modifying existing products, extending the product line or introducing new products.

Appropriate forecasting of product market adoption enables optimal planning of resources, investments, revenue, marketing and sales. Quantitative forecasting methods for the new product adoption rely on the S-shaped (sigmoidal) growth model such as the logistic, the Richards and Bass growth models. This paper presents adaptations of these models by introducing explanatory and marketing variables which are suitable for the forecasting prior to product launch or in the early phases of the product life cycle.

M-competition results are often cited as evidence that complex models do not always produce more accurate forecasts than simple models. Although the results have provided forecasters with informative insights, the results of the competition are generated with an aggregate model for all series instead of an individually selected model for each series. Following Shah (1997), this study introduces an individual model selection procedure for forecasting by employing a multinomial logit (MNL) model to relate best forecast method to data moments (mean, variance, skewness and kurtosis) and selected time-series characteristics (coefficient of variation, number of outliers, step changes, turning points, trend direction, number of observations and ARCH effects). The MNL procedure is trialled on the M3 competition data. Encouragingly, the MNL model based on the Relative GRMSE in particular, is able to indicate the better forecast model reasonably well. Not surprising, results differ by error statistic and higher frequency data are more difficult to forecast. The study is exploratory in nature and another set of data characteristics may be more appropriate for a different series.

The aim of our research was to create a functional algorithm of preprocessing of input data taking into account the specific aspects of teletraffic and properties of neural networks. The practical application to forecasting telecommunication data sequences shows that the procedure of data preprocessing decreases the time of learning (what is particularly important in the case of large data sets) and increases the plausibility and accuracy of the forecasts. The algorithm can be applied to forecasting the intensity of plain telephone networks and IP networks.

The pattern by which new technology is adopted is reasonably well-understood and, assuming there is data, there are mathematical models and methods to help forecast. However, many of the most strategic forecasts involve not much data and lots of uncertainty. There are ‘big’ methods – alternate scenarios, for example – to address such issues, but sometimes the practitioner needs to make a good forecast quickly and with few resources. In the process, the same issues often come up. Some examples: When will a new technology be introduced? Will anyone adopt it? If so, how many and how quickly? What are the factors of success? How long before the technology is obsolete? Which of two or more technologies will win? Each of these questions gives rise to other questions, the answers to which enable a good forecast. Or, in some cases, the answers lead to the conclusion that a definitive forecast would be premature. This paper provides examples, philosophy, and practical advice for addressing these questions so that quick – and dirty – forecasts are long-lasting and beautiful.

Terms and Acronyms in Telecommunications Forecasting