Expert Dialogues Mohsen HAMOUDIA mohsen.hamoudia@francetelecom.com Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Examining Forecasting Methodologies To Assist And Support Operators In The Transition From A Regulated Environment To Deregulated Markets ITU - International Telecommunication Union Market, Economics and Finance Unit in cooperation with IIF - International Institute of Forecasters Geneva, Switzerland 25-26 October 2004 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Table of Contents n Introduction n A Changing Environment : From a Monopoly Situation to Liberalized Markets n Limitations of “Traditional” Forecasting Methods and Models n Case Study from an Important Operator experience in International Traffic : outlining the forecasting methods undertaken and how they held up to the challenges of Telecom Deregulation in 1998 n Conclusion 2 A Changing Environment From a Monopoly Situation to Liberalized Markets Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (1/10) In Europe, 1998 was a very important event because the European Telecom Industry has been entirely deregulated. The main effects and consequences were : l Increasing number of operators, carriers and Internet providers in the deregulated markets l Increasing (and aggressive) competition and new Alliances l Explosion of new products and services l Evolution of the Customer Choices and usages : From Telecommunication to Communication l Falling Prices and Falling Revenue l Pressure on Settlement Rates on Bilateral Routes 4 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (2/10) n The Telecom Deregulation level From 1 – Regulated markets to 5 – Fully Liberalized markets 5 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (3/10) n Increasing and aggressive competition Int'l Traffic by Carrier Type Outgoing Int'l in Minutes (Billions) 140 120 100 New Carriers in Competitive Markets Existing Carriers in Competitive Markets Monopolies 80 60 40 20 0 1990 1992 1994 1996 6 1998 1999 2000 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (4/10) n The International Carrier Boom Number of Carriers 2805 3000 2500 2000 1760 1500 1000 586 500 367 0 1995 1997 1999 7 2000 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (5/10) n A general overview of Worldwide Telecom Industry Fixed Telephony Data Services Mobile Internet 1460 bEuros 1162 bEuros 780 bEuros 67 145 83 161 123 62 573 529 203 448 1998 427 +10% / year 581 + 5% / year 2002 2006 2005 1,2 billion of fixed lines 1,5 billion of mobiles 900 millions of Internet Access Source : Idate 8 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (6/10) n Explosion of new products and services n n n P2P Communications : + 40 % Fixed Voice Calls : only + 18 % Fixed Voice Share in the Global Call volumes falled Monthly Number of communication by FT’s Customer Structure of communications Mail 1,8 mobile voice SMS 5,2 fixed voice Text 1,0 mobile voice fixed voice 19,6 32,6 97% 49% 4% Internet SMS 11% mobile voice 42% voice Voice + 18% fixed voice 20,2 1997 3% fixed voice 51% 43% 2002 1997 Source : FT/DPS 2003 9 2002 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (7/10) n Evolution of the Customer usages and Choices France : Monthly Number of communication by Customer 200 Customer wants Communication rather Telecommunication 160 140 Internet 120 SMS and mobile data 100 60 Communication synchrone 80 Mobile voice 40 20 Fixed voice 0 1990 Fixed voice 1992 Mobile voice 1994 1996 SMS and mobile data 1998 2000 Minitel 2002 2004 Internet mail & IM Communication asynchrone 180 2006 2008 2010 Internet WEB &t P-to-P Source : FT/DPS 2003 10 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (8/10) n Evolution of the Customer usages in Enterprises France : Number of communications by employee and by day 8 0,08 7 1,92 6 5 SMS 1,49 eMails Mobile 4 3 Fixed Traffic 3,84 2 1 0 1998 1999 2000 2001 2002 2003 2004 2005 2006 Source : FT/DPS 2003 11 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (9/10) n Falling Prices and Falling Revenue Lowest Available Retail Price for Calls from Germany Australia Poland Turkey 2000 Austria 1997 UK France USA 0 0,5 1 DM/minute 12 1,5 2 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment (10/10) n Price Evolution : France Telecom’s Prices for Fixed Voice Traffic CAGR (1997 – 2002) 100 90 80 Line rental (€/month) 70 Local call (€/100) 60 + 5.0% - 2.8% National call (€/100) -13.3% 50 International call (€/100) 40 -12.4% Fixed to mobile (€/100) 30 Mobile to mobile (€/100) 20 - 9.5% -16.9% 10 0 1994 1996 1998 2000 2002 13 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Changing Telecom Environment : Conclusion The Modeling and Forecasting Techniques in the Telecom Industry are significantly and highly impacted by this changing environment : l It is necessary to Understand the real issues facing Operators in a newly Liberalized & deregulated environment l These changes must lead the historical players (incumbents) to “re-engineer”, adapt and reshape their Forecasting processes and techniques to face and challenge the new environment l These market changes must drive “new” and more suitable forecasting Techniques 14 Limitations of “Traditional” Forecasting Methods and Models in the Context of Competitive Markets Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Using inadequate Forecasting Model : an example 1600 1400 1600 Traf_D 1400 Forecast_D Traf_D Traf_D 1600 Forecast_D 1400 1200 1200 1000 1000 1000 800 800 800 600 600 600 400 400 400 200 200 0 0 200 Univariate B&J Univariate B&J Monopoly situation Transition • Jan 1998 : Deregulation of Telecom Market • Traffic Forecast for 1998 accurate. • Forecasting Model = Univariate B&J for 1998 • Until early 1998 : Telecom environment was stable. • In 1998 : Some New players entered the market • 1999 : Deregulation of Telecom Market has produced its early effects • Traffic Forecast for 1999 not accurate. • Forecasting Model = Univariate B&J for 1999 • Starting 1999 : Telecom environment was instable because full competition • In 1999 : Increase of the number of New players in the market 16 jan v-9 6 ma i-96 se pt96 jan v-9 7 ma i-97 se pt97 jan v-9 8 m ai98 se pt98 jan v-9 9 m ai99 se pt99 jan v-0 0 ma i-00 se pt00 0 jan v-9 6 av r-9 6 jui l-9 6 oc t-9 6 jan v-9 7 av r-9 7 jui l-9 7 oc t-9 7 jan v-9 8 av r-9 8 jui l-9 8 oc t-9 jan 8 v-9 9 av r-9 9 jui l-9 9 oc t-9 9 jan v-9 6 av r-9 6 jui l-9 6 oc t-9 jan 6 v-9 7 av r-9 7 jui l-9 7 oc t-9 jan 7 v-9 8 av r-9 8 jui l-9 8 oc t-9 8 jan v-9 9 av r-9 9 jui l-9 9 oc t-9 9 1200 Forecast_D Univariate B&J Competitive Market • 2000 : Deregulation of Telecom Market has produced its full effects • Traffic Forecast for 2000 not accurate et MAPE has shown an increasing error. • Forecasting Model = Univariate B&J for 2000 • Very important change in the actual traffic • In 2000 : Telecom market was widely opened and fully competitive • In 2000 : Increase of the number of New players in the market Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Adapting the Forecasting Model to the new context 1600 1400 Traf_D 1600 Forecast_D 1400 Traf_D Traf_D 1600 Forecast_D 1400 1200 1200 1000 1000 1000 800 800 800 600 600 600 400 400 400 200 200 0 0 Univariate B&J Monopoly situation • Jan 1998 : Deregulation of Telecom Market • Traffic Forecast for 1998 accurate. • Forecasting Model = Univariate B&J for 1998 200 Univariate B&J with Intervention Function Hubbing & Refiling Transition jan v-9 6 m ai96 se pt96 jan v-9 7 ma i-9 7 se pt97 jan v-9 8 ma i-9 8 se pt98 jan v-9 9 ma i-9 9 se pt99 jan v-0 0 ma i-0 0 se pt00 0 jan v-9 6 av r-9 6 jui l-9 6 oc t-9 6 jan v-9 7 av r-9 7 juil -97 oc t-9 jan 7 v-9 8 av r-9 8 jui l-9 8 oc t-9 jan 8 v-9 9 av r-9 9 juil -99 oc t-9 9 jan v-9 6 av r-9 6 jui l-9 6 oc t-9 jan 6 v-9 7 av r-9 7 juil -97 oc t-9 jan 7 v-9 8 av r-9 8 jui l-9 8 oc t-9 8 jan v-9 9 av r-9 9 jui l-9 9 oc t-9 9 1200 Forecast_D Multivariate B&J Transfer Function Prices, Competition, LCR Competitive Market • 1999 : Deregulation of Telecom Market has produced its early effects • 2000 : Deregulation of Telecom Market has produced its full effects • Traffic Forecast for 1999 accurate. • Traffic Forecast for 2000 accurate et MAPE has shown a stable error. • Forecasting Model = Univariate B&J with Intervention Function for 1999 • Until early 1998 : Telecom environment was stable. • Starting 1999 : Telecom environment was instable because full competition • In 1998 : Some New players entered the market • In 1999 : Increase of the number of New players in the market 17 • Forecasting Model = Multivariate B&J for 2000 • The new model feets well the changing trend of traffic • In 2000 : Telecom market was fully competitive • In 2000 : Increase of the number of New players in the market Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Selection of the Explanatory Variables Main drivers in a Monopoly situation The main drivers in a Monopoly Telecom market are the Economy and the Customer features : Economy Economy : Demand Forecast Consumption, Price index, Investments , Foreign Trade … Traffic Revenues Capacity Users Customers ……. Customer : demand, segmentation, price elasticity by customer segment, …. Customers The main drivers in a Competitive Telecom market are : Economy Customers needs and their perceived value, detailed segmentation, … Regulation interconnexion rules, offerings Pricing Strategy adaptative pricing, yield Main drvers in a Competitive Environment Economy Competitors Substitute Products Regulation Innovation/Technology management Competition, market shares, players, … Product substitution Technology & Innovation GNP, GDP, Income, Customers Demand Forecast Traffic Revenues Capacity Users Customers ……. QoS (Quality of Services) Company Developments Company Goals 18 Pricing / Yield Management Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Main Forecasting Methods used Before the 1998 Liberalization n An Example of a Previous Econometric Model used for the Global Outgoing IDD Traffic Explanatory Variable LOG(Price) LOG(Time)*LOG(Price) LOG(Foreign Trade) AR(1) R2 Adj. R2 Durbin-Watson Coefficient -0.471823 0.048264 0.477754 0.373937 0.984826 0.984320 2.369308 Stand. Error 0.013197 0.003987 0.001810 0.090575 t-Statistic -35.75160 12.10482 263.9157 4.128461 Prob. 0.0000 0.0000 0.0000 0.0001 n Before 1998, this type of Econometric model has generally provided accurate forecasts : l For Global Traffic l For Main Customer Segments : Business and Residential l For Main Geographical Segmentation : Europe, Asia, NA, … n For regulated markets, this type of Econometric model could provide accurate forecasts : l By country (Algeria, Viet-Nam, Cambodia, China, Saudi Arabia, Tunisia, …) l By area : Middle East, North Africa, South America, … 19 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Main Forecasting Methods used Before the 1998 Deregulation n An Example of Previous Forecasting Models by Geographical Segments for Global Outgoing IDD Traffic Geographical Segments Models/Methods used Explanatory variables NAFTA USA-CANADA-MEXICO Econometric Multivariate ARIMA Market model Econometric Univariate ARIMA Market model Econometric/Dynamic Reg. Uni.ARIMA/H&W Market model Imports/Exports index Call prices communication consumption Competition, Hubbing Refiling Imports/Exports index GNP/GDP - Call prices communication consumption Competition Hubbing Refiling Imports/Exports index GNP/GDP - Call prices communication consumption Imports/Exports index GNP/GDP - Call prices communication consumption Econometric/Dynamic reg. Uni. ARIMA/H&W Market model X11 Regression / H&W Market model Imports/Exports index GNP/GDP - Call prices Global investments in ME Tourism index GNP/GDP - Call prices Western Europe Japan/HongKong Taiwan/S.Korea South America Middle East/Africa Caribbean (ex. Martinique & Guadeloupe) Econometric Uni/Multivar. ARIMA Market model 20 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector The Forecasts Accuracy n An Example of a Previous Econometric Model for the Global Outgoing IDD Traffic 21 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Enhancing the Modeling Process (1/4) n The Econometric Modeling • To enhance the quality of the forecast (accuracy) • To have a better comprehension and understanding of the Market Structure : Ø for example, how each explanatory variable drive the demand ? • To learn the relationship between the dependent variable (to be forecasted) and its environment : evaluation of the Economy, Price, Competition, … impacts on the dependent variable (contribution) • To assess and test and simulate some alternative scenarios and hypothesis : Ø For example, what is the impact on mobile voice traffic of 15% reduction of price ? • To catch the dynamic of the market environment (lagged explanatory variables) Ø For example, when (after how many months or weeks) a 10% reduction of price will increase the demand ? This could allow the operator to adapt the periodicity of some actions through the time (special offers, …) 22 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Enhancing the Modeling Process (2/4) n Time series Modeling • We can integrate in some Time Series Models the effects of : Ø specific actions/events (dummy variables, intervention variable) Ø Multivariate Time Series Analysis : including explanatory variables (Transfer Function Modeling, i.e Box & Jenkins with Transfert function or Kalman Filtering) n Growth Curves/S-Curves • These models are adequate for new product studies Ø The Logistic (Local, Extended, …) Ø Gompertz Ø Exponential 23 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Enhancing the Modeling Process (3/4) n A specific Model for each Customer Segment : by customer, by market, by product, ... For example : • By Age • By Income • By Occupation • By overall Spend • By business sector The benefit is to learn for example how each Type of Customer reacts to the Price, Economic, Competition … variables - For example, the Young Customer has a high Price elasticity . The « High Revenues » Customer has a high Economy elasticity than Price elasticity. - This information could help the operator for its Marketing strategy (Packages for Young Customers, Prepaid cards for…..) 24 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Enhancing the Modeling Process (4/4) n The need of Information and Data in the Competitive markets There is a very important need of onformation and data concerning « new » drivers to be considered in a competitive market : • Competition • Innovation • Pricing • Regulation • Customer attitudes and choices • QoS • Substitution Products • Marketing • Strategy • Sales How to collect them : • Market intellingence • Benchmark studies • Consultancy sources : Ovum, IDC, Giga, Yankee Group, Idate, Frost& Sullivan, …… • Customer surveys / Trade Off studies • Int’l Offices os Statistics : ITU, WB, IMF, EU, • National Offices os Statistics : INSEE, Census, … • Other sources : Wefa, …. ØIt is absolutely important to improve the « quality » of the Data to be integrated in the forecasting models ØIt is « vital » to track « new » information concerning the competitive market 25 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Limitations of traditional forecasting methods : Conclusion The Forecasting Process (Modeling, Estimating, Accuracy) must be adapted continuously to the changing environment : l The market changes must drive “new” and more adequate forecasting Techniques (for example the Simultaneous Equation Models) l Telecom Forecasting Models must take into account : l l More adequate and accurate explanatory variables l The dynamic of demand and supply : lagged variables l The Necessity of more detailed Demand Segmentation l The Players behaviour and Strategies (Alliances, …) Using Econometric Models solely could lead to inconsistent and not accurate Forecasts : l It is necessary to integrate in the Forecasting Process other contributions outside the Econometric Model (see Case Study in 3 rd Part) : Learning from Customer Surveys, Market Studies, Trade Off studies, … and experience l Integrating a flexible pricing strategy : stable Price elasticity ? l Learning from the experience of deregulated and competitive telecom markets : UK, USA, … l Need for frequent re-forecasting and adaptative forecasts l Integrating new traffic routing strategies (LCR, hubbing, refiling,…) 26 A Case Study from an important Operator Experience in the International Traffic Outlining the Forecasting Methods undertaken and how they held up to the Challenges of Liberalization in 1998 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Adoption of New Approaches and Forecasting Models n Building a new approach for Incoming International Traffic (IDD Traffic) Module Forecast Econometric Model Routing : Fragmentation Routing : - Concentration - Repartition Expected Results Total Incoming Traffic to France Total Traffic from origin country to France by operator / carrier Total Incoming Traffic to France - by Route & by operator - b y d e s t i n a t i o n o p e r a t o r/ c a r r i e r 28 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (1/9) n Module 1 : Global Traffic between Origin country and France : l “Traditionnal” Econometric Modelling & Estimation l But with More suitable explanatory variables Forecast Econometric Model Tran sit » g n i ubb «H Country A Total Incoming Traffic to France Estimation of GlobalTraffic from Country A to France France n Transit & “ hubbed” traffics to country A are not included in the Global traffic l They should be added outside the model 29 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (2/9) n Information & Data Set : l Annual Data from 1990 to 1998 l Traffic between 54 countries , i.e 9 years x 54 countries x 53 country-pairs n Model Specification & Estimation 30 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (3/9) n Actual & Predicted data 31 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (4/9) n Module 2 : Fragmentation of the Incoming Traffic from Origin Country to France : Routing : Fragmentation Total Traffic from origin country to France by operator / carrier OP1 CAR1 ….. Tran sit » ng i b ub «H Country A Traffic to France by Route and by Operator/Carrier France CAR2 OP2 n Transit & “ hubbed” traffics to country A by Route and by Carrier/Operator must be added to the global Traffic n Assumptions and hypothesis are defined for actual and future situation of players & market : l Alliances and Partnership Policy l Liberalization evolution and Telecom players behaviour in the origin country 32 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (5/9) Detail Resellers Local Access C's C ORI2 ORI1 Resellers Methodology Local Access n Fragmentation of Traffic in the Origin Country Main categories of operators/carriers considered (couldiffers from a country to another country) Int’l Traffic origination to France Traffic Concentration chain in the Origin country (Retail → wholesale) Improbable Case ORI1 Could happend but rarely Traffic to be sent from the origin country to France ORI2 C's C Traffic from origin Country A to France Traffic to France (excluding hubbing) Int’l Traffic from other countries to A (‘hubbed’) Total Int’l Traffic to France From concentration module in third country To concentration & Distribution modules 33 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (6/9) n Fragmentation of Traffic in the Origin Country 34 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (7/9) n Module 3 : Traffic Concentration in the Origin Country & distribution in France Routing : - Concentration - Distribution OP1 Tra nsi t ng bbi u «H Total Incoming Traffic to France - by Route & by operator - by destination operator/carrier OP2 Route 1 » Country A OP1 Route 2 OP2 France Route n OP4 OP3 OP4 35 OP3 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (8/9) n Concentration of Traffic to France by Route and by Carrier/Operator Countries and Operators/Carriers Geostrategic Criterion Possible Routes Economic Criterion Percentage of Traffic Managed by LCR Tariffs of Half-circuts Int’l Leased Lines Cost of End-to-End Route Traffic on the Considered Route interconnection Cost In France Cost of « Hubbing » Route Commutation Cost 36 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Forecasting the Int’l Incoming Traffic (9/9) n Distribution of Int’l Incoming Traffic to France by Carrier/Operator Operator Traffic Traffic on Operator bilateral route + Traffic on « Third » party bilateral route + End to End Traffic (if Affiliated) Competitors Traffic Traffic on the Carrier’s bilateral route + Traffic on « Third » party bilateral route + End to End Traffic (if partnership) 37 Conclusion Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector Conclusion : The Forecasting process (modeling, estimating, accuracy checking) must be adapted continuously to the changing environment : l l l The market changes must drive “new” and more adequate forecasting Techniques l “Refresh” the forecasting model & approaches l Build alternative model & approaches when necessary l Use alternative scenarios to assess the demand & supply Telecom Forecasting Models must take into account : l More adequate explanatory variables l The dynamic of demand and supply : lagged variables l The Necessity of Demand Segmentation l The Players behaviour and Strategies (Alliances, …) l The flexibility of pricing strategy l New traffic routing strategies (LCR, hubbing, refiling,…) It is necessary and very useful : l to integrate in the Forecasting Process other contributions outside the Econometric Model : Learning from Customer Surveys, Market Studies, Trade Off studies. l To learning from the experience of deregulated and competitive telecom markets : UK, USA, Australia, Canada, … l To set-up and use a re-forecasting process 39 Adjusting Forecasting Methods To the Needs Of The Telecommunication Sector THANK YOU FOR YOUR ATTENTION mohsen.hamoudia@francetelecom.com Tel : Mobile : Fax : + 33 1 56 66 34 50 + 33 6 82 93 15 72 + 33 1 56 66 53 83 40