Forecasting Forecasting Problems and Methods New product forecasting Forecasting using Diffusion Models (to forecast trial or adoption) Forecasting using Pre-Test Market Models (to forecast both trial and repeat purchase) 1 Managerial Issues Related to Forecasting What is the purpose of developing the forecast? What, specifically, do we want to forecast (e.g., market demand, technology trends)? How important is the past in predicting the future? What influence do we have in constructing the future? What method(s) should we use to develop the forecast? What factors could change the forecast? © DecisionPro 2007 Principles Chapter 5: Forecasting - 2 Forecasting Methods Judgmental Salesforce composite Jury of executive opinion Delphi methods Scenario analysis Market and Survey Analysis Time Series Buyer intentions Naïve methods Product tests Moving averages Chain ratio method Exponential smoothing Box-Jenkins method Decompositional methods © DecisionPro 2007 Causal Analyses Regression analysis Econometric models Input-output analysis MARMA Neural networks Principles Chapter 5: Forecasting - 3 Methods for Forecasting New Product Sales Early stages of development Chain ratio method Judgmental methods Scenario analysis Diffusion model Later stages of development Pre-test market methods Test-market methods © DecisionPro 2007 Principles Chapter 5: Forecasting - 4 Chain Ratio Method (Estimate of Online Grocery Sales) Number of households (2000 census) Grocery purchases per household per year (52x120) % of sales from Supermarkets and grocery stores 105 million $5300 84% (Progressive Grocer) Households with children (married and unmarried – Census) % of households with Internet access (Census Bureau) Will order groceries online if available (Survey) Discount of survey intentions Online grocery shopping availability (guess) Awareness given availability (guess) 35% 58% 25% 50% 40% 50% Market forecast: $ ??? © DecisionPro 2007 Principles Chapter 5: Forecasting - 5 Intent-to-Buy Scale Used for Generating Some Inputs to Chain Ratio 1. Definitely would buy 2. Probably would buy 3. May or may not buy (May be excluded from the scale) 4. Probably would not buy 5. Definitely would not buy © DecisionPro 2007 Principles Chapter 5: Forecasting - 6 Who Are They? © DecisionPro 2007 Principles Chapter 5: Forecasting - 7 New Product Forecasting Models That We Consider Forecasting the pattern of new product adoptions (Bass Model) Forecasting market share for new products in established categories (Assessor pre-test market model) Forecasting using conjoint analysis © DecisionPro 2007 Principles Chapter 5: Forecasting - 8 Hi Forecasting Based on “Newness” of Products • Repositioning Pre-test market model New to World Lo • Line Extensions Simple pre-test market models (e.g., Bases) • Breakthroughs—Major Product Modifications Bass model/Conjoint • “Me Too” Products Conjoint/Pre-test market models Hi Lo New to Company © DecisionPro 2007 Principles Chapter 5: Forecasting - 9 Overview of “Stage-Gate” New Product Development Process Opportunity Identification Reposition Market definition Idea generation Harvest Life-Cycle Management Go Market response analysis & fine tuning the marketing mix; Competitor monitoring & defense Innovation at maturity No Design Identifying customer needs Sales forecasting Product positioning Engineering Marketing mix assessment Segmentation Go No Introduction Go No Launch planning Tracking the launch Testing Advertising & product testing Pretest & prelaunch forecasting Test marketing Go © DecisionPro 2007 No Principles Chapter 5: Forecasting - 10 The Bass Diffusion Model of New Product Adoption The model attempts to answer the question: When will customers adopt a new product or technology? Why is it important to address this question? Helps in planning major investments (e.g., building a factory) with respect to the product. © DecisionPro 2007 Principles Chapter 5: Forecasting - 11 Non-cumulative Adoptions, n(t) Graphical Representation of The Bass Model (Cell Phone Adoption) Adoptions due to internal influence pN Adoptions due to external influence Time © DecisionPro 2007 Principles Chapter 5: Forecasting - 12 Number of Registered Users eBay (by Quarter) million 225 210 195 180 165 150 135 120 105 90 75 60 1997 45 Q1 0.09 Q2 0.15 30 Q3 0.25 15 Q4 0.40 0 1997 '98 '99 '00 '01 '02 '03 '04 '05 '06 Source: eBay/SEC filings © DecisionPro 2007 Principles Chapter 5: Forecasting - 13 The Bass Diffusion Model for Durables nt = p Remaining Potential + q Adopter Proportion Remaining Potential Innovation Effect Imitation Effect nt = n umber of adopters at time t (Sales) p = “coefficient of innovation” (External influence) q = “coefficient of imitation” (“internal” to the society in which the diffusion spreads) N = Eventual number of adopters # Adopters = n0 + n1 + • • • + nt–1 Remaining = Total Potential – # Adopters Potential © DecisionPro 2007 Principles Chapter 5: Forecasting - 14 Assumptions of the Basic Bass Model Diffusion process is binary (consumer either adopts, or waits to adopt). Constant maximum potential number of buyers ( ). Eventually, all will adopt the product. N N or replacement purchase. No repeat purchase, The impact of word-of-mouth is independent of adoption time. Innovation is independent of substitutes. The marketing strategies supporting an innovation are not explicitly included. Uniform influence or complete mixing. That is, everyone in the population knows everyone else, or is at least able to communicate with, or observe everyone else. © DecisionPro 2007 Principles Chapter 5: Forecasting - 15 Representation as an Equation N (t ) n( t ) [ N N ( t )] p q N ...(1) N(t) : Cumulative number of adopters until time t. © DecisionPro 2007 Principles Chapter 5: Forecasting - 16 Parameters of the Bass Model in Several Product Categories Product/ Technology B&W TV Color TV Room Air conditioner Clothes dryers Ultrasound Imaging CD Player Cellular telephones Steam iron Oxygen Steel Furnace (US) Microwave Oven Hybrid corn Home PC Innovation parameter (p) Imitation parameter (q) 0.065 0.021 0.010 0.073 0.003 0.028 0.005 0.036 0.001 0.018 0.000 0.003 0.335 0.583 0.454 0.389 0.506 0.368 0.506 0.318 0.456 0.337 0.798 0.253 A study by Van den Bulte and Stremersch (2004) suggests an average value of 0.03 for p and an average value of 0.42 for q, The average was taken across a couple of hundred categories. © DecisionPro 2007 Principles Chapter 5: Forecasting - 17 Estimating the Parameters of the Bass Model Estimation using data Regression Specialized nonlinear estimation Estimation using analogous products Select analogous products based on the similarity in environmental context, market structure, buyer behavior, marketing-mix strategies of the firm, and innovation characteristics. © DecisionPro 2007 Principles Chapter 5: Forecasting - 18 Forecasting Using the Bass Model— Room Temperature Control Unit Quarter Market Size = 16,000 (At Start Price) Innovation Rate = 0.01 (Parameter p) Imitation Rate = 0.41 (Parameter q) Initial Price = $400 Final Price = $400 0 1 4 8 12 16 20 24 28 32 36 Sales 0 160 425 1,234 1,646 555 78 9 1 0 0 Cumulative Sales 0 160 1,118 4,678 11,166 15,106 15,890 15,987 15,999 16,000 16,000 Example computations n( t ) pN (q p) N ( t 1) (q / N ) N 2 ( t 1) Sales in Quarter 1 = 0.01 16,000 + (0.41–0.01) 0 – (0.41/16,000) (0)2 = 160 Sales in Quarter 2 = 0.01 16,000 + (0.40) 160 – (0.41/16,000) (160)2 = 223.35 © DecisionPro 2007 Principles Chapter 5: Forecasting - 19 Factors Affecting the Rate of Diffusion Product-related High relative advantage over existing products High degree of compatibility with existing approaches Low complexity Can be tried on a limited basis Benefits are observable Market-related Type of innovation adoption decision (e.g., does it involve switching from familiar way of doing things?) Communication channels used Nature of “links” among market participants Nature and effect of promotional efforts Source: Everett Rogers © DecisionPro 2007 Principles Chapter 5: Forecasting - 20 Some Extensions to the Basic Bass Model Varying market potential As a function of product price, reduction in uncertainty in product performance, and growth in population, and increases in retail outlets. Incorporating marketing variables Incorporating repeat purchases Multi-stage diffusion process Awareness Interest Adoption Word of mouth Incorporating Network Structure © DecisionPro 2007 Principles Chapter 5: Forecasting - 21 Example Application of Bass Model DirecTV (History and Technology) 1984 FCC grants GM Hughes approval to construct a Direct Broadcast Satellite system (DBS) High Ku Band frequency Early 1990’s technological breakthrough in digital compression. Result: Affordable product and nonobtrusive dish and equipment Changed economics of DTH broadcasting 1991 DIRECTV founded © DecisionPro 2007 Principles Chapter 5: Forecasting - 22 DirecTV Data Collection Method CATI (Computer-Assisted Telephone Interview) data collection - nationally representative sample of TV viewers. 15-minute phone interview. “Eligibles” assigned to one of two monadic concept-price cells (“Intent to Buy”). Respondents mailed a color brochure that described DIRECTV/RCA branded Direct Broadcast System concept. Phone callback interview (22 minutes)-Key inputs: Stated Intentions (Probability of Acquire and Perceived value and Affordability). © DecisionPro 2007 Principles Chapter 5: Forecasting - 23 Obtaining p, q, and N Guessed p and q from analogous previously introduced product N obtained from stated intentions in survey Average stated intent from survey = 32% Stated intentions overstate actual choices. How much to discount stated intent to adopt? (They discounted by 50%) Also, have to adjust each year’s predicted sales for actual levels of awareness and availability of product in the entire market. © DecisionPro 2007 Principles Chapter 5: Forecasting - 24 Adjusting Stated Intentions to Get Actual Purchase Behavior Probability of purchase given stated intent for new durable and non-durable products. From Jamieson, Linda F. and Frank M. Bass "Adjusting Stated Intention...To Predict Trial Purchase of New Products," JMR, August 1989. 45 Probability of Purchase (within six months) Increases with Stated Intention 40 Probability of Purchase 35 30 Some Who Say They Will, Don’t 25 20 Some Who Say They Won’t, Do! Purchase Increases with Stated Intention 15 10 5 0 Definitely Will Not Buy Probably Will Not Buy Might or Might Not Buy Actual Purchase Probablity Given Stated Intention for 5 Non-Durable Products © DecisionPro 2007 Probably Will Buy Definitely Will Buy Actual Purchase Probability Given Stated Intention for 5 Durable Products Principles Chapter 5: Forecasting - 25 Multi-Year Forecast and Actual Year 7/01/94 - 6/30/95 7/01/95 - 6/30/96 7/01/96 - 6/30/97 7/01/97 - 6/30/98 7/01/98 - 6/30/99 1992 Forecast Number of TV Homes Acquiring Satellite Television (Million) 0.875 2.269 4.275 6.775 9.391 Actual Number of TV Homes Acquiring Satellite Television (Million) 1.15 3.076 5.076 7.358 9.989 1992 Forecast of Percent of TV Homes with Satellite Television (Percentage) 0.92 2.37 4.42 6.95 9.55 Actual Yearly Percent of TV Homes with Satellite Television (Percentage) 1.21 3.21 5.25 7.55 10.16 9.4 Million TV homes forecast for June 99; Actual = 9.9 Million Forecast based on p and q of Cable TV (other alternative considered was Color TV) and maximum penetration set to 16% of population (half that in the stated intent survey). © DecisionPro 2007 Principles Chapter 5: Forecasting - 26 Multi-Year Forecast-Actual Graph 92 Forecast Was Not Updated © DecisionPro 2007 Principles Chapter 5: Forecasting - 27 Using Scenario Analysis for Calibrating the Bass Model Structure a scenario as a flowing narrative, not as a set of numerical parameters. Include verbal descriptions such as “rapid experience effects,” “FCC adoption of digital standard,” etc. Ideally, each scenario should also include how the situation described in the scenario will be reached from the present position. Construct several scenarios that capture the richness and range of the “possibilities” relevant to a decision situation. Describe all the scenarios in the same manner, i.e., one is not more “vivid” than another. Focus your further analyses on scenarios that are internally consistent and plausible. Develop forecasts and strategies that are compatible with the scenarios. The strategies include: Robust actions that are resilient across scenarios (e.g., hedging, concurrent pursuit of multiple options, etc.) Contingent actions that postpone major commitments to the future. © DecisionPro 2007 Principles Chapter 5: Forecasting - 28 Steps in Scenario Planning (Example for Zenith HDTV) Identify the major stakeholders. Summarize the core trends that are relevant (technological, economic, social, etc.) within the time frame of interest. Articulate the main uncertainties (e.g., TV studio adoption of new filming methods). Construct an initial set of scenarios. Assess the consistency and plausibility of the scenarios. Create “themes” (i.e., a story with a name) that combine some trends into meaningful composites (e.g., a Japanese domination of hardware and American domination of software). Identify areas where you need more research (e.g., consumer acceptance) and seek additional information. Associate the final set of scenarios with potential product analogs for diffusion model, select p and q, and generate the forecasts. Evaluate strategic and tactical choices that will help you realize the forecasts in the most cost effective manner. © DecisionPro 2007 Principles Chapter 5: Forecasting - 29 Example “Middle of the Road” Scenario (Zenith HDTV case) The FCC makes a commitment to the 16:9 NTSC HDTV standard in 1994, with promises to release details in a year. Initial HDTV sets cost over $3,000 and are seen as a luxury item, little programming is available so new features (such as use as computer monitors and compatibility with analog signals) are integrated to justify purchases. Art studios and other display locations become innovators as they purchase units for displays. Interior designers realize the benefits of HDTV plasma screens and suggest purchases to their wealthiest clients. HDTV becomes a “nouveau riche” item, a status symbol much like luxury cars. By 2000, the manufacturing costs of Plasma and other flat-screen displays decrease drastically from standards integration and increased competition. Middle-class customers can now afford HDTV displays. The movie industry embraces digital recordings because of the ease in editing and persistent quality. New movie features (screen and TV) are filmed in 16:9 digital format. Subsequent releases on DVD show higher quality. Public TV stations cannot justify the cost of upgrading, but cable channels such as HBO and Showtime commit to upgrading in 2003. Their recent entry into movie-making and their purchase of new high-tech digital recording equipment coincides with the need to upgrade transmission hardware. Customers are then driven to adopt technology not for increased quality on regular programming, but for movie watching, design, and display of other items. © DecisionPro 2007 Principles Chapter 5: Forecasting - 30 Comparative Trajectories of Population/GDP From Global Scenario Group Gross World Product ($ trillions) 250 Conventional Worlds Great Transition Eco-communalism Policy Reform Market Forces New sustainability paradigm Fortress World 20 1990 5 Breakdown Population (billions) © DecisionPro 2007 Barbarization 10 Principles Chapter 5: Forecasting - 31 Pretest Market Models Objective Forecast sales/share for new product before a real test market or product launch Conceptual model Awareness Availability Trial Repeat Commercial pre-test market services Yankelovich, Skelly, and White Assessor Others (e.g., BASES) © DecisionPro 2007 Principles Chapter 5: Forecasting - 32 Yankelovich, Skelly and White Model (Chain Ratio Method) Forecast market share = S N C R U K where: S = Lab store sales (indicator of trial), N = Novelty factor of being in lab market. Discount sales by 20–40% based on previous experience that relate trial in lab markets to trial in actual markets, C = Clout factor which retains between 25% and 75% of SN determined, based on proposed marketing effort versus ad and distribution weights of existing brands in relation to their market share, R = Repurchase rate based on percentage of those trying who repurchase, U = Usage rate based on usage frequency of new product as compared to the new product category as a whole, and K = Judgmental factor based on comparison of S N C R U K with Yankelovich norms. The comparison is with respect to factors such as size and growth of category, new product’s share derived from category expansion versus conversion from existing brand. © DecisionPro 2007 Principles Chapter 5: Forecasting - 33 Overview of ASSESSOR Modeling Procedure Consumer Research Input (Laboratory Measures) (Post-Usage Measures) Management Input (Positioning Strategy) (Marketing Plan) Preference Model Trial & Repeat Model Reconcile Outputs Draw & Cannibalization Estimates Brand Share Prediction © DecisionPro 2007 Unit Sales Volume Diagnostics Principles Chapter 5: Forecasting - 34 Overview of ASSESSOR Measurement Process Design O1 O2 X1 [O3] X2 O4 X3 O5 Procedure Measurement Respondent screening and recruitment (personal interview) Pre-measurement for established brands (self-administrated questionnaire) Exposure to advertising for established brands and new brands Measurement of reactions to the advertising materials (selfadministered questionnaire) Simulated shopping trip and exposure to display of new and established brands Purchase opportunity (choice recorded by research personnel) Home use/consumption of new brand Post-usage measurement (telephone Criteria for target-group identification (e.g., product-class usage) Composition of ‘relevant set’ of established brands, attribute weights and ratings, and preferences Optional, e.g. likability and believability ratings of advertising materials Brand(s) purchased New-brand usage rate, satisfaction ratings, and repeat-purchase propensity; attribute ratings and preferences for ‘relevant set’ of established brands plus the new brand O = Measurement; X = Advertising or product exposure © DecisionPro 2007 Principles Chapter 5: Forecasting - 35 Predicted and Observed Market Shares for ASSESSOR Product Description Deodorant Antacid Shampoo Shampoo Cleaner Pet Food Analgesic Cereal Shampoo Juice Drink Frozen Food Cereal Etc. Average Average Absolute Deviation Standard Deviation of Differences Initial Adjusted Actual Deviation (Initial – Actual) 13.3 9.6 3.0 1.8 12.0 17.0 3.0 8.0 15.6 4.9 2.0 9.0 ... 11.0 10.0 3.0 1.8 12.0 21.0 3.0 4.3 15.6 4.9 2.0 7.9 ... 10.4 10.5 3.2 1.9 12.5 22.0 2.0 4.2 15.6 5.0 2.2 7.2 ... 2.9 –0.9 –0.2 –0.1 –0.5 –5.0 1.0 3.8 0.0 –0.1 –0.2 1.8 ... 0.6 –0.5 –0.2 –0.1 –0.5 –1.0 1.0 0.1 0.0 –0.1 –0.2 0.7 ... 7.9 — — 7.5 — — 7.3 — — 0.6 1.5 2.0 0.2 0.6 1.0 © DecisionPro 2007 Deviation (Adjusted – Actual) Principles Chapter 5: Forecasting - 36 ASSESSOR Trial & Repeat Model Market Share Due to Advertising Response Mode •Max trial with unlimited Ad •Ad$ for 50% max. trial •Actual Ad $ •Max awareness with unlimited Ad •Ad $ for 50% max. awareness •Actual Ad $ Manual Mode % buying brand in simulated shopping Awareness estimate % making first purchase GIVEN awareness & availability 0.42 Prob. of awareness 0.70 Distribution estimate Prob. of availability 0.80 Switchback rate of non purchasers 0.16 Generalization of Assessor implementation As implemented in Assessor Repurchase rate for purchasers 0.42 © DecisionPro 2007 % making first purchase due to advertising 0.235 Long-term market share from advertising 0.049 Retention rate GIVEN trial for those who saw ad 0.211 Source: Adapted from Thomas Burnham Principles Chapter 5: Forecasting - 37 ASSESSOR Trial & Repeat Model Market Share Due to Sampling Sampling, Number Delivered 30M Proportion of market using samples 12.96/40 = 0.32 Correction for sampling/ad overlap 0.075 Cumulative trial (previous chart) 0.235 % Delivered 0.90 % of those delivered hitting target 0.80 Assumes 40 million households in target market Sample use in simulation 0.60 Switchback rate for non-purchasers in previous time period Repurchase rate of those not buying in simulation Net incremental trial 0.245 Prob. of switching to brand 0.15 Prob. of repurchase of brand 0.26 Long term repeat rate for sample receivers 0.169 First repeat for those not buying in simulation 0.26 Long-term market share from sampling 0.011 Source: Adapted from Thomas Burnham © DecisionPro 2007 Principles Chapter 5: Forecasting - 38 ASSESSOR Preference Model Summary Pre-use preference ratings Pre-use constant sum evaluations Beta (B) for choice model Pre-entry market shares Pre-use choices Post-use constant sum evaluations Post-use preference ratings Cumulative trial from ad (T&R model) 0.202 Proportion of consumers who consider product 0.235 Post-entry market shares (assuming consideration 0.243 Predicted post entry market shares 0.057 Draw & cannibalization calculations Source: Adapted from Thomas Burnham © DecisionPro 2007 Principles Chapter 5: Forecasting - 39 ASSESSOR Market Share to Financial Results Diagrams Market share 0.06 Market size 40M Industry average sales for realized market share 52.8M Average annual sales per household $22 Company factory sales 49.6M Average unit margin 0.581 Ad/sampling expense 4.0/6.0M Company factory sales 49.6M Unit-dollar adjustment 0.94 Frequency of use differences 0.9 Net Contribution 18.82M Company factory sales 49.6M Price differences 1.04 Return on sales 38% Note: Market share from Trial/Repeat Model: 0.060 Market Share from Preference Model: 0.057 Source: Adapted from Thomas Burnham © DecisionPro 2007 Principles Chapter 5: Forecasting - 40 Recap Judgmental methods and Chain ratio approach can be applied in a wide range of forecasting situations. We will cover one judgmental method (Delphi method) when discussing Resource Allocation models developed based on managerial judgment. Bass diffusion model is useful for forecasting the adoptions of a new to the world product (e.g., a new technology or trend) Pre-test market models are useful for forecasting products that have repeat purchase potential (e.g., consumer packaged goods). © DecisionPro 2007 Principles Chapter 5: Forecasting - 41