Forecasting Methods for Strategy Development

1
Forecasting Methods for Strategy Development

• Market Demand Analysis

2
Quote for the Day

• There are two kinds of forecasters. Those who
  dont know and those who dont know that they
  dont know.
• - John Kenneth Galbraith

3
Two Kinds of Forecasts

• Passive forecasts (assumes demand exists and we
  simply measure it). Can help us determine when
  to allocate strategic resources.
• e.g., time series methods
• Active forecasts (assumes we can influence demand
  through marketing efforts). Can help us
  determine when, where, and how much of
  strategic resources to invest.
• e.g., causal analysis

4
Forecasting Methods

• Market Judgmental Survey Time
  Causal Methods Series Analyses
• Salesforce Buyer Naive methods Regression
  analysis composite intentions Moving
  averages Econometric modelsJury of
  executive Product tests Exponential Input-output
  opinion smoothing analysisDelphi
  methods Box-Jenkins MARMA method Neural
  networks Decompositional MNL analysis methods

5
Three Methods We Will Use

• Regression
• Multinomial (and Binary) Logit Analysis
• Neural Networks

6
MNL Model for Response to Direct Mail

• Probability of function of (past response
  bahavior,
• responding to marketing effort,
• direct mail characteristics of
• solicitation customers)

7
Individual-Level MNL Models

• Multinomial logit model to represent probability
  of choice. Individual is probability of
  choosing brand 1 is
• eAi1
• Pi1
• å eAij
• j
• where Aij å wk bijk
• k
• Aij is the attractiveness of alternative j, and
  is a linear composite of selected
• variables (bijk) that vary by brand (j) and
  individual (i).

8
Variables used in Bookbinders Case

• Amount purchased in the past
• Frequency of purchase
• Last purchase
• First purchase
• Number of childrens books purchased
• Number of youth books purchased
• Number of cookbooks purchased
• Number of DIY books purchased
• Number of art books purchased

9
Model Estimation

• Dependent variable is choice - whether someone
  purchased/selected a product on a purchase
  occasion (i.e., variable is 0 or 1).
• Estimation is by maximum likelihood methods.

10
Artificial Neural Networks for Modeling Market
Response

• What is a neural network?
• Determinants of network properties
• Description of feed-forward network with back
  propagation
• Potential value of neural networks

11
What Is An Artificial Neural Network?

• An artificial neural network is a general
  response model that relates inputs (e.g.,
  advertising) to outputs (e.g., product
  awareness).
• A neural net attempts to mimic how the human
  brain processes input information and consists of
  a richly interlinked set of simple processing
  mechanisms (nodes).

12
Characteristics of Biological Neural Networks

• Massively parallel
• Distributed representation and computation
• Learning ability
• Generalization ability
• Adaptivity
• Inherent contextual information
• Fault tolerance
• Low energy consumption

13
Artificial Neural Network
Neurons
Inputs In humanssensory data. In
4Thoughtadvertising, selling effort, price, etc.
Outputs In humansmuscular reflexes. In
4Thoughtsales model.
Synapses
14
Determinants of the Behavior of Artificial Neural
Network

• Network properties (depends on whether network is
  feedforward or feedback number of nodes, number
  of layers in the network, and order of
  connections between nodes).
• Node properties (threshold, activation range,
  transfer function).
• System dynamics (initial weights, learning rule).

15
Processing Mechanism of Individual Neurons

• Convert input signals into overall signal value.
• Z å Wi Xi
• i
• Transform overall signal value into output signal
  (transfer function).

16
Transfer Function Formulations

• Hard limiter (Y 1 if Z T else 0)
• Sigmoidal (0 Y 1)
• 1
• Y g(Z)
• 1 e(ZT)
• Tanh (1 Y 1)
• Y g(Z) tanh (Z T)

17
System Dynamics(Learning Mechanism)

• Supervised learning using back propagation of
  errors. Goal of this process is to reduce the
  total error at output nodes
• EP å (tPk OPk)2
• k
• where
• EP error to be minimized
• tPk target value associated with the kth
  input values
• OPk Output of neural net as calculated from
  current set of weights.

18
Error Propagation

• The error is calculated at each node for each
  input set k
• The error at the output node is equal to
• diL g (ZiL)tiL YiL
• where
• TiL Target value on the i-th output node
  (layer L of network)
• diL Error to be back propagated from node i
  in layer L
• g gradient of transfer function.

19
Error Propagation

• Error is propagated back as follows
• dil g (Zil) å wijl1 djl1
• j
• for l (L1), . . . 1.
• The weights are then adjusted using an
  optimality rule (in conjunction with a learning
  rate) to minimize overall error EP.

20
So, Whats the Big Deal?

• With a sigmoidal transfer function and back
  propagation, the neural network can learn to
  represent any sampled function to any required
  degree of accuracy with a sufficient number of
  nodes and hidden layers.
• This allows us to capture underlying
  relationships without knowing the form of the
  relationship.

21
Some Successful Applications

• Recognizing handwritten characters (zip codes)
• Recognizing speech
• Estimating response to direct mail operations

22
Choosing a Forecasting Method
Objective data available?
Judgmental method
No
Yes
New product situation?
New product methods (ch. 7)
Yes
No
Large changes in environment?
Extrapolation/Time Series methods
No
Yes
No
Good information on relationships?
Neural nets
No
Yes
No
Much data on causal variables?
Causal method
No
Yes
No
Major data problems?
Yes