Scribd
Upload
77 views

Purchase Incidence:: Purchase Incidence Purchase Timing Brand Choice Integrated Models of Incidence, Timing and Choice

1. The document discusses stochastic models of consumer behavior including purchase incidence, purchase timing, and brand choice models. 2. It describes several purchase incidence models including the Poisson, Negative Binomial, and Zero-Inflated Poisson models. The Negative Binomial model incorporates heterogeneity across consumers. 3. Purchase timing models include hazard models which account for right-censoring of interpurchase times. Brand choice models include Markov, Bernoulli, and learning models as well as models incorporating marketing variables like the Multinomial Logit and Markov Response models. 4. Integrated models combine purchase incidence, timing, and choice, obtaining a joint Negative Binomial-Dirichlet Multinomial distribution through heterogeneity distributions.

Uploaded by

Vishal Tomar
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
77 views

Purchase Incidence:: Purchase Incidence Purchase Timing Brand Choice Integrated Models of Incidence, Timing and Choice

1. The document discusses stochastic models of consumer behavior including purchase incidence, purchase timing, and brand choice models. 2. It describes several purchase incidence models including the Poisson, Negative Binomial, and Zero-Inflated Poisson models. The Negative Binomial model incorporates heterogeneity across consumers. 3. Purchase timing models include hazard models which account for right-censoring of interpurchase times. Brand choice models include Markov, Bernoulli, and learning models as well as models incorporating marketing variables like the Multinomial Logit and Markov Response models. 4. Integrated models combine purchase incidence, timing, and choice, obtaining a joint Negative Binomial-Dirichlet Multinomial distribution through heterogeneity distributions.

Uploaded by

Vishal Tomar
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 5

Stochastic Consumer Behavior Models

Stochastic models of consumer behavior are often classified according to the type of behavior they attempt to describe. The major categories are:

• Purchase Incidence
• Purchase Timing
• Brand Choice
• Integrated models of incidence, timing and choice

Purchase incidence, purchase timing and brand choice are described by stochastic processes, i.e., families of random variables (X1) indexed by t varying in an index set T(t ϵT). A random or
stochastic variable is a variable which can assume different values (or alternative value intervals) with some probability other than one. The values the random variable can take are defined in
the state space which is the collection of possible outcomes of the stochastic process under study. For instance, if the process is the purchasing of a specific brand, the state space for one trial is
Y(es), N(o), and for two trials it is (YY, YN, NY, NN).

1. Purchase Incidence:

• THE POISSON PURCHASE INCIDENCE MODEL:


These models are based on the Poisson process, which has the property that the distribution of the number of units purchased in any interval depends only on the length of that interval. The
random variable ( Yit), denoting the number of units purchased by consumer i in a certain time period t, then follows a Poisson distribution with parameter λ.

• HETEROGENEITY AND THE NEGATIVE BINOMIAL (NBD) PURCHASE INCIDENCE MODEL:


The assumptions underlying the Poisson process are quite restrictive in many marketing applications. For example, the assumption that all consumers have the same value of λ is unrealistic.
Heterogeneity has been accommodated in several ways, most frequently by assuming that λ is a random variable that follows a gamma distribution across individuals:

Here, α & β are parameters of the gamma distribution and Γ(β ) is the gamma function. The gamma distribution is a very flexible distribution that can take on a variety of shapes
In this model, the number of purchases for a (randomly selected) individual can be shown to follow a Negative Binomial Distribution:

• THE ZERO-INFLATED POISSON (ZIP) PURCHASE INCIDENCE MODEL:


Another critique of the Poisson and NBD purchase incidence models is that they do not accommodate individuals who never buy. Both models predict that every individual will eventually buy
the product, as t increases.
One solution is to add a component to the model that allows for an additional spike at zero, due to the class of non-buyers, with proportion Π 0. The Zero-Inflated Poisson (ZIP) model is:

2. Purchase timing:
Purchase timing models are intrinsically related to purchase incidence models in that the choice of a distribution for interpurchase times at the same time defines the distribution of purchase
incidence.

• HAZARD MODELS:
The major advantage of this approach is that it accounts for so-called right-censoring. Right censoring occurs if a sample of consumers or households is observed for a time period of fixed
length only, causing longer interpurchase times to have a larger probability of falling (partially) outside the observation period. If one does not account for right censoring, we obtain biased
estimates.
In hazard models the probability of a purchase during a certain time interval, say t to t + Δt, given that is has not occurred before t, is formulated as:

where T is the random interpurchase time variable.

In discrete-time hazard models, the probability of a purchase is specified directly for given values of Δt :

where y is the number of events that has occurred during the interval [t, t + Δt ].
In the continuous-time approach Δt approaches zero to yield a continuous hazard rate λ(t). The hazard rate can be interpreted as the instantaneous rate of purchasing at time t. The distribution
function of interpurchase times can then be derived:

3. Brand choice models:


Brand choice models can be distinguished according to several criteria:
a. Assumptions about the stochastic process underlying the model. The brand choice behavior being constant in time (stationary) or not (non-stationary) yield, for example, Bernoulli,
Markov, and learning models.
b. Whether brand choice behavior is assumed to be homogeneous or heterogeneous across consumers.
c. Whether or not the effects of marketing variables are accommodated.

• MARKOV AND BERNOUILLI MODELS:


Discrete time Markov brand choice models are based on Markov chains, in which one considers probabilities such as:

i.e. the probability that brand j is purchased at time t, given that brand i was purchased at t-1, brand k at t-2, .. .. These probabilities are called transition probabilities.

Here, we will talk about 2 types of Markov Models: Zero order & First order model.

• ZERO ORDER MODELS:


One refers to a zero-order (Markov) model, if the probability of purchasing a particular brand at t does not depend on purchasing behavior at t -1, t-2, etc. In other words, a zero-order
model applies when current and future purchasing behavior does not depend on past purchase history. Thus,

• FIRST-ORDER MARKOV MODELS:


A first-order Markov model applies if only the last purchase has an influence on the present one, i.e.,
• LEARNING MODELS:
In learning models the entire purchase history is taken into account in each subsequent purchase. Learning may have a positive effect in the sense that increased familiarity with the product
increases the chance that it will be re-purchased in the future. The effect of learning may also be negative, for example due to satiation.

• BRAND CHOICE MODELS WITH MARKETING DECISION VARIABLES:

Here, we will see 2 models: Multinomial Logit Model & Markov Response Model.

• MULTINOMIAL LOGIT MODEL:


In a multinomial choice model an individual chooses between n alternatives. The assumption is that a consumer will choose the alternative that gives him maximal utility. Here, Each
individual i chooses the brand with the maximal utility. Thus the observed choice variable y j is defined as:

• MARKOV RESPONSE MODEL:


In the Markov response models the transition probabilities, Pijt with i, j = 1, ... , n, t = 1, ... , T are related to decision variables. It is assumed that transition probability matrix is non-
stationary & can be written as:

Considering 2 brands, 1 & 2, the brand loyalty for brand 1 is defined as:

Similarly, brand switching to brand 1 is defined as:

Here, the values of the transition probabilities are not restricted to the range zero to one.
4. Integrated models of incidence, timing and choice:
The basic setup of this approach is as follows:
• Assume a Poisson distribution for purchase frequency, y: P(y I λ)
• Assume a gamma heterogeneity distribution for the purchase rate: G(λ)
• Obtain the unconditional distribution of y by integrating out the heterogeneity distribution. This leads to a NBD : NBD(y) = f P(y I λ )G(λ)d λ.
• Assume a multinomial distribution for choice, x: M(x I p).
• Assume a Dirichlet heterogeneity distribution for the choice probabilities: D(p).
• 6. Obtain the unconditional distribution of x by integrating out the heterogeneity distribution. This leads to a Dirichlet-Multinomial (DM) distribution: DM(x) = ʃ M(x I p)D(p)dp.
• In the final step, the joint distribution of purchase frequency and choice is obtained, assuming independence: NBDM(y, x) = NBD(y) · DM(x).

You might also like