Harald Hruschka

Market definition and segmentation using fuzzy clustering methods

Abstract The main theme of this paper is the use of so-called fuzzy clustering methods for market definition and segmentation. Market boundaries are formed according to substitutive relations between brands as seen by homogeneous segments of potential or actual customers. The classification types partition, overlapping classification and fuzzy partition are explained. Segmentation of customers is conceived as a pre-processing step of market definition. Segments are aggregates of customers homogeneous with respect to the substitutive relations between brands. Several fuzzy clustering methods are presented. Evaluation procedures specialized in fuzzy classifications or suited for fuzzy and non-fuzzy classifications are dealt with. One of these procedures is an enlargement of the ADCLUS-model for fuzzy membership values of objects. An empirical pilot study demonstrates and compares the methods discussed. Diverse hard and fuzzy clustering methods are used to form homogeneous segments of customers and sets of competitive brands. External and internal validity of this classifications are determined.

Comparing performance of feedforward neural nets and K-means for cluster-based market segmentation

We compare the performance of a specifically designed feedforward artificial neural network with one layer of hidden units to the K-means clustering technique in solving the problem of cluster-based market segmentation. The data set analyzed consists of usages of brands (product category: household cleaners) in diAerent usage situations. The proposed feedforward neural network model results in a two segment solution that is confirmed by appropriate tests. On the other hand, the K-means algorithm fails in discovering any somewhat stronger cluster structure. Classification of respondents on the basis of external criteria is better for the neural network solution. We also demonstrate the managerial interpretability of the network results. ” 1999 Elsevier Science B.V. All rights reserved.

Determining market response functions by neural network modeling: A comparison to econometric techniques

Abstract The article starts by discussing basic concepts of artificial feedforward networks with hidden units. Backpropagation being the most wide-spread learning method to determine weights of connections between units is described. On the basis of data on a consumer brand, market response functions represented by artificial neural networks are compared to econometric models mainly on the basis of error measures. Economic interpretation of neural network results is given consideration. Finally advantages and limitations of neural networks are summarized.

Cross-category sales promotion effects

Abstract We introduce a multivariate binomial logit model measuring cross-category dependence and sales promotion effects of a retail assortment. This model requires as data both the market baskets of individual shoppers and the categories currently promoted in a retail outlet. A special section describes the stepwise procedure used to estimate parameters of this model. Its application is demonstrated analyzing 6147 purchases that were acquired in a medium-sized supermarket. We finally discuss the managerial relevance of this model for sales promotion decisions of retail firms.

Genetic Algorithms for product design: how well do they really work?

Recently, Balakrishnan and Jacob (1996) have proposed the use of Genetic Algorithms (GA) to solve the problem of identifying an optimal single new product using conjoint data. Here we extend and evaluate the GA approach with regard to the more general problem of product line design. We consider profit contribution as a firms economic criterion to evaluate product design decisions and illustrate how the genetic operators work to find the product line with maximum profit contribution. In a Monte Carlo simulation, we assess the performance of the GA methodology in comparison to Green and Kriegers (1985) greedy heuristic.

Computer-assisted travel counseling

Abstract Reliable and efficient access to relevant travel alternatives is a key ingredient for successful travel counseling. Computers can now assist travel counselors to provide better service to potential tourists. Lack of experience of a counselor may be offset by following the suggestions prompted on a computer monitor during the sales talk. Computer assistance in travel counseling may take two main forms: expert systems and retrieval approaches. This paper examines prototypes of these two approaches with a view to describing their main features as well as their respective strengths and weaknesses. The problems of implementation are discussed and recommendations for further action are proposed.

An empirical comparison of the validity of a neural net based multinomial logit choice model to alternative model specifications

Applications of choice models to brand purchase data as a rule specify a linear deterministic utility function. We estimate deterministic utility by means of a neural net able to approximate any continuous multivariate function and its derivatives to a desired level of precision. We compare this model to related alternatives both with linear and nonlinear utility functions. Alternatives with nonlinear utility functions are based on generalized additive modeling and Taylor series expansion, respectively. We analyze purchase data of the six largest brands in terms of market share for two product groups. Neural choice models outperform the alternative models studied w.r.t. posterior probabilities. They also attain the best crossvalidated log-likelihood values. These results demonstrate that the increase in complexity caused by the neural choice model is justified by higher validity. In the empirical study the neural choice models imply elasticities different from those obtained by linear utility multinomial logit models for several predictors. Neural choice models discover inversely S-shaped, saturation and interaction effects on utility.

Relevance of functional flexibility for heterogeneous sales response models: A comparison of parametric and semi-nonparametric models

So far studies estimating sales response functions on the basis of store-specific data either consider heterogeneity or functional flexibility. That is why in this contribution a model is developed possessing both these features. It is a multilayer perceptron with store-specific coefficients which is specified in a hierarchical Bayesian framework. An appropriate Markov Chain Monte Carlo estimation technique is introduced capable to satisfy theoretical constraints (e.g. sign constraints on elasticities). The empirical study refers to a data base consisting of weekly observations of sales and prices for nine leading brands of a packaged consumer good category. The data were acquired in 81 stores over a time span of at least 61 weeks. The multilayer perceptron is compared to a strict parametric multiplicative model and approaches the maximum value of posterior model probability. This indicates the benefits of using a flexible model even if heterogeneity is dealt with. Estimated sales curves and elasticities demonstrate that both models differ in their implications about price response. Bisher haben Untersuchungen zur Schatzung von Absatzreaktionsfunktionen auf Grundlage outletspezifischer Daten entweder Heterogenitat oder funktionale Flexibilitat berucksichtigt. Daher entwickelt der vorliegende Beitrag ein Modell, das beide Eigenschaften besitzt. Es handelt sich um ein Mehrschichtperzeptron mit outletspezifischen Koeffizienten, das mittels eines hierarchischen Bayesschen Ansatzes spezifiziert wird. Zur Schatzung dieses Modells wird eine geeignete Markov-Ketten-Monte-Carlo Technik eingefuhrt, die theoretisch begrundete Restriktionen einhalt (z.B. Vorzeichenrestriktionen von Elastizitaten). Die empirische Untersuchung bezieht sich auf einen Datensatz, der aus wochentlichen Beobachtungen von Absatzmengen und Preisen fur neun Marken einer Konsumguterkategorie besteht. Diese Daten wurden in 81 Outlets uber eine Zeitspanne von mindestens 61 Wochen erhoben. Das Mehrschichtperzeptron wird mit einem strikt parametrischen multiplikativen Modell verglichen und erreicht den Maximalwert der a-posteriori Modellwahrscheinlichkeit. Dieses Ergebnis zeigt die Vorteilhaftigkeit der Verwendung eines flexiblen Modells auch bei Berucksichtigung von Heterogenitat auf. Geschatzte Absatzkurven und Elastizitaten verdeutlichen, dass beide Modelle jeweils unterschiedliche Preiseffekte implizieren.

Market segmentation by maximum likelihood clustering using choice elasticities

We determine market segments by clustering households on the basis of their average choice elasticities across purchases and brands w.r.t. price, sales promotion and brand loyalty. The cluster analysis technique used is a maximum likelihood method which allows varying size and orientation and assumes constant volume. Elasticities originate from choice models with alternatively linear and nonlinear utility functions. Choice models are estimated on the basis of household scanner data. Segments are interpreted by means of multiple discriminant analysis and multinomial logit models whose predictors are elasticities of predictors and external variables (i.e. number of purchases, number of brands bought, income and household size), respectively.

Market share analysis using semi-parametric attraction models

Abstract Attraction models used to analyze the effects of marketing instruments on market share hitherto assume certain strict functional forms. We introduce semi-parametric models whose parametric components are equivalent to an exponential or multiplicative function. The nonparametric part is estimated on the basis of penalized generalized least squares taking into account smoothness of nonlinear functions. In the empirical study presented market share models with semi-parametric additive brand attractions attain better fits both according to an information criterion that penalizes a model for degrees of freedom (df) consumed and according to error measures determined by bootstrapping.