Market Definition Using Consumer Characteristics and Cluster Analysis

• DOI: http://doi.org/10.1111/saje.12221
• Author: Willem Boshoff,Rossouw Jaarsveld
• Date: 01 September 2019
• Published date: 01 September 2019

© 2019 Economic Society of Sout h Africa.

South African Journal of Economics Vol. 87:3 September 2019

doi : 10.1111/ saje .122 21

302

MARKET DEFINITION USING CONSUMER

CHAR ACTERISTICS AND CLUSTER ANALYSIS

WILLE M BOSHOFF* AND ROSSOU W VAN JAARS VELD†

Abstract

Competition authorities commence merger investigations with a definition of the relevant market,

i.e. the collection of products or regions that compete with the firm(s) under investigation. Market

definition relies on evidence of cross-price elasticity. When the econometric estimation of cross-

price elasticities is not possible, analysts have to rely on qualitative assessment and a set of

quantitative tools relying only on a subset of information or less precise quantitative evidence.

Under these conditions, a statistical analysis of the similarity of consumer characteristics of

different products can offer a useful addition to this toolkit. Cluster analysis can be used to identify

meaningful groups of substitutes (and hence markets) on the basis of homogeneity of products’

consumer characteristics. Cluster analysis is useful not only in drawing market boundaries, but

also in ranking the competitors within the boundaries. In a recent radio merger case in South

Africa, competition authorities and counter parties struggled to obtain systematic evidence for

market definition and ultimately relied on mostly qualitative impressions based on a range of

evidence from industry experts and company strategy documents. The substitutes suggested by the

cluster analysis matches the closest substitutes suggested by the Competition Tribunal, although

the results do not support the Tribunal’s list of weaker substitutes.

JEL Classification: L4, L1, C1, D1

Keywords: Market de finition, cluster analysis

1. INTRODUCTION

The definition of relevant product and geographic markets remains a key first step in

merger analysis in most competition jurisdictions, including South Africa. In fact, the

current South African policy debate on the level of market concentration in South

African product market s are predicated on clearly delineated markets. Market defi nition

in competition cases relies on empirical evidence of substitution patterns. Cross-price

elasticity estimates, measuring the sensitivity of product prices to price changes in other

products, are often preferred forms of evidence. Unfortunately, the industrial organisa-

tion (IO) models from which elasticity estimates are derived have high data and time

* Corresponding author: A ssociate Professor, Centre for Competition Law and Economics,

Department of Economics, Stel lenbosch University, Private Bag X1, Matieland 7602, South

Africa. E-ma il: wimpie2@sun.ac.za

† Department of Economics, C entre for Competition Law and Economics, Stellenbosch

Uni ve rsi ty.

South African Journal

of Economics

303South African Journal of Economics Vol. 87:3 September 2019

© 2019 Economic Society of Sout h Africa.

requirements, often preventing their us e1. Practitioners therefore often rely on qualitative

assessment as well as a set of less sophisticated tools when defining markets, which in-

clude price co-movement tests (see Boshoff, 2007), analyses of product flows (Elzinga

and Hogarty, 1998) and diversion ratios (Katz and Shapiro, 2002). These tools each rely

only on a subset of information, but combining different tools significantly expands the

available information set and broadens the basis for inferences. It is therefore important

to continue exploring alternative approaches to studying economic data for the purposes

of defining markets, especially because of the implications of market share for competi-

tion cases.

The paper considers an addition to the existing set of market definition tools, namely

a cluster analysis of consumer characteristics. Cluster analysis form part of the statistical

learning literature, and these techniques are generally underappreciated in competition

economics. In fact, until recently, statistical lea rning techniques have received far less at-

tention from economists, but this is changing, g iven the chal lenge to identify patterns i n

large datasets ( Varian, 2014) and as the techniques have progressed to deal with identif i-

cation problems (Einav and Levin, 2014). Consequently, this paper is the f irst to demon-

strate the utility of a statistical learning technique in def ining markets in a competition

case. The focus of the cluster analysis is on consumer characteristics. As argued below,

this is a novel focus, as consumer cha racteristics directly determine substitutability.

Cluster analysis of consumer (or, indeed, product) characteristics can be a use ful com-

plementary tool under conditions where data is limited and hence where traditional

econometrics-based approaches are infeasible. This tool complements the insights from

price co-movement tests or other quantitative tools, as well as insights from qualitative

assessments, by shedding light on the heterogeneity of consumer characteristics.

Furthermore, even under conditions where econometrics-based approaches are feasible

(i.e. where elasticity estimates can be obtained), such approaches also face challenges2

and cluster analysis and other limited information tools can offer complementary

insights.

To develop these arguments, this paper first considers the extent to which a study of

consumer characteristic s is consistent with the concept of a market in competition policy.

Thereafter, the paper shows how cluster analysis can be used to compare the consumer

characteristics of different products and to identify meaningful groups of substitutes

based on the homogeneity of their consumer characteristics. Cluster analysis enforces

consistency in the comparison of characteristics, while recent advances also allow infer-

ences robust to data errors. The relatively simple implementation and graphical output,

which assist interpretation, further recommend cluster analysis as an additional tool for

market definition. The paper demonstrates clus ter analysis as a tool for defining markets

by applying it to consumer characteristics data in a recent radio merger case in South

Africa. As discussed, the substitutes identified by the cluster analysis match the close

substitutes identified by the competition authority, although cluster analysis highlights

1 This practica l constraint partly explains why the direct assessment of merger effec ts, which

relies on cross-price elast icity estimates and doe s not require delineated markets, has not dis-

placed the traditiona l approach based on market shares.

2 See, for example, Bishop and Walker (1999), who discuss the diff iculty of translating market

definition concepts i n law into technical statements about elasticity e stimates.