Estimating Brand Equity from Aggregate Data

Abstract

Brands are now widely recognized as among a firm’s most valuable assets. Hence the assessment of the value or equity of brand assets has attracted a lot of academic and practitioner attention in recent years. Various brand equity (henceforth, BE) measurement schemes have been proposed and utilized. These methods rely on either on surveys or experimental approaches that require costly or complex data collection at the individual level, on valuation approaches requiring firm financial data, or on aggregate level product-market outcomes often utilizing fixed effects approaches that interpret brand intercepts as BE.

In this dissertation, I propose and estimate a model of BE measurement based on a random effects approach that uses readily available data sources. In contrast to current methods based on aggregate data that model BE as a fixed unknown constant for each brand, I posit that realized BE values in any market are outcomes of a complex process that is difficult to specify completely, but which can be modeled as draws from an underlying distribution. The model is data-driven in that it does not explicitly specify how BE is generated from the underlying complex system. Rather, it makes identifying assumptions on the shape and domain of the unknown distribution based on the profit maximizing actions of rational economic agents. I develop the general conceptual framework for the model and adapt this framework to three different applications in the course of three essays: (1) Brand Equity as a Revenue Multiplier, an application based on a national dataset; (2) The Brand Equity of Private Labels, an application based on a multi-retailer dataset; and (3) Spatial Patterns in Brand Equity, based on a multi-market dataset.

The random effects approach presents a simple, single-stage BE estimation and inference procedure that yields six distinct advantages: (i) an implicit baseline product for reference, (ii) easily interpretable results because BE is now a residual measure, (iii) useful downstream BE measures such as a BE dollar metric, which aids in assessing the financial value of brand assets, (iv) the use of a reduced form specification that does not make assumptions about the nature of the game and behavior of the players involved, (v) the incorporation of SKU based competitive effects modeled in product-attribute space, and all this (vi) in the presence of complex product structures and branding hierarchies. The tradeoff is that parametric assumptions on the underlying BE distribution need to be made.

The first essay introduces the BE measurement problem, develops the modeling approach, tests the model empirically on national data in the beer category, and seeks to validate the analysis results using independent, external sources. I find that the results bear face and external validity, and parsimoniously account for inter-product competition. The second essay is an application of the proposed model to BE estimation of Private Labels in addition to those of National Brands. The conceptual model is adapted to a multi-retailer setting in which the Private Labels are all Store Brands, and response heterogeneity across retailers is incorporated using a random coefficients specification. I find that there is considerable heterogeneity in BE across brands and retailers, that some but not all Private Labels have BE and thereby are ‘brands’, and that Private Label BE estimates present evidence favoring an ‘economic benefits and costs’ value proposition.

The third essay examines whether the demographic characteristics of U.S. metropolitan markets in conjunction with geospatial location information aid in understanding the level and distribution of BE across these markets. Unlike traditional spatial analyses that assume a multivariate normal error covariance structure, in this study I develop a distribution-free approach based on generalized least squares. I find heterogeneity in BE levels across brands and markets, and for many of the brands sampled I find substantial spatial correlation between the BEs of beer brands across geographic markets, and a distinct regional pattern in the BE distribution. I find that the demographic composition of markets are related to the BE for many brands and further, that the parameter estimates of the demographic BE antecedents are substantially improved by the incorporation of spatial information into the model.

Keywords: Doctoral dissertation