how useful are brand valuation methods for brand ......brand valuation philosophies and prior tests...
TRANSCRIPT
1
How useful are Brand Valuation Methods for Brand Management? A Validation Study
Marc Fischer1)
Tobias Hornig2)
August 2014
1 Professor of Marketing and Market Research, University of Cologne, and Associate Professor
of Marketing at UTS Business School, Sydney. Contact: University of Cologne, The Faculty of Management, Economics, and Social Sciences, Chair for Marketing and Market Research, Albertus-Magnus-Platz, 50923 Cologne, Germany, Phone: +49 (221) 470-8675, Fax: +49 (221) 470-8677, e-mail: [email protected]
2 Siemens AG, e-mail: [email protected] The authors gratefully acknowledge the support of Corebrand and Harris Interactive for the usage of their data in this study. They also thank Jan-Benedict Steenkamp and participants of seminars at the University of Groningen, UTS Business School, and Marketing Science Conference 2014 in Atlanta for valuable comments on this paper.
This study is fully independent. The authors declare that they have no professional relationship with the providers of investigated brand valuation models. They did not receive financial support from the providers.
2
ABSTRACT
Many methods for measuring the financial value of a brand have been suggested in the past.
Their results, however, differ to a great extent, which raises serious concerns about the validity
of these methods. In this large-scale study, we adapt the established construct validation
methodology in social sciences to assess the validity of financial brand valuation methods.
Specifically, we test the validity of brand valuation methods with respect to their
reliability/stability, convergent validity, discriminant validity, nomological validity, and
predictive validity.
We apply the test procedure to nine prominent valuation methods that cover the basic
philosophies in brand valuation, which are cost-based, market-based, and income/DCF-based
approaches. The data cover a period of 22 years from 1990 to 2012. The sample includes 36,992
financial values of 4,879 brands that originate from 87 countries and represent more than 70
industries.
Generally speaking, brand valuation models produce reliable and stable results. Only few models
show convergent validity across different valuation approaches. Nomological validity cannot be
established for any method and a few models demonstrate predictive validity in stock-return
forecasts. Considering all validity test criteria together, it appears that the market-based methods
perform best in meeting the validity requirements.
Keywords: Brand valuation, brand equity, scale validation, time-series tests
3
INTRODUCTION
The financial value of a brand is of high interest to many stakeholders and decision
makers. As a result, several methods for measuring the brand value have been suggested in the
past. Their results, however, differ to a great extent, which raises serious concerns about the
validity of these methods. Marketing’s seat at the table in the boardroom is severely hampered if
the value of the asset that attracts most of marketing’s expenditures cannot be established in a
credible way. Hence, it is crucial that marketers agree on a generally accepted valuation method
that produces valid results and makes marketing accountable. This study is a first step into that
direction.
Unfortunately, the true value of a brand cannot be observed. Without knowing the true
value it is hard to assess how close a model’s estimate comes to the true value. While
unpromising at first glance, this dilemma is not new but well known in the social sciences that
frequently need to measure unobserved constructs. In fact, this discipline has developed a
rigorous framework of test procedures and statistics for evaluating the validity of a construct
(e.g., Churchill 1979; Peter 1979, 1981). Construct validity refers to the correspondence between
a construct at the conceptual level and the operational procedure to measure that construct.
Assessing construct validity usually involves several tests including tests on reliability,
convergent validity, discriminant validity, nomological validity, and predictive validity of the
measurement model. Construct validity cannot be assessed directly but is inferred from the
measure’s variance and covariance with other measures.
A drawback of this validation approach is that it is ignorant about the measurement level,
i.e. the absolute metric. The absolute brand value is particularly relevant to applications for
accounting and transaction purposes, such as mergers and acquisitions, licensing, or
4
securitization. The other major group of applications includes valuation for brand management
purposes, such as brand strategy, brand portfolio management, or marketing budget allocation.
Here, the absolute brand value plays a minor role. It is rather essential that the metric reflects the
effects of brand decisions in a consistent and valid way. According to a recent survey among
senior executives, the reasons for brand valuation are equally distributed across the two groups
of applications (PWC 2012). Hence, even though testing for construct validity does not solve the
issue of diverging absolute brand valuation results, it helps establish confidence in using a
method for brand management purposes that are by no means less important in practice. This is
the focus of our study.
We assess nine brand valuation methods that cover the three basic philosophies in brand
valuation, i.e. cost-based, market-based, and income/DCF-based approaches. Specifically, we
compare brand valuation results produced by the following models: historical cost of creation
and advertising stock (cost-based approaches), Simon and Sullivan (1993) and CoreBrand
(market-based approaches), and Interbrand, Millward Brown, Semion, Brand Finance, and
Ailawadi, Lehmann, and Neslin (2003) (income/DCF-based approaches). Our data covers a
period of 22 years ranging from the start of brand valuation in 1990 to 2012. It includes 36,992
financial values of 4,879 brands that originate from 87 countries and represent more than 70
industries. The scope of our data thus provides the basis for truly generalizable results on the
validity of brand valuation methods.
This study offers several contributions. It is the first attempt to bring light into the jungle
of brand valuation methods. We achieve this by testing the construct validity of established
valuation methods, which is a prerequisite for using brand values to assess and monitor the
5
outcomes of important brand management decisions. We show which methods and which
valuation philosophies satisfy the validity criteria.
Second, since the test procedures for validating social constructs were mainly developed
for cross-sectional data we need to make appropriate adaptations to our data that are generated
across both brands and time. Specifically, we extend the test framework by established time-
series tests that account for both stationary and non-stationary time-series. We also introduce
Granger-causality tests as a powerful means to test for nomological validity, which is not
applicable with purely cross-sectional data.
Finally, we provide important implications from our findings for the use of brand
valuation methods in management practice. In addition, we discuss consequences for using brand
values in research on brand management. Our conclusions also stimulate the development of an
ideal, generally accepted measurement approach.
The paper is structured as follows: In the next section we briefly review the literature on
brand valuation philosophies and prior tests of methods. We then describe our test methodology.
In the following sections, we introduce our data and present the empirical test results. We
continue with discussing the managerial and research implications. We conclude the paper with
its limitations and suggestions for further research.
LITERATURE REVIEW
We define brand value as the incremental discounted future cash flows accruing from a
branded product compared with an identical but unbranded product (Simon and Sullivan 1993).
Salinas (2009) provides a complete summary of 39 proprietary financial brand valuation
methods. We do not repeat the detailed discussion here but refer to this excellent review. The
6
main purpose of this section is to delineate the ideas behind the three major brand valuation
philosophies that emerge from the literature (e.g., ISO Standard 2010; Salinas 2009). These
philosophies are: cost-based, market-based, and income/DCF-based approaches.
Approaches of Brand Valuation
Cost-based approach. This approach determines the brand value in terms of prior
investments into developing and building the brand. A straightforward method is to measure the
historical cost of creation by accumulating all expenditures on the brand until the period of
valuation (Barwise et al. 1989). A more realistic approach is to follow the idea of an advertising
stock where brand expenditures build up the stock that decays at a constant rate over time
(Nerlove and Arrow 1962). While cost-based measures are attractive due to the objective and
easy collection of data they are heavily criticized for their theoretical weaknesses. Prior brand
expenditures measure past efforts for brand building but not future outcomes, such as excess
profits, that accrue from the brand (Salinas 2009; 60f).
Market-based approach. Following the efficient market hypothesis (Fama and French
2006), the market price for an asset represents the fair value for that asset, provided that all
investors have the same amount of information available and engage in many transactions.
Unfortunately, there does not exist a liquid market for brand transactions, i.e. transaction prices
are not readily available. But given that a brand is part of the firm value, its value is implicitly
included in the market value of a company. The idea of a market-based approach, such as the
CoreBrand model or the model by Simon and Sullivan (1993), is to separate the brand value
from the observed market capitalization of the firm. The major advantage of this approach is its
consistency with capital asset pricing theory and fair valuation principles. But there is no general
7
agreement about the right approach to isolate the brand value from the company’s market value
and applications are limited to publicly listed firms (Fischer 2007).
Income/DCF-based approach. The income/DCF-based approach attempts to project the
(future) stream of profits or cash flows, respectively, due to the brand (e.g., Salinas 2009; Fischer
2007). In theory, the present value of this stream equals the market price of the brand. There are
multiple ways to arrive at the estimate for the present value of brand-generated cash flows, such
as the relief-from-royalty method (Brand Finance), income split method (Interbrand, Millward
Brown), or incremental income method (Semion). We also consider the revenue premium model
by Ailawadi, Lehmann, and Neslin (2003) as a current-period income-based method as it can
easily be used to determine the incremental cash flow that is attributable to the brand. The major
advantage of these models is that they are consistent with the basic principles of fair asset
valuation that is inherent to DCF models. Major concerns exist about the subjectivity and
uncertainty that is often associated with the forecast and separation of expected brand cash flows
(Ailawadi, Lehmann, and Neslin 2003).
Prior Literature on Validating Brand Valuation Models
Little research exists that offers orientation in the validation of brand valuation models. In
fact, many (commercial) models are presented in a way that the model and its results stand alone
and convince by its face validity. Additional validity tests are not performed. Rigorous tests of
model validity are also very limited in the academic literature. Simon and Sullivan (1993) show
in a descriptive way how the estimated brand value for Coke and Pepsi changes in response to
important events in the soft drink market. Fischer (2007) compares model-implied market
capitalization with actual values for two automobile firms. Only Ailawadi, Lehmann, and Neslin
(2003) adopt a more rigorous approach and test the stability of their measure and its correlation
8
with several other measures in two different data sets. To the best of our knowledge, no study
has attempted to develop and apply a comprehensive testing framework to test several
established brand valuation models at once.
METHODOLOGICAL FRAMEWORK FOR INTERNAL AND EXTERNAL VALIDATION
Our suggested framework for testing the validity of brand valuation methods follows the
paradigm for measuring marketing constructs (Churchill 1979; Peter 1979, 1981). According to
this paradigm, construct validity is established by internal and external validation of the measure.
Internal validation refers to trait validity and is achieved by demonstrating reliability, convergent
validity, and discriminant validity for the measure. Internal validation is a necessary condition
for the validity of a measure, but it is not sufficient. The brand value method must also enable
observable predictions derived from theoretical propositions (Peter 1979). Hence, we also
require the method to demonstrate nomological and predictive validity (external validation).
In the following, we describe how we approach the measurement of the various types of
validity. We use established procedures of analyzing correlations and covariances among
constructs (e.g., DeVellis 2012; Netemeyer, Bearden, and Sharma 2003). When necessary, we
also introduce adaptations and extensions to the framework that result from the time-series
nature of brand value data. Table 1 summarizes the methodological framework with its test
statistics and thresholds, which we follow in this study.
== Table 1 about here ==
Reliability/Stability over Time
Reliability measures the degree to which a measure is free from random error. It is
assessed in terms of (1) test-retest correlation and (2) internal consistency (Cronbach’s Alpha)
9
(Netemeyer, Bearden, and Sharma 2003, 10). We consider reliability as an integral part of a
construct’s validity because any subsequent correlation-based validity test depends on the
measures’ reliability. In fact, the correlation between two measures cannot be higher than the
square root of the smaller reliability (Peter 1981).
Test statistics. The test-retest correlation is the key measure for our reliability test.
Internal consistency tests do not apply to brand valuation methods as they refer to multi-item
scales. For each brand valuation method, we calculate the correlation of the measure in t with its
lagged value in t-1, where t refers to the year of valuation. Strictly speaking, the autocorrelation
coefficient does not only reflect the reliability of the method since the underlying brand value
may change from one year to the next year. However, stickiness or stability over time,
respectively, is inherent to the concept of brand equity. Thus, low autocorrelation suggests poor
reliability of the method (Ailawadi, Lehmann, and Neslin 2003). In addition, we make use of the
combined time-series and cross-sectional nature of our data. Specifically, we decompose the total
variance in a valuation method’s results into its cross-sectional variance and time variance.
Following the same argument about stickiness, time variance should be significantly smaller
relative to cross-sectional variance.
Threshold values. In order to attest a method a high reliability we need to define the
desired value for the test statistic. Unfortunately, there is no theory from which we can deduce
the appropriate threshold level (DeVellis 2012, 67; Churchill 1979). We will rely on thresholds
used in the literature and consider the amount of shared variance that is implicit to the
correlation. For reliability (correlation) coefficients, the literature considers a value of .90 as the
minimum and a value of .95 as desirable standard for measures in applied settings where
important decisions are made (e.g., Ailawadi, Lehmann, and Neslin 2003; Churchill 1979;
10
DeVellis 2012, 109). We adopt the value .90 for the brand value’s correlation with its prior year
value. As a consequence, the shared variance over the two periods is larger than 80%. For
methods with two measurement points within the same year, we set a stricter threshold of .95.
When decomposing the observed variance of a method, we require that the (cross-sectional)
variance across brands be significantly larger than the (time) variance across time, which is
satisfied at a ratio of 3 (see table 1).
Convergent Validity
Convergent validity describes the extent to which independent brand valuation methods
yield similar results. Conceptually, this means that they should be highly correlated among each
other (Churchill 1979; Peter 1981).
Test statistics. The standard test statistic here is the correlation coefficient for two
independent valuation methods, which may be corrected for the known unreliability in the
measures (DeVellis, 66). It should be noted, however, that the correlation between two time
series requires that their means and variances remain constant over time (Aldrich 1995). Such
series are said to be stationary. If they are non-stationary, the traditional validation test is
meaningless and produces spurious correlation coefficients. For such cases, we suggest applying
Kao’s (1999) panel co-integration test. Two series are co-integrated if they converge towards a
common and stable equilibrium in the long run (Engle and Granger 1987). By accounting for the
trend in the time-series, this test reveals whether the two measures are associated with each other
or not. Figure 1 summarizes the steps we suggest for analyzing the association between measures
with time-series (panel) data.
== Figure 1 about here ==
11
Threshold values. Again, we need to define desired values for the correlation statistics.
The psychometric literature does not suggest a minimum level of correlation but only requires
that the correlation among alternative measures be higher than the correlation with distinct
measures (discriminant validity) (e.g., Peter 1981; DeVellis 2012, 67ff). This rule leaves room
for different threshold values, which is not acceptable for our application. Consistent with prior
research in branding (e.g., Fischer, Völckner, and Sattler 2010), we require an average minimum
correlation of .50 that corresponds to an average shared variance of at least 25% to demonstrate
convergent validity. Unlike most scale validation studies, our study includes 9 alternative
measures leading to 36 potential pairwise correlations. Before obtaining the average correlation,
we require that 75% of all correlations for a measure with other measures must be manifest and
statistically significant (p < .05) (see table 1).
For non-stationary time-series, we require that ADF < tADF. ADF denotes the augmented
Dickey-Fuller statistic that is obtained for a panel, i.e. we estimate brand-specific constants, and
tADF is the respective t-statistic (Engle and Granger 1987; Kao 1999).1
Discriminant Validity
Discriminant validity refers the extent to which brand valuation methods correlate with
constructs that are designed to measure distinct concepts. The measures should be uncorrelated if
they represent different constructs (Churchill 1979; Peter 1981).
Test statistics. The standard test statistic is either the correlation coefficient between the
brand value and the divergent measure or the panel co-integration test if series are non-stationary
(see figure 1 again). We assess discriminant validity with respect to two important constructs:
customer satisfaction and corporate reputation. Customer satisfaction measures the degree to
1 Note that we do not compute correlation coefficients for differenced, non-stationary time-series. Although
differencing translates the series into stationary series, it also removes any cross-sectional variance. As a result, correlations tend to be much lower and are no longer comparable with correlations in levels.
12
which purchase and consumption experiences confirm purchase expectations (Fornell et al.
1996). It focuses on the relationship of the customer with the company. In contrast, brand value
focuses on the product. Corporate reputation reflects the overall credibility and respect that an
organization has among a broad set of constituents (e.g., employees, investors, regulators,
customers). Reputation arises from several dimensions including financial soundness,
innovation, product/services quality, and quality of management (Fombrun and Shanley 1990).
Hence, the concept is broader than that of the brand.
Threshold values. Correlations between brand values and distinct measures should be
lower than the correlation among alternate brand values (e.g., Peter 1981; DeVellis 2012, 67ff).
We consider a maximum correlation of .50 (see our threshold for convergent validity before) as
too high since this would allow for a shared variance of up to 25%. We rather require the
correlation not to exceed .30, i.e. the shared variance between the divergent constructs should be
less than 10%. For non-stationary time-series, we require that the co-integration test be rejected
(ADF < tADF) (see table 1).
Nomological Validity
Nomological validity describes the extent to which brand valuation methods are
associated with measures of other constructs that is consistent with theory (Netemeyer, Bearden,
and Sharma 2003, 86). Recall that we investigate the use of brand valuation methods for brand
management purposes such as brand portfolio strategies, brand investment decisions, etc. In
these applications, brand value is a key intermediate marketing variable that fits into a logical
chain of brand value creation (Keller and Lehmann 2003). From the brand value chain, we derive
the nomological network of antecedents and consequences as shown in Figure 2. Previous
advertising and other brand expenditures that are included in selling, general, and administrative
13
(SG&A) expenditures drive brand value. We also consider customer-based brand equity
measured by Harris EquiTrend scores as an antecedent. The strength of the brand in the heart and
minds of customers is a prerequisite for future returns due to the brand (Fischer 2007). If the
brand value increases we suppose this to have positive economic consequences for the firm in
the future. Specifically, we expect that sales, profit and finally the market valuation of the firm
improve (see figure 2 again).
== Figure 2 about here ==
Test statistics. The standard test statistics is again the correlation coefficient between the
brand value and the antecedent and consequence measures. If time-series are non-stationary we
need to test for co-integration. However, correlation does not imply causation. Since the
nomological framework implies a causal ordering of variables it can be tested more rigorously by
using the concept of Granger Causality. The idea here is to use the temporal ordering of events to
distinguish between leading (antecedents) and lagging (consequences) variables on empirical
grounds (Granger 1969). We employ the established regression-based methods (see Appendix A
for details) to test the validity of the Granger-causal ordering of figure 2.
Threshold values. For the correlation analysis, we require that 75% of pairwise
correlations between the brand value measure and both its antecedent and consequence measures
must be statistically significant (p < .05). In addition, the average correlation should be .40 or
greater, i.e. the shared variance is at least 15% (e.g., Fischer, Völckner, and Sattler 2010; Finn
and Kayande 2005). For non-stationary time-series, we require that the co-integration test be not
rejected (ADF > tADF) (see table 1).
In our Granger-causality tests, we allow for up to 4 lags. For each antecedent, we test
whether it Granger-causes the respective brand value measure. The same applies to each
14
consequence, which we assume to be Granger-caused by the brand value measure. Here again,
we require that 75% of these tests be not rejected to establish confidence in the causal
relationships. More importantly, we also test the reverse of all postulated variable relationships.
In fact, if the reverse relationship is supported by the Granger-causality test it casts doubt on the
presumed causal ordering of variables. We require that the number of supported Granger-causal
relationships must be significantly greater than the number of unexpected reverse relationships
across all antecedents and consequences. The ratio should be at least 3 to 1 (see table 1).
Predictive Validity
Predictive validity demonstrates the ability of a measure to predict an external criterion
that is independent of the valuation result but related to it in a practical and meaningful way.
Unlike nomological validity, predictive validity does not require that the reason for the empirical
association is understood (DeVellis 2012, 61). It is rather important that the external criterion can
be considered a “gold standard” and has relevance for actual decision-making.
Real brand transaction prices obviously represent a “gold standard” that reflects
managerial decisions. Unfortunately, we could only obtain a very limited number of transactions
prices. Sample size is not sufficient to correlate M&A values with individual method’s results.
But we can obtain insights from a pooled correlation across methods.
The stock price of a firm is another useful external criterion for prediction. Consistent
with the call for marketing accountability, changes in brand values as a result of brand
management decisions should have an effect on firm value. Prior research (Mizik and Jacobson
2008) has shown that a perceptional brand measure, which is an antecedent in our nomological
framework (figure 2), indeed drives stock prices. Hence, testing whether brand value estimates
15
impact stock prices qualifies as a powerful way to demonstrate the predictive validity of a
valuation method (Srinivasan and Hanssens 2009).
Model and test statistic. We implement the stock return-modeling framework of Mizik
and Jacobson (2008) that builds on the efficient market hypothesis (Fama and French 2006) and
test the predictive ability of brand value measures. The model controls for risk factors according
to Carhart’s (1997) four-factor model, period fixed effects, and unanticipated changes in
accounting performance. Based on the residuals of a fixed-effect, first-order autoregressive
model, we add unanticipated changes in brand value to the model because investors only respond
to new information. For further modeling details, we refer to Mizik and Jacobson (2008) and
Appendix B. We estimate the immediate impact on stock return. Since it may take some time
until investors fully understand and appreciate the financial implications a change in brand value,
we also estimate the effect for 1 month, 5 months, and 11 months after the announcement.
Predictive validity is shown if the coefficient estimate associated with the brand value measure is
positive and statistically significant (p < .05) (see table 1 again).
DATA AND MEASURES
Data Sources
We test nine brand valuation models: historical cost of creation (Barwise et al. 1989),
advertising stock (Nerlove and Arrow 1962), Simon and Sullivan (1993), CoreBrand (2013),
Interbrand (2012), Millward Brown (2013), Semion (2013), Brand Finance (2013), and
Ailawadi, Lehmann, and Neslin (2003). We do not repeat details on the methodology of the
models but refer to the respective original source. Salinas (2009) also covers the models in her
review. CoreBrand, Interbrand, Millward Brown, Semion, and Brand Finance are commercial
16
vendors that developed their own proprietary models. Semion is based in Germany, the country
with the second most model contributions. All commercial vendors publish brand valuation
results for international brands in ranking lists that are either cross-sectoral or focus on a specific
industry. We collected these information from their websites, press releases, and outlets such as
Financial World or Business Week, respectively. We also gathered data on revised brand
valuation results whenever possible. We approached each vendor to obtain brand values that
were not published in the ranking list. Only CoreBrand was willing to share their proprietary data
for the purpose of our study.
For the other four (academic) models, we collected published brand values and
reproduced brand values according to the methodology described by the authors. For this
purpose we collected financial, marketing, and capital market data from various sources
including COMPUSTAT, Thomson Banker One, SEC filings, and the Center for Research in
Security Prices (CRSP).
Finally, for the tests on discriminant and nomological validity, we used data provided by
the American Customer Satisfaction Index (ACSI), Fortune’s corporate Reputation Index, and
Harris EquiTrend. The data collection spans a period of 22 years from 1990 to 2012. Depending
on the model, brand values are available from the start in 1990 or at a later date. A detailed
summary of data sources and periods for each model is available in Appendix C.
Measures
We need to report on a few decisions and assumptions we had to make when compiling
our database.
Advertising stock model. We follow the Nerlove and Arrow (1962) approach to measure
brand value in terms of an advertising stock. The advertising stock in a given year t is composed
17
of the advertising expenditures in t and the stock value in t-1 that is carried over at a constant rate
measured by the carryover coefficient. Since we do not observe brand-specific carryover
coefficients, we use a value of .50 that has been established as an empirical generalization
(Sethuraman, Tellis, and Briesch 2011). Following Fischer et al. (2011), we estimate the stock
value in the initial year of our observation period by dividing brand expenditures in that year by
the decay rate of .50 (= 1 – carryover).
Ailawadi, Lehman, and Neslin (2003) model. This model suggests computing brand value
by the revenue premium a brand has over a benchmark product with no or low brand equity. This
could be a private label brand in a consumer packaged goods category or simply the lowest-share
brand in a market. We use the lowest-share brand as benchmark that can be consistently applied
across industries (p. 15). To obtain a profit-based measure, we follow the authors’ suggestion (p.
6) and apply a profit margin that reflects average profitability in the market by period. However,
we do not project revenues and costs into the future to obtain a future-oriented measure. Thus,
though valuation results from this model are income-based they refer only to the current period.
Customer satisfaction and corporate reputation. Fornell et al. (1996) provide details on
the calculation of the satisfaction rating scale (0-100) used for the ACSI. Ratings are collected on
an annual basis for more than 230 companies. Fortune’s annual corporate reputation index is an
overall reputation score ranging from 0-10. It builds on ratings from eight dimensions (Fombrun
and Shanley 1990).
Customer-based brand equity. Harris Equitrend provides the data for our customer-based
brand equity measure. The Equitrend measure is a brand rating scale from 0-100 that captures
US customer perceptions on a representative basis for more than 1,500 brands per year.
18
M&A brand values and stock returns. We reviewed SEC filings of publicly listed firms to
obtain brand values from real market transactions. Since the revision of international accounting
standards in 2001, firms are required to recognize the brand value that is part of the purchase
price in an acquisition (for details, see Fischer 2007; Salinas 2009). We estimate expected stock
returns according to the four-factor model (Carhart 1997), where risk factors are provided by the
CRSP database. The dependent variable in our stock return model is the geometric 12-month
mean of abnormal stock returns since our temporal interval is the year.
Comparability and Time Alignment
Two independent experts evaluated each brand to ensure that brand values and other
information are comparable and appropriately aligned on the time scale. Both evaluators had to
agree that the brand satisfies the following rules.
Brand values always reflect the value of a specific product or service, respectively, not a
portfolio. When company data had to be used to obtain brand values (e.g., for the Simon and
Sullivan model), we verified that the valued brand is a corporate brand or clearly dominates the
company’s business. For example, BMW satisfies this requirement as the brand accounts for
more than 90% of firm sales. In contrast, Colgate-Palmolive, Procter and Gamble, or Unilever do
not satisfy the rule since we cannot identify one major brand. Note, however, that their product
brands are still be part of our database if they are produced by other valuation models such as
Interbrand or Millward Brown.
A brand’s value always refers to the global value of the brand in a given year. Foreign
currencies are converted into US dollars at the average annual exchange rate, i.e., all figures are
in US dollar.
19
Finally, we carefully align brand values and other variables, so that they are comparable
in terms of the period of measurement. Commercial vendors announce brand values at different
dates during the year. We treat the year of announcement as the valuation year by assuming that
it incorporates the most recent information available to the valuator. We use financial and other
information from that year to produce brand values for the other methods. For nomological
validity tests (see figure 2 again), we consider financial and Equitrend data of the preceding and
following year to generate variables for antecedents and consequences, respectively. For the
stock return response model, we align abnormal returns and brand value data at the monthly
level. For example, if Interbrand announces the new brand value in August of year t we calculate
the 12-month abnormal return for the parent company that covers the preceding 12 months
including August of year t. The abnormal returns within 1 month after the announcement
includes only the returns of September, the returns after 5 months only the returns for September
thru January, etc.
Descriptive Statistics
Our data collection effort resulted into a healthy sample size of 36,992 brand values that
span a period of 22 years from 1990 to 2012. This sample includes 3,879 brands that originate
from 87 countries and cover virtually all industries. Table 2 summarizes our database with
respect to periods, brands and observations.
== Table 2 about here ==
Since valuation models were introduced at different points in time, time-series for brand
values have different lengths and starting years across the methods. Due to time and other data
limitations the number of brands by method varies from 78 (Semion) to 2,752 (Brand Finance).
On average, each method contributes 727 brands to our analysis. 4,110 brand value observations
20
are, on average, available across methods. The mean length of individual brand time-series varies
from 2.16 years (Brand Finance) to 13.22 years (historical cost and advertising stock models).
The average brand time-series length amounts to 5.66 years. Thus, our sample exhibits a typical
panel structure. Actual brand values (not shown) range from zero to more than US$150 bn. The
mean value is US$ 2.54 bn. The standard deviation amounts to US$ 6.04 bn. We observe a
highly skewed and leptokurtic distribution of values for each method. This suggests that there are
only a few brands carrying very large values. The vast majority of brands, however, have
relatively low values compared to the top brands.
An important characteristic of our database is that it provides enough observations for
conducting the various validity tests. As a rule, we only compute a correlation or regression
coefficient if more than 50 observations are available. As table 2 shows, the number of joint
observations between two methods averages 706 observations. The number of joint observations
only falls below the minimum of 50 for Semion with respect to three other methods. Considering
the variables for testing discriminant and nomological validity, all methods reach the minimum
level. The sample sizes average several hundred observations. Appendix D provides detailed
information on joint observations by method and variable.
Panel Unit Root Tests
Consistent with the nature of a data, we apply the panel unit root test according to Levin,
Lin, and Chu (2002) to test for stationarity of variables. Unit root is rejected for most brand
valuation methods and validation variables. Brand values by the historical cost model and the
Simon-Sullivan model as well as SG&A expenditures and firm revenues do not pass the test.
Hence, we need to make appropriate changes to our tests following the procedure of figure 1.
21
EMPIRICAL RESULTS
We present the aggregated results of our validity tests in Tables 3 to 8. More detailed test results
are provided in Tables E.1 to E.4 in Appendix E. Recall that all tests are based on at least 50
(joint) observations. The vast majority of tests exceed this minimum level by far.
Reliability and Stability Results
Table 3 shows that test-retest correlations are very high for the nine valuation methods.
All methods satisfy the correlation threshold of .90 (between-year correlation) or .95 (within-
year correlation), respectively. The picture does not change if we consider the decomposition of
the variance of brand values into its cross-sectional and time variance. Cross-sectional variance
is more than three-times larger than time variance, except for the historical cost model. We
conclude that the reliability and stability is very strong for most methods.
== Table 3 about here ==
Convergent Validity Results
Table 4 summarizes the pairwise correlations of brand values produced by different
methods. Note that only 7 correlation coefficients are available for the cost-based methods and
CoreBrand due to insufficient joint observations with the Semion model. Most methods correlate
positively and significantly with each other. If time-series are non-stationary, the panel co-
integration test suggests that model results converge towards a common equilibrium in the long
run, i.e. they are associated with each other. All methods pass the threshold for the proportion of
significant correlations, except for the Millward Brown model. This model correlates only with
50% of the other models. In fact, Millward Brown only correlates with other (commercial)
models of the same category, i.e. future-oriented income/DCF-based approaches such as
Interbrand.
22
== Table 4 about here ==
Considering the average significant correlation among methods, it turns out that four
models do not pass the threshold of .50. These models are historical cost of creation, Interbrand,
Semion, and Ailawadi, Lehmann, and Neslin. As we show later in our robustness checks, it does
not matter whether we include Millward Brown and/or Semion or not into this analysis. If we
consider only correlations among models within their subcategory of valuation approach all
methods except for Semion pass the threshold, i.e. demonstrate sufficient convergent validity.
Discriminant Validity Results
Our discriminant validity tests in Table 5 suggest that brand valuations of all models are
distinct from the concept of customer satisfaction (correlation < .30). However, correlations of
CoreBrand, Millward Brown, and Brand Finance with Fortune’s Corporate Reputation Index are
slightly larger than our threshold of .30. These correlations are still smaller than their average
significant correlations in the convergent validity test. But concerns about these models’
differentiation from overall firm reputation remain.
== Table 5 about here ==
Nomological Validity Results
Correlations. The correlation results inform about the strength of supposed relations
between brand values and their antecedents and consequences (see Figure 2 again). For a strong
association, we require that the majority of correlations is significant and averages .40 or higher.
Non-stationary time-series should be co-integrated to support the association assumption. With
respect to consequences, Table 6 shows that these requirements are satisfied by all methods
except for Millward Brown. For antecedents, we find that Interbrand, Millward Brown, and
Ailawadi, Lehmann, and Neslin do not pass the minimum correlation of .40. Hence, full
23
nomological validity in terms of the strength of associations cannot be established for these
models.
== Table 6 about here ==
Granger-causality tests. Table 7 summarizes our findings from Granger-causality tests.
With these tests, we check for the plausibility of the assumed direction of causality according to
our nomological framework. When testing for Granger-causal relations between antecedents and
consequences variables we account for up to four lags. A relation between two variables is
counted in table 7 if we find support for at least one lag. As the results show, we find support for
most expected relations between brand values and its antecedents and consequences across the
various models. One exception is again Millward Brown’s model. Another observation,
however, is most striking. We also find that the reverse of the assumed relations is supported for
most variables and across models. Considering the threshold of 3 for the ratio of supported
versus reverse relations, we conclude that not a single method demonstrates sufficient
nomological validity.
== Table 7 about here ==
Predictive Validity Results
Correlation. We start our predictive validity tests with the correlation of brand value
estimates with real brand transaction prices, which we consider as a ‘gold standard’. Since only
102 transaction prices are available to us, we can only correlate them with the pooled outcomes
of the valuation models. The correlation, which is corrected for measurement error due to
varying model estimates (DeVellis 2012, 66), amounts to .482 (p < .01). We interpret this a first
evidence for predictive validity of brand valuation models in general with respect to real
24
transaction prices. Due to data limitations, however, we cannot establish that result for any
specific valuation method.
Stock return response model. Table 8 presents the results of the stock return response
model. Recall that we control for several risk factors and the impact of unanticipated accounting
performance information in this model (Mizik and Jacobson 2008). We consider immediate
stock response and future stock response within 1, 5, and 11 months after the new brand value
information is available to the public.
Generally, we find only weak support for predictive validity across brand valuation
models. For most models, the coefficient associated with the unanticipated change in brand value
is not significant. However, we do find positive and significant parameter estimates for the
models of CoreBrand, Interbrand, and Semion in the immediate stock response model. We also
find a significant estimate for the Ailawadi, Lehmann, and Neslin model if we consider future
stock return response within 1 month after announcement. Most interestingly, the CoreBrand
model further demonstrates a strong predictive ability for future returns with respect to 5 and 11
months after new brand value estimates are available.
== Table 8 about here ==
Robustness Checks
We performed several analyses to check the robustness of our results. First, we
substituted the valuation year of the commercial methods for the preceding year. It could be that
the brand values announced in year t in fact represent the information level of year t-1. The
change in periods does not have a substantial impact on our results. Second, we performed the
convergent validity analysis again by excluding the models of Millward Brown model and/or
Semion. Recall that Millward Brown’s performance is very poor. So, we could argue that the
25
inclusion of this model adversely affects the results for other models, especially those that do not
pass the threshold. Another adverse effect might result from including the Semion model since
we have significantly less brand values available from this model. Excluding these models from
the analysis does not change our conclusions. Third, we took the log of all variables in the
nomological validity tests (except for EBIT due to potentially negative values) to check the
stability of Granger-causality results. Our conclusions do not change. Fourth, we considered all 4
lags in calculating the ratio threshold with respect to the Granger-causality tests. Again, our
conclusions do not change.
Summary of Validation Test Results
Table 9 integrates the findings from all validation tests, which helps developing an
overall evaluation of the brand valuation methods. The check marks and crosses indicate whether
a method has passed a specific test or not. We summarize first results across methods and focus
then on the performance of individual methods.
== Table 9 about here ==
The last row of table 9 shows the number of methods that passed a specific test. From this
summary, we draw the following conclusions:
1. Brand valuation methods generally produce reliable and stable results. 2. The results also appear to be largely distinct from other constructs (discriminant
validity). 3. Convergent validity, however, is only sufficiently established within the valuation
category but not across categories. Only market-based methods demonstrate unconstrained convergent validity across all categories.
4. While common correlation analysis suggests nomological validity, there is no support at all from Granger-causality tests. Hence, nomological validity cannot be shown for any method.
5. Only a minority of methods provides evidence of predictive validity. None of the cost-based models shows predictive validity.
26
It is apparent from table 9 that no single method does pass all tests together. If we claim
this to be the standard we have no choice but to conclude that none of the brand valuation
methods is valid and thus useful. However, such a standard is probably too strict. Simply
counting the check marks selects the two market-based methods and the advertising stock model
as best performing models. However, such counting might ignore differences in importance of
the validity dimensions. From a brand management perspective, we consider the
reliability/stability of a method, its convergent validity and its predictive validity as relatively
more important than discriminant and nomological validity. This is because a measurement
method must be reliable if management continuously uses it for decision-making. Convergent
validity strongly indicates that the measure does reflect the conceptual essence of the construct,
which is important to establish a shared understanding about investment activities and objectives
in the company. Finally, predictive validity ensures that the measure is indeed linked with
relevant external performance criteria such as stock price. Following this prioritization, we draw
the following conclusions about individual methods:
1. Only the market-based CoreBrand model convinces across all tree priority validity dimensions reliability, convergent validity, and predictive validity.
2. If we accept that convergent validity within only the valuation category of Income/DCF-based methods is sufficient, the Interbrand model is also performening quite well.
3. The value of the Millward Brown model is highly questionable. It does not pass a single validity test, except for the reliability/stability test.
4. Although the cost-based valuation approach lacks theoretical foundation, the performance of the ad-stock model with respect to reliability, convergent validity, and discriminant validity is remarkable and warrants further attention.
27
DISCUSSION
Managerial Implications
Companies that invest a lot in building and nurturing their brands want to monitor the
outcome of these investments. Measuring the brand value thus is of utmost importance to brand
managers and ensures the attention of top management. While it is crucial to know the absolute
value for transactional purposes it is less relevant in brand management applications. Here, it is
important that management can trust the measure that is to reflect the effects of brand decisions
in a meaningful and consistent way. Our validation study is a first step to provide guidance
through the jungle of brand valuation models available to managers. Managers may learn from
our study in several ways. We cannot fully solve the issue of the best brand valuation model, as
our validation test does not reveal the one superior model that passes every single test. But we
can differentiate between models and valuation approaches that come closer to the ideal and
those that are further away.
It appears that the market-based methods generally perform best along the various
criteria. Specifically, the CoreBrand model turns out to be reliable and to converge with the
results across different valuation categories. Most importantly, it has an impact on both
immediate and future stock returns. Conceptually, we agree with Salinas (2009, 388-395) that the
model setup raises concerns. It uses brand familiarity and favorability from surveying a specific
audience - top managers and decisions makers of other corporations. This might explain why the
model does not fully discriminate from the corporate reputation construct. But it simply performs
better than many other suggested models. It would be worthwhile to further investigate the
reasons for this better performance.
28
The major drawback of the CoreBrand model is that it only applies to corporate brands or
dominant brands of a company. But many companies manage a portfolio of brands that requires
monitoring the value of individual product brands. For these applications, our analysis suggests
the use of the Interbrand model. Conceptual and methodological concerns are also associated
with this model (e.g., Salinas 2009, 215-232). Putting these limitations aside, our study shows
that the model is reliable and has predictive relevance. However, we could not establish
convergent validity across valuation categories but only within its category. If the user is willing
to accept this limitation, the model certainly represents the best choice among future-oriented
Income/DCF-based valuation approaches.
Alternatively, brand managers should consider the use of the Ailawadi, Lehmann, and
Neslin model. It demonstrates both high reliability/stability and predictive validity. Its rather low
performance on convergent validity may be due to the fact that this approach focuses on current-
period brand revenues and profits. All other models involve longer time horizon by either
incorporating expected future returns or accumulating past brand investments. If a focus on only
current-period results is acceptable, the Ailawadi, Lehmann, and Neslin model presents a
valuable alternative that is particularly easy to implement.
Research Implications
Our brand validation results also have implications for researchers and hopefully
stimulate the further development and refinement of valuation models. First, we think that the
suggested methodological framework is a valuable contribution to the measurement and scaling
literature. The availability of time-series data offers new possibilities for testing concepts such as
nomological validity or predictive validity in a more rigorous way than before. Although
generating data over several periods involves higher costs it also offers the chance to verify the
29
causal ordering of nomological networks. Such verification is important if researchers want to
use new constructs for theory testing in their empirical models. In addition, it seems worthwhile
to carefully think about meaningful external criteria that are linked with management objectives
and can be used for testing predictive validity.
Second, in contrast to managerial applications, nomological validity has a higher value to
researchers. Our validation results obviously do not attest nomological validity but rather suggest
that brand valuation results are highly endogenous with their antecedents such as advertising
expenditures and consequences such as profit. This finding is not necessarily surprising, as brand
value is defined as the incremental cash profit that accrues from effective brand investments.
Hence, the construct essentially is defined in terms of its antecedents and consequences. As a
consequence, financial brand equity measures do not qualify as a measure to be used in empirical
models of firm performance unless the performance or criterion variable, respectively, can be
separated in a convincing manner.
Finally, our results should stimulate the research on new valuation models. The surprising
good performance of the ad-stock model deserves more attention. Although it uses past brand
investments it produces reliable results that correlate strongly with other valuation approaches.
Apparently, it shows empirical evidence of capturing the value of a brand. Conceptually, the
concept of an advertising stock is consistent with the idea of brand equity that provides the
potential for future performance even if investments are cut back to a minimum. This might
explain why the ad stock provides a meaningful way to reflect brand value. It should be noted
that we had to use the generalized carryover coefficient of .50 due to the lack of more specific
data. Brand-specific carryover coefficients will better capture the true differences in brand
strength. We encourage more research in this direction, which might lead into new, powerful
30
approaches for brand valuation. In fact, an important advantage of the advertising stock approach
is that it is consistent with the accounting practice of recognizing the value of an asset, which is
depreciated over time.
Limitations and Further Research
Our study is subject to limitations. The results of our study do not inform about the model
that performs best with respect to the absolute metric. Given the large differences between brand
value estimates, this is still an issue of high importance that warrants a solution. The true value
of a brand is not observed. A promising avenue is to use data from real brand transactions
provided the sample is large enough. While we consider nine valuation models, there are many
more models. Salinas (2009) discusses 39 models. Provided that sufficient data are available it
would be interesting to extend our validation tests to other models. Finally, we had to make an
assumption about the choice of the lowest-share brand as benchmark brand for the Ailawadi,
Lehmann, and Neslin (2003) model. This is fully consistent with the recommendation by the
authors (p. 15) and they present evidence that their measure is robust with respect to this
assumption. But it would be interesting to learn more about the potential influence of this
assumption on our validation results.
31
REFERENCES
Ailawadi, Kusum L., Donald R. Lehmann, and Scott A. Neslin (2003), “Revenue Premium as an Outcome Measure of Brand Equity,” Journal of Marketing, 67, 1–17.
Aldrich, John (1995), “Correlations Genuine and Spurious in Pearson and Yule,” Statistical Science, 10 (4), 364–376.
Barwise, Patrick, Christopher Higson, Andrew Likierman, and Paul Marsh (1989), “Accounting for Brand,” London Business School/The Institute of Chartered Accountants in England and Wales, 1-84.
Brand Finance (2013), “Methodology,” (accessed August 24, 2013), [available at http://brandirectory.com/methodology].
Carhart, Mark M. (1997), “On Persistence in Mutual Fund Performance,” Journal of Finance, 52 (1), 57–82.
Churchill, Gilbert A. Jr. (1979), “A Paradigm for Developing Better Measures of Marketing Constructs,” Journal of Marketing Research, 16 (1), 64–73.
CoreBrand (2013), “Methodology,” (accessed November 12, 2013), [available at http://www.corebrand.com/brandpower/methodology].
DeVellis, Robert F. (2012), Scale Development - Theory and Applications. Thousand Oaks: Sage Publications.
Engle, Robert F. and Clive W. J. Granger (1987), “Co-Integration and Error Correction: Representation, Estimation, and Testing,” Econometrica, 55 (2), 251–276.
Fama Eugene F. and Kenneth R. French (2006), “The Value Premium and the CAPM,” Journal of Finance, 61 (5), 2163–2185.
Finn, Adam and Ujwal Kayande (2005), “How fine is C-OAR-SE? A Generalizability Theory Perspective on Rossiter’s procedure,” International Journal of Research in Marketing, 22, 11-21
Fischer, Marc (2007), “Valuing Brand Assets: A Cost-Effective and Easy-to-Implement Measurement Approach,” Marketing Science Institute Working Paper Series, Report No. 07-002, 23–50.
———, Franziska Völckner, and Henrik Sattler (2010), “How Important Are Brands? A Cross-Category, Cross-country Study,” Journal of Marketing Research, 47 (5), 823–839.
Fombrun, Charles J. and M. Shanley (1990), “What’s in a Name: Reputation Building and Corporate Strategy”, Academy of Management Journal, 33(2), 233-258.
32
Fornell, Claes, Michael D. Johnson, Eugene W. Anderson, Jaesung Cha, and Barbara E. Bryant (1996), “The American Customer Satisfaction Index: Nature, Purpose, and Findings,” Journal of Marketing, 70(4), 7–18.
Granger, Clive W. J. (1969), “Investigating Causal Relations by Econometric Models and Cross-
spectral Methods,” Econometrica, 37 (3), 424–438.
Interbrand (2012), “Methodology,” (accessed August 24, 2013), [available at http://www.interbrand.com/de/best-global-brands/2012/best-global-brands-methodology.aspx].
ISO (2010), ISO10668: 2010, “Brand valuation –Requirements for monetary brand valuation” [available at http://www.iso.org/iso/catalogue_detail?csnumber=46032].
Kao, Chiwa (1999). “Spurious Regression and Residual-Based Tests for Cointegration in Panel Data,” Journal of Econometrics, 90(1), 1–44.
Keller, Levin L. and Donald R. Lehmann (2003), “The brand value chain: Optimizing strategic and financial brand performance,” Marketing Management, (May/June), 26–31.
Levin, Andrew, Chien-Fu Lin, and Chia-Shang James Chu (2002), “Unit Root Tests in Panel Data: Asymptotic and Finite-sample Properties,” Journal of Econometrics, 108 (1), 1–24.
Millward Brown (2013), “Methodology,” (accessed August 24, 2013), [available at http://www.millwardbrown.com/BrandZ/Top_100_Global_Brands/Methodology.aspx].
Mizik, Natalie and Robert Jacobson (2008). The Financial Value Impact of Perceptual Brand Attributes. Journal of Marketing Research, 45(1), 15-32
Nerlove, Marc and Kenneth J. Arrow (1962), “Optimal Advertising Policy under Dynamic Conditions,” Economica, 29 (114), 129–142.
Netemeyer, Richard G, William O. Bearden, and Subash Sharma (2003), Scaling Procedures: Issues and Applications. Thousand Oaks: Sage Publications.
Peter, Paul J. (1979), “Reliability: A Review of Psychometric Basics and Recent Marketing Practices,” Journal of Marketing Research, 16 (1), 6–17.
——— (1981), “Construct Validity: A Review of Basic Issues and Marketing Practices,” Journal of Marketing Research, 18 (2), 133–145.
Salinas, Gabriela (2009), The International Brand Valuation Manual. Chichester: Wiley.
Semion (2013), “semion® brand€valuation,” (accessed August 24, 2013), [available at http://www.semion.com/d/wert01b.htm].
33
Sethuraman, Raj, Gerard J. Tellis, and Richard Briesch (2011), “How Well Does Advertising Work? Generalizations from Meta-Analysis of Brand Advertising Elasticities,” Journal of Marketing Research, 48 (3), 457–471.
Simon, Carol J. and Mary W. Sullivan (1993), “The Measurement and Determinants of Brand Equity: A Financial Approach,” Marketing Science, 12 (1), 28–52.
Srinivasan, Shuba and Dominique M. Hanssens (2009), “Marketing and Firm Value: Metrics, Methods, Findings, and Future Directions,” Journal of Marketing Research, 46 (3), 293–312.
34
FIGURE 1 TESTING FOR MEANINGFUL ASSOCIATIONS WITH TIME-SERIES DATA
No
Correlation analysis Co-integration test (Kao
1999)
Yes
Are both time-series evolving?
Are both time-series stationary?
No
Yes
Test for unit root (Levin, Lin, and Chu 2002)
35
FIGURE 2
NETWORK OF NOMOLOGICAL RELATIONSHIPS
Antecedents(Period't)1' Period't' Period't+1'
Consequences(
Log'of'Adver1sing''expenditures'
Log'of'Selling'&'general'administra1on'expenditures'
Brand'value'
measure'
Firm'market'capitaliza1on'
Price)to)'book'value'
Customer)''based'brand''equity'
Sales'
Profit'
36
TABLE 1 METHODOLOGICAL FRAMEWORK
Criterion Test statistic Threshold Source Reliability/ stability
1. Test-retest correlation r between brand value in year t and t-1
2. Test-retest correlation r of brand value within year t
rt-1 ≥ .90 rwithin ≥ .95
Ailawadi, Lehmann, and Neslin (2003); Churchill (1979); DeVellis (2012, 109)
3. Variance decomposition into cross-sectional (brands) and time variance
Cross-sectional varianceTime variance
≥ 3
Convergent validity
4. Correlation r between brand values of different methods (stationary series)
a) At least 75 % significant pairwise correlations (p < .05) b) Average significant rconv ≥ .50
Fischer, Völckner, and Sattler (2010)
5. Panel co-integration test for brand values of different methods (non-stationary series)
ADFconv > tADF Kao (1999)
Discriminant validity
6. Correlation r between brand values and distinct measures (stationary series)
Average rdiscr < .30
7. Panel co-integration test for brand values and distinct measures (non-stationary series)
ADFdiscr < tADF Kao (1999)
Nomological validity
8. Correlation r between brand values and nomological variables (stationary series)
a) More than 50 % significant pairwise correlations (p < .05) b) Average significant rnomol ≥ .40
Finn and Kayande (2005); Fischer, Völckner, and Sattler (2010)
9. Panel co-integration test for brand values of different methods (non-stationary series)
ADFnomol > tADF Kao (1999)
10. Granger causality test for relations between brand values and both antecedents and consequences
a) More than 50 % significant supported relations (p < .05)
b)
# supported relations# reverse relations
> 2
Predictive validity
11. Stock return response model (t-test)
Positive and significant estimated coefficient for brand value measure (t > 1.96; p < .05)
Carhart (1997); Fama and French (2006); Mizik and Jacobson (2008)
37
TABLE 2 OVERVIEW OF BRAND VALUE DATABASE
Period Brands Observations Mean no. of
observations per brand
Median no. of observations
per brand
Mean no. of joint observ. per
method Cost-based methods Ad-stock model 1990-2011 186 2,458 13.22 14.00 818 Historical costs 1990-2011 186 2,458 13.22 14.00 818 Market-based methods
Simon and Sullivan (1993) 1992-2012 438 5,571 12.72 14.00 913
CoreBrand 2002-2012 672 3,979 5.92 6.00 982 Income/DCF-based methods
Future-oriented Interbrand 1992-2012 1,027 3,841 3.74 3.00 557 Millward Brown 2006-2012 324 1,175 3.63 4.00 305 Semion 1997-2012 78 774 9.92 12.00 58 Brand Finance 2006-2012 2,752 5,950 2.16 2.00 521 Current-period oriented
Ailawadi, Lehmann, and Neslin (2003)
1997-2012
876
10,786
12.31
13.00
1,381
Total 1990-2012 3,879 36,992 5.66 3.00 706
38
TABLE 3 RELIABILITY AND STABILITY TEST RESULTS
Test-retest correlation Variance decomposition
Year t
with year t-1
Within year t
Ratio of cross-sectional to time variance
Cross-sectional variance
Time variance
Threshold ≥ .90 ≥ .95 ≥ 3.00 ≥ 75.0 % ≤ 25.0 % Cost-based methods Ad-stock model .996 n.a. 3.26 76.5 % 23.5 % Historical costs .998 n.a. 1.64 62.1 % 37.9 % Market-based methods
Simon and Sullivan (1993)
.993 n.a. 7.70 88.5 % 11.5 %
CoreBrand .964 n.a. 10.49 91.3 % 8.7 % Income/DCF-based methods
Future-oriented Interbrand .990 .983 8.35 89.3 % 1.7 % Millward Brown .952 .997 5.76 85.2 % 14.8 % Semion .987 .970 14.63 93.6 % 6.4 % Brand Finance .942 .984 13.08 92.9 % 7.1 % Current-period oriented
Ailawadi, Lehmann, and Neslin (2003)
.901 n.a. 3.12 76.1 % 23.9 %
Notes: All correlation coefficients are highly significant at p < .01; n.a. = not applicable due to missing observations.
39
TABLE 4 CONVERGENT VALIDITY TEST RESULTS
Within and across valuation categories Within valuation category No. of significant
correlations (p < .05) Average significant
correlation
Average significant correlation
Threshold ≥ 75 % ≥ .50 ≥ .50 Cost-based methods Ad-stock model 6 (86%)1) .554 .928 Historical costs 6 (86%)1) .4622) .928 Market-based methods Simon and Sullivan (1993) 7 (88%) .6072) .693
CoreBrand 7 (100%)1) .583 .693 Income/DCF-based methods
Future-oriented Interbrand 8 (100%) .494 .660 Millward Brown 4 (50%) .540 .543 Semion 5 (100%)3) .4423) .401 Brand Finance 8 (100%) .520 .625 Current-period oriented Ailawadi, Lehmann, and Neslin (2003) 7 (88%) .397 n.a
Notes: n.a. = not applicable since only one current-period income model 1)
The total number of pairwise correlations is 7 due to insufficient joint observations with the Semion model. 2)
Correlation between historical costs model and Simon and Sullivan (1993) model not included due to non-stationary time-series. Series are co-integrated (ADF = 38.10; p < .01). 3)
The total number of pairwise correlations is 5 due to insufficient joint observations with all models.
40
TABLE 5 DISCRIMINANT VALIDITY TEST RESULTS
Average correlation across methods
(convergent validity)
American Customer Satisfaction Index
Fortune Corporate Reputation Index
Threshold < .30 < .30 Cost-based methods Ad-stock model .554 .219 .047NS Historical costs .462 .258 .007NS
Market-based methods Simon and Sullivan (1993) .607 .061NS .051NS
CoreBrand .583 .101 .339 Income/DCF-based methods
Future-oriented Interbrand .494 -.267 .251 Millward Brown .540 -.078NS .310 Semion .442 n.a. -.073NS Brand Finance .520 .002NS .345 Current-period oriented Ailawadi, Lehmann, and Neslin (2003) .397 .016NS .073
Notes: NS = not significant (p > .05; two-sided t-test); n.a. = not applicable since less than 50 observations available.
41
TABLE 6 NOMOLOGICAL VALIDITY TEST RESULTS (CORRELATIONS)
Three antecedents (year t-1) Four consequences (year t+1) No. of significant
correlations (p < .05) Average significant
correlation No. of significant
correlations (p < .05) Average significant
correlation Threshold > 50% ≥ .40 > 50% ≥ .40 Cost-based methods ! = .634 ! = .430 Ad-stock model 2 (66%) .653 3 (75%) .456 Historical costs 2 (66%) .6141) 3 (75%) .4031) Market-based methods ! = .519 ! = .584 Simon and Sullivan (1993) 2 (66%) .6241) 3 (75%) .5881)
CoreBrand 3 (100%) .414 4 (100%) .579 Income/DCF-based methods
Future-oriented ! = .413 ! = .431 Interbrand 3 (100%) .359 4 (100%) .407 Millward Brown 3 (100%) .235 4 (100%) .277 Semion 3 (100%) .627 4 (100%) .449 Brand Finance 3 (100%) .431 3 (75%) .591 Current-period oriented Ailawadi, Lehmann and Neslin (2003) 3 (100%) .266 3 (75%) .732
Across all methods .469 .498 Notes: ! refers to the average correlation within the respective subcategory. 1) Correlation between Log(SG&A) expenditures and both historical costs model and Simon and Sullivan (1993) model not included due to non-stationary time-series. Series are co-integrated (ADF = 12.52 and 40.52, respectively; p < .01).
42
TABLE 7 NOMOLOGICAL VALIDITY TEST RESULTS (GRANGER CAUSALITY)
Notes: Non-stationary time-series are transformed into first differences for the Granger-causality tests. 1) Only two variables considered due to insufficient sample size.
No. of significant supported relations
(p < .05)
No. of significant reverse relations
(p < .05)
Ratio of supported to reverse relations
(p < .05) Antecedents Consequences Antecedents Consequences Antecedents Consequences
Threshold > 50% > 50% > 2 > 2 Cost-based methods Ad-stock model 3 (100%) 3 (75%) 3 (100%) 3 (75%) 1 1 Historical costs 1 (33%) 3 (75%) 3 (100%) 3 (75%) .33 1 Market-based methods Simon and Sullivan (1993)
2 (66%) 3 (75%) 3 (100%) 3 (75%) .66 1
CoreBrand 2 (66%) 3 (75%) 2 (66%) 3 (75%) 1 1 Income/DCF-based models
Future-oriented Interbrand 2 (66%) 3 (75%) 1 (33%) 4 (100%) 2 .75 Millward Brown 0 (0%) 2 (50%) 1 (33%) 3 (75%) 0 .66 Semion 1 (50%) 1) 3 (75%) 2 (100%) 1) 3 (75%) .50 1 Brand Finance 3 (100%) 3 (75%) 3 (100%) 3 (75%) 1 1 Current-period oriented Ailawadi, Lehmann and Neslin (2003)
2 (66%) 3 (75%) 3 (100%) 3 (75%) .66 1
Across all methods 1.8 (60%) 2.9 (72%) 2.3 (81%) 3.1 (78%) .79 .93
43
TABLE 8 PREDICTIVE VALIDITY TEST RESULTS (STOCK RETURN RESPONSE MODEL)
Ad-stock model
Historical costs
Simon and Sullivan (1993)
CoreBrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann, & Neslin (2003)
Immediate stock return response UΔROA1) 6.095
(0.99) 6.042 (.990)
5.315 (0.395)
3.114 (0.763)
2.637 (1.262)
4.595NS (2.895)
-1.880NS (2.092)
5.649 (1.497)
9.853 (.512)
UΔBV1) -.003NS (.019)
-.017NS (.015)
-.008NS (.455)
.006 (.003)
.007 (.004)
.002NS (.002)
.032 (.014)
.004NS (.005)
.000NS (.001)
F-Value 6.41 6.49 16.33 3.38 2.56 4.25 4.19 6.67 26.96 N 1,759 1,759 4,215 2,736 1,099 380 575 1,520 8,005 Future stock return response within 1 month after announcement UΔBV1) -.083NS
(.071) -.083NS (.054)
-.001NS (.017)
.014NS (.011)
-.012NS (.013)
-.015NS (.008)
-.025NS (.054)
.003NS (.018)
.017 (.005)
F-Value 2.69 2.75 2.92 4.91 2.64 7.36 1.51NS 4.63 4.08 N 1,802 1,802 4,226 2,738 1,101 380 576 1,519 8,029 Future stock return response within 5 months after announcement UΔBV1) -.500NS
(.305) -.493 (.232)
-.033NS (.068)
.144 (.043)
.000NS
(.005) .000NS (.004)
.003NS (.021)
-.007NS
(.007) -.003NS (.021)
F-Value 3.25 3.35 5.87 3.21 4.79 1.53NS 5.94 6.96 5.94 N 1,802 1,802 4,420 2,736 1,101 380 576 1515 8,013 Future stock return response within 11 months after announcement UΔBV1) -.446
(.019) -.045 (.015)
-.009NS (.005)
.011 (.003)
-.003NS (.004)
-.001NS (.002)
-.005NS
(.014) -.003NS (.004)
.000 NS (0.001)
F-Value 3.23 3.46 5.75 2.12 3.14 2.05NS 4.70 5.30 8.91 N 1,799 1,799 4,206 2,734 1,101 380 576 1,502 7,973 Notes: Standard error are in parentheses; NS = not significant (p < .05; one-sided t-test); UΔROA = unanticipated change in accounting performance; UΔBV = unanticipated change in brand value; N = sample size. Parameter estimates for intercept, annual dummies, and accounting performance (for 1, 5, and 11 months future stock return models) are not reported in the table. 1) For reading convenience we multiply coefficients by 10,000.
44
TABLE 9 SUMMARY OF VALIDATION TEST RESULTS
Reliability/stability Convergent validity Discriminant validity
Nomological Validity Predictive validity
Test statistic Test-retest correlation
Variance decomposition
Correlation / Co-integration test
Correlation / Co-integration test
Correlation / Co-integration test
Granger causality test
Stock return response model (t-test)
Threshold rt-1 ≥ .90 / rwithin ≥ .95
Ratio of cross-sectional to time variance ≥ 3
rconv ≥ .50 / ADFconv > tADF rdiscr < .30 /
ADFdiscr < tADF rnomol ≥ .40 / ADFnomol > tADF Ratio of supported
to reverse relations > 2
t > 1.96, p < .05
Across all methods
Within categories
Cost-based methods Ad-stock model ✓ ✓ ✓ ✓ ✓ ✓ ✕ ✕ Historical costs ✓ ✕ ✕ ✓ ✓ ✓ ✕ ✕
Market-based methods Simon and Sullivan (1993)
✓ ✓ ✓ ✓ ✓ ✓ ✕ ✕
CoreBrand ✓ ✓ ✓ ✓ ✕ ✓ ✕ ✓
Income/DCF-based methods
Future-oriented Interbrand ✓ ✓ ✕ ✓ ✓ ✕ ✕ ✓ Millward Brown ✓ ✓ ✕ ✓ ✕ ✕ ✕ ✕ Semion ✓ ✓ ✕ ✕ ✓ ✓ ✕ ✓ Brand Finance ✓ ✓ ✓ ✓ ✕ ✓ ✕ ✕ Current-period oriented
Ailawadi, Lehmann, and Neslin (2003)
✓ ✓ ✕ n.a. ✓ ✕ ✕ ✓
Across all methods 9 of 9 8 of 9 4 of 9 7 of 8 6 of 9 6 of 9 0 of 9 4 of 9 Notes: ✓ passed, ✕ not passed; r denotes the correlation coefficient; ADF is the augmented Dickey-Fuller test statistic applied in co-integration tests;
and t refers to the respective t-statistic.
45
APPENDIX
Appendix A: Granger-causality tests
We test Granger (1969) causality between a antecedent variable, x (e.g., advertising), and a
brand value metric, y (e.g., Interbrand), by the following regressions:
(A.1a)
yt =α0,1a + α1,1a,l yt−l +l=1
L=4
∑ α 2,1a,l xt−l + ε1al=1
L=4
∑ (supposed relation)
(A.1b)
xt =α0,1b + α1,1b,l xt−l +l=1
L=4
∑ α 2,1b,l yt−l + ε1bl=1
L=4
∑ (reverse relation)
where denotes the parameters to be estimated, the error term, t the year, and L the
number of lags. We further specify the estimation equations for brand value metrics and each
consequence variables, z (e.g., firm profit):
(A.2a) zt =α0,2a + α1,2a,l zt−l +
l=1
L=4∑ α2,2a,l yt−l + ε2a
l=1
L=4
∑ (supposed relation)
(A.2b)
yt =α0,2b + α1,2b,l yt−l +l=1
L=4
∑ α 2,2b,l zt−l + ε2bl=1
L=4
∑ (reverse relation)
Note that we take the first differences for non-stationary time series in each regression of eq.
(A.1a) to (A.2b) to ensure that variables are integrated by the same order.
Granger-causality is supported if the F-value for the joint hypothesis of α 2,i,l = ...=α 2,i,L = 0
exceeds the respective Wald statistic. The effective number of estimation coefficients relates to the number of lags l that eventually has been specified. We test four lag structures l
= 1,..., 4.
α ε
α
46
Appendix B: Stock return response model We adopt the framework of Mizik (2009) and expect abnormal returns to depend on
unexpected changes in profitability and financial brand values. Abnormal return is the
difference between actual and expected stock return. We estimate expected stock return
according to the four-factor model (Carhart 1997, Fama and French 2006). The four factors
are provided by the Center for Research in Security Prices (CRSP) and are based on regional
values depending on the primary listing of the company. They reflect the country-specific
risk more precisely than the Global factors, i.e. the aggregate of the regional factors. Regional
factors are available for Europe, Japan, Asia Pacific, and the US. Only firms that are listed in
these regions entered our analysis. We use monthly stock returns to ensure correct alignment
with the month in which financial brand equity metrics become available. Since financial
brand values are measured on an annual basis we take the geometric mean of abnormal
returns within the corresponding 12 months brand value measurement period.
We measure unexpected changes in profitability and brand value as the residuals of a fixed-
effect, first-order autoregressive model of firm profitability and the financial brand equities,
respectively. The usage of unanticipated changes instead of actual levels follows the efficient
market hypothesis (Fama 1970). They also avoid error terms to be serially correlated (Greene
2011). We additionally control for economy-wide risk by yearly dummies and estimate
abnormal returns on financial brand values by the following regression:
(B.1) AStkRetit = β0 + β1UΔAccPit + β2UΔBVit + βk+1YD +
k=1
K−1∑ ηit , with
where AStkRetit denotes abnormal stock return of brand i‘s parental firm in period t, UΔAccPit
and UΔBVit unanticipated changes in profitability and financial brand values, respectively, β
parameter estimates, YD yearly dummies, and and the error term and variance.
Recent research further considers capital markets not only to respond when new information
arrives but also when their total financial implications have been fully understood (e.g., Brav
and Heaton 2002, Srinivasan and Hanssens 2009). We thus test for lagged effects that might
occur within 1, 5, and 11 months after the brand value announcements. In these cases, the
brand value variable effectively precedes future abnormal stock returns in eq. (B.1). Again,
abnormal stock returns are measured as geometric mean.
ηit~
i.i.d
N (0,ση2 )
η ση
2
47
Appendix C: Overview of periods and data sources Variable Period Data source Commercial financial brand values
Interbrand 1992-2012 Interbrand, www.interbrand.com Millward Brown 2006-2012 Millward Brown, www.millwardbrown.com Semion 1997-2012 semion® brand-broker Gmbh, www.semion.de
Brand Finance 2006-2012 Brand Finance, www.brandfinance.com Corebrand 2002-2012 Corebrand, www.corebrand.com
Academic financial brand values Cost-based metrics
Advertising 1990-2011 Compustat Simon & Sullivan (1993)
Intangible asset value, concentration ratio, advertising spending, -share, market share, R&D share, brand age, order of entry
1992-2012
Compustat, Thomson Banker One, company reports, company founding dates (Field and Karpoff 2002, Loughran and Ritter 2004), Internet research
Ailawadi, Lehmann & Neslin (2003) Revenues, industry profit margin 1997-2012 Compustat, SIC classification
Brand transactions prices Brand values derived from M&As 2001-2012 SEC filings
Additional data for construct validity tests
American Customer Satisfaction Index 1994-2012 www.theacsi.org Fortune Reputation Index 1992-2012 Fortune’s Most Admired Companies Advertising spending, SG&A spending, revenues, EBIT, market capitalization, market-to-book value
1992-2012 Thomson Banker One
Harris EquiTrend 2005-2012 Harris Interactive Additional data for predictive validity tests
Stock return 1992-2012 Thomson Banker One Market factor, size factor, value factor, momentum factor
1992-2012 Center for Research in Stock Prices (CRSP)
Profitability 1992-2012 Thomson Banker One
48
Appendix D: Overview of joint observations Table D.1: Number of joint observations between financial brand valuation methods
Ad stock model
Capitalized costs
Simon and Sullivan (1993)
Corebrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann & Neslin
(2003) Cost-based models Ad-stock model 2,458 Historical costs 2,458 2,458 Market-based models Simon and Sullivan (1993)
763 763 5,571
Corebrand 929 929 1,489 3,979 Income/DCF forecast-based models
Future-oriented Interbrand 628 628 566 507 3,841 Millward Brown 207 207 131 285 539 1,175 Semion 51) 51) 98 191) 101 53 774 Brand Finance 299 299 435 556 728 681 102 5,950 Current period-oriented Ailawadi, Lehmann, and Neslin (2003)
1,258 1,258 3,147 3,140 756 333 84 1,069 10,786
1) We do not calculate correlation coefficients of these relationships since the no. of joint observations is less than 50.
Table D.2: Number of joint observations with distinct marketing constructs
American Customer Satisfaction Index
Fortune Corporate Reputation Index
Cost-based methods Ad-stock model 888 967 Historical costs 888 967
Market-based methods Simon and Sullivan (1993)
351 700
Corebrand 726 1,183 Income/DCF forecast-based methods
Future-oriented Interbrand 437 598 Millward Brown 182 257 Semion 151) 89 Brand Finance 292 565 Current period-oriented Ailawadi, Lehmann, and Neslin (2003)
1,135 1,723
1) We do not calculate the correlation coefficients of this relationship since the
no. of joint observations is less than 50.
49
Table D.3: Number of joint observations with nomological variables
Antecedents (t-1) Consequences (t+1) Log Ad
spending Log SG&A spending
Harris EquiTrend
Sales EBIT Market capitalizatio
n
Market-to-book value
Cost-based methods Ad-stock model 2,046 1,903 409 2,096 2,096 2,124 2,116 Historical costs 2,046 1,903 409 2,096 2,096 2,124 2,116 Market-based methods Simon and Sullivan (1993)
1,973 4,984 339 4,729 4,748 4,757 4,733
Corebrand 1,877 3,167 658 3,384 3,384 3,369 3,347 Income/DCF-forecast based methods
Future-oriented Interbrand 946 1,624 520 1,820 1,820 1,798 1,793 Millward Brown 328 510 642 507 507 492 490 Semion 60 583 50 651 651 643 643 Brand Finance 908 2,402 875 3,114 3,114 3,094 3,117 Current-period oriented Ailawadi, Lehmann, and Neslin (2003)
3,777 7,566 807 8,978 8,978 8,871 8,823
50
Appendix E: Detailed validation test results Table D.1: Results on convergent validity
Ad stock model
Capitalized costs
Simon and Sullivan (1993)
CoreBrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann & Neslin
(2003) Required thresholds Cost-based models Ad-stock model 1 Historical costs .928 1 Market-based models Simon and Sullivan (1993)
.817 .2101) 1
Corebrand .472 .406 .693 1 Income/DCF-based models
Future-oriented Interbrand .375 .373 .319 .658 1 Millward Brown .018NS .040NS .092NS .530 .672 1 Semion n.a. n.a. .795 n.a. .525 .264 1 Brand Finance .334 .276 .510 .698 .770 .690 .411 1 Current-period oriented Ailawadi, Lehmann, and Neslin (2003)
.401 .328 .510 .623 .240 -.015NS .209 .470 1
Notes: Bold = significant, NS = Not significant (p > .05; one-sided t-test); n.a. = not applicable since less than 50 observations 1)
The correlation coefficient between these series bases on first differences since they are non-stationary. The panel co-integration test indicates convergence towards a common and stable equilibrium for these variables in the long run (ADF = 38.10; p < 01).
51
Table D.2: Results on nomological validity
Antecedents (t-1) Consequences (t+1) Log Ad
spending Log SG&A spending
Harris EquiTrend
Sales EBIT Market capitalizatio
n
Market-to-book value
Cost-based methods Ad-stock model .720 .585 -.015NS .469 .420 .480 -.018NS Historical costs .614 -.0461) .039NS .0781) .386 .420 -.019NS Market-based methods Simon and Sullivan (1993)
.624 .1561) -.021NS .1141) .538 .637 -.001NS
CoreBrand .519 .517 .207 .680 .710 .901 .023NS Income/DCF- based methods
Future-oriented Interbrand .442 .436 .198 .474 .444 .641 .070 Millward Brown .217 .346 .143 .169 .309 .552 .078 Semion .845 .758 .277 .798 .524 .583 -.110 Brand Finance .543 .550 .201 .646 .456 .670 .024NS Current-period oriented Ailawadi, Lehmann, and Neslin (2003) .457 .432 -.091 .890 .670 .636 -.008NS
Notes: No. of observations in parentheses; Bold = significant, NS = Not significant (p > .05; one-sided t-test) 1)
The correlation coefficient between these series bases on first differences since they are non-stationary. The panel co-integration test indicates a stable and long run relationship between these variables. The corresponding ADF statistic for Simon and Sullivan (1993) and for the capitalized costs model is 12.52 and 40.52, respectively (both p < .10).
52
Table D.3: F-values of bivariate Granger-causality regressions (Eq. A.1a and A.1b)
Dependent variable
Ad-stock model
Historical costs1)
Simon and Sullivan (1993)1)
CoreBrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann & Neslin (2003)
Predictor variable Advertising Lags (t) 1 4577.99 (2044) .00 (1934) 6.10 (1550) .11 (1557) 3.44 (661) 2.43 (219) 2.15 (58) 2.08 (528) 1.4 (3555) 2 2.54 (1813) n.a. 62.04 (1335) 5.54 (768) 5.93 (488) .84 (149) n.a. 5.04 (312) .12 (3014) 3 8.30 (1619) n.a. 24.11 (1148) 6.36 (990) 2.37 (377) .46 (97) n.a. 2.15 (161) 3.86 (2531) 4 n.a. n.a. 19.05 (981) 5.54 (768) 1.03 (301) .17 (57) n.a. 1.07 (55) 3.12 (2077) SGA1) 1 174.90 (1701) 2.76 (1636) .02 (4108) 2.28 (2556) .02 (1100) 2.42 (355) 17.19 (510) 7.64 (1208) 277.81 (6607) 2 2.80 (1521) 3.65 (1460) 5.60 (3696) 53.73 (1299) .27 (810) 1.20 (263) 8.93 (446) 3.54 (594) 9.77 (5768) 3 7.13 (1355) 5.64 (1297) 21.37 (3310) 37.48 (1667) .51 (594) .99 (181) 12.62 (392) .65 (299) 51.88 (5029) 4 4.42 (1224) 4.47 (1169) 2.76 (2953) 53.73 (1299) 1.66 (457) .47 (111) 9.77 (342) .91 (107) 31.63 (4303) Harris
Equitrend
1 1.32 (392) 1.05 (387) .42 (304) .57 (600) 5.90 (452) 1.68 (488) n.a. 11.63 (593) .53 (766) 2 1.08 (299) .96 (295) .63 (224) 2.26 (185) 4.20 (328) 2.84 (352) n.a. 5.76 (387) 1.74 (579) 3 2.37 (214) 2.45 (211) .06 (161) .64 (299) 1.17 (233) .40 (232) n.a. 3.10 (230) .24 (434) 4 2.54 (134) 1.98 (132) 2.33 (100) 2.26 (185) .61 (168) 1.78 (140) n.a 1.37 (97) .60 (295) Predictor variable
Ad-stock model
Capitalized costs1)
Simon and Sullivan (1993)1)
Corebrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann & Neslin (2003)
Dependent variable
Advertising Lags (t) 1 1.15 (2044) 35.67 (1934) 2.88 (1550) 65.84 (1557) .13 (661) 4.88 (219) 6.01 (58) 5.88 (528) 2.45 (3555) 2 .03 (1813) n.a. 5.48 (1335) 59.25 (768) .51 (488) 3.52 (149) n.a. 2.12 (312) 16.83 (3014) 3 2.09 (1619) n.a. .55 (1148) 64.08 (990) 1.98 (377) 2.29 (97) n.a. .39 (161) 1.93 (2531) 4 n.a. n.a. 14.13 (981) 5.54 (768) 1.24 (301) 4.14 (57) n.a. 3.34 (55) 9.43 (2077) SGA1) 1 .70 (1701) 6.15 (1636) .08 (4108) 19.02 (2556) .26 (1100) .05 (355) 11.14 (510) 4.49 (1208) 331.53 (6607) 2 22.62 (1521) 21.25 (1460) 3.20 (3696) 6.82 (1299) .12 (810) .05 (263) 4.14 (446) 1.14 (594) 159.81 (5768) 3 14.68 (1355) 17.21 (1297) 4.47 (3310) 9.35 (1667) .03 (594) .20 (181) 2.98 (392) 2.14 (299) 92.42 (5029) 4 12.86 (1224) 14.06 (1169) 56.55 (2953) 6.82 (1299) .03 (457) .96 (111) 2.51 (342) 2.34 (107) 63.18 (4303) Harris
Equitrend
1 .14 (392) .16 (387) 4.31 (304) .34 (600) 1.88 (452) .79 (488) n.a. 1.31 (593) 2.97 (766) 2 3.09 (299) 4.33 (295) 3.98 (224) .86 (185) 5.37 (328) .30 (352) n.a. 6.64 (387) 2.95 (579) 3 2.09 (214) 4.40 (211) 3.21 (161) .39 (299) 4.20 (233) .08 (232) n.a. 5.92 (230) 6.02 (434) 4 4.42 (134) 5.41 (132) 1.45 (100) .86 (185) 3.09 (168) 1.36 (140) n.a. 3.68 (97) 4.90 (295) Notes: No. of observations in parentheses; Bold = significant, otherwise p > .05; n.a. = not applicable 1) First differences are used to ensure that time series are integrated by the same order.
53
Table D.4: F-values of bivariate Granger-causality regressions (Eq. A.2a and A.2b)
Dependent variable
Ad-stock model
Historical costs1)
Simon and Sullivan (1993)1)
CoreBrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann & Neslin (2003)
Predictor variable Sales1) Lags (t) 1 101.31 (1761) .13 (1688) 3.92 (3973) 29.26 (2792) 19.68 (1238)
(1238) 2.65 (430) 36.96 (619) 47.52 (2088) 203.32 (8290)
2 1.31 (1567) .75 (1500) 4.70 (3551) 16.37 (1425) 9.84 (884) .32 (313) 17.60 (551) 11.78 (1096) 175.63 (7332) 3 8.58 (1386) 5.37 (1324) 5.84 (3156) 12.54 (1826) 7.14 (643) .16 (212) 11.66 (489) 7.63 (535) 45.43 (6471) 4 4.13 (1251) 4.55 (1192) 3.89 (2796) 16.37 (1425) 5.09 (492) .89 (129) 13.42 (432) .63 (155) 101.89 (5624) EBIT 1 22.72 (1886) 10.43 (1708) 27.75 (4020) 3.89 (2805) 3.24 (1255) 1.57 (431) .00 (623) 101.59 (2091) .01 (8362) 2 12.38 (1688) 13.99 (1518) 36.85 (3593) 2.97 (1431) 14.43 (893) 5.74 (314) 11.63 (555) 5.16 (1101) .89 (7399) 3 1.06 (1500) 8.08 (1339) 23.08 (3197) 1.72 (1835) 11.72 (650) 2.33 (212) 5.88 (493) 13.39 (538) 2.05 (6532) 4 6.68 (1324) 4.51 (1206) 18.53 (2833) 2.97 (1431) 7.26 (497) 1.20 (129) 5.80 (435) 2.55 (156) 3.88 (5682) Market capitalization 1 67.86 (1911) 14.09 (1737) 162.52 (4122) 2.61 (2818) 47.78 (1274) 18.30 (434) .83 (616) 335.44 (2140) 284.49 (8382) 2 26.61 (1716) 24.32 (1548) 138.69 (3712) 12.89 (1451) 53.70 (908) 42.99 (319) 11.87 (549) 285.66 (1135) 176.77 (7435) 3 16.85 (1528) 13.36 (1369) 54.85 (3326) 19.67 (1854) 3.04 (662) 24.48 (217) 1.04 (488) 66.25 (559) 157.69 (6578) 4 1.18 (1350) 11.07 (1236) 47.00 (2971) 12.89 (1451) 2.26 ( 510) 14.83 (133) 12.56 (431) 12.74 (158) 119.76 (5734) Price-to-book value 1 .04 (1898) .02 (1725) .93 (4085) .00 (2803) 9.60 (1271) 4.57 (433) 1.04 (616) .24 (2151) .05 (8321) 2 .03 (1697) .03 (1530) .66 (3667) .05 (1441) 1.83 (908) 1.86 (318) .83 (549) .68 (1139) .03 (7367) 3 .08 (1503) .09 (1346) .62 (3277) .02 (1844) .87 (663) .17 (216) 1.06 (488) .23 (560) .00 (6516) 4 .08 (1321) .06 (1210) .62 (2921) .05 (1441) .50 (511) .04 (132) 1.70 (431) .77 (158) .00 (5679) Predictor variable
Ad-stock model
Capitalized costs 1)
Simon and Sullivan (1993)1)
Corebrand Interbrand Millward Brown
Semion Brand Finance
Ailawadi, Lehmann, and Neslin (1993)
Dependent variable
Sales1) 1 .62 (1761) 4.88 (1688) 98.19 (3973) 190.60 (2792) 9.94 (1238) .97 (430) 14.67 (619) 12.31 (2088) 268.80 (8290) 2 8.07 (1567) 25.92 (1500) 75.7 (3551) 52.96 (1425) 7.56 (884) 6.22 (313) 5.49 (551) 1.11 (1096) 173.42 (7332) 3 2.22 (1386) 21.72 (1324) 62.28 (3156) 83.50 (1826) 5.46 (643) 7.40 (212) 7.47 (489) 7.26 (535) 44.25 (6471) 4 15.86 (1251) 20.37 (1192) 41.64 (2796) 52.96 (1425) 5.10 (492) 11.06 (129) 8.66 (432) 4.97 (155) 18.58 (5624) EBIT 1 33.09 (1824) 54.02 (1708) 27.17 (4020) 470.81 (2805) 7.98 (1122) 2.6 (357) 27.73 (529) 4.45 (1223) 14.16 (6727) 2 35.61 (1636) 7.08 (1518) 31.03 (3593) 58.61 (1431) .16 (819) .21 (265) 11.71 (465) .74 (601) 169.41 (5882) 3 36.70 (1460) 7.67 (1339) 31.06 (3197) 94.44 (1835) .09 (601) .33 (182) 8.73 (408) 3.22 (305) 101.49 (5135) 4 27.50 (1297) 4.92 (1206) 36.57 (2833) 58.61 (1431) .10 (462) .87 (112) 6.95 (357) 4.60 (107) 68.61 (4406) Market capitalization 1 7.56 (1911) 7.95 (1737) 5.12 (4122) 51.77 (2818) 23.89 (1274) 27.14 (434) 2.04 (616) 32.27 (2140) .71 (8382) 2 2.44 (1716) 0.93 (1548) 5.79 (3712) 45.71 (1451) 2.68 (908) 11.32 (319) 8.74 (549) 13.42 (1135) 3.97 (7435) 3 1.19 (1528) 2.24 (1369) 7.71 (3326) 69.83 (1854) 1.20 (662) 16.20 (217) 4.99 (488) 2.82 (559) 6.57 (6578) 4 2.7 (1350) 4.06 (1236) 4.61 (2971) 45.71 (1451) 2.85 (510) 4.60 (133) 9.35 (431) 2.32 (158) 1.83 (5734) Price-to-book value 1 .27 (1898) .05 (1725) .95 (4085) .17 (2803) 2.75 (1271) 1.71 (433) .60 (616) 1.19 (2151) .45 (8321) 2 .15 (1697) .54 (1530) .54 (3667) .21 (1441) .56 (908) 1.96 (318) .49 (549) .08 (1139) .20 (7367) 3 .46 (1503) .48 (1346) .78 (3277) .10 (1844) .68 (663) .74 (216) 1.26 (488) .91 (560) .14 (6516) 4 .46 (1321) .36 (1210) .25 (2921) .21 (1441) .64 (511) .47 (132) 1.20 (431) 1.36 (158) .10 (5679) Notes: No. of observations in parentheses; Bold = significant (p < .05), NS = Not significant (p > .05); n.a. = not applicable 1) First differences are used to ensure that time series are integrated by the same order.
54
Appendix References
Ailawadi, Kusum L., Donald R. Lehmann, and Scott A. Neslin (2003), “Revenue Premium as an Outcome Measure of Brand Equity,” Journal of Marketing, 67, 1–17.
Brav, Alon and J.B. Heaton (2002), “Competing Theories of Financial Anomalies,” Review of Financial Studies, 15 (2), 575–606.
Carhart, Mark M. (1997), “On Persistence in Mutual Fund Performance,” Journal of Finance, 52 (1), 57–82.
Fama Eugene F. (1970), “Efficient Capital Markets: A Review of Theory and Empirical Work,” Journal of Finance, 25 (2), 383–417.
——— and Kenneth R. French (2006), “The Value Premium and the CAPM,” Journal of Finance, 61 (5), 2163–2185.
Granger, Clive W. J. (1969), “Investigating Causal Relations by Econometric Models and Cross-spectral Methods,” Econometrica, 37 (3), 424–438.
Greene, William H. (2011), Econometric Analysis. 6 ed. New Jersey: Prentice Hall.
Mizik, Natalie (2009), „Assessing the Total Financial Performance Impact of Marketing Assets with Limited Time-series Data: A Method and an Application to Brand Equity Research,” MSI Report No. 09-116, 1–23.
Simon, Carol J. and Mary W. Sullivan (1993), “The Measurement and Determinants of Brand Equity: A Financial Approach,” Marketing Science, 12 (1), 28–52.
Srinivasan, Shuba and Dominique M. Hanssens (2009), “Marketing and Firm Value: Metrics, Methods, Findings, and Future Directions,” Journal of Marketing Research, 46 (3), 293–312.