International Journal of u- and e-

e- Service, Science and Technology

Vol. 2, No. 4, December,
December, 2009
2009

The Development of Dynamic Brand Equity Chase Model and Its

Application to Digital Industry Based on Scanner Data

1
Nam Yongsik and 2Kwak Youngsik
1
OpenTide China, Beijing, China
2
Jinju National University
[email protected], [email protected]

Abstract

The purpose of this research is to develop a comprehensive modeling for measuring

dynamics of brand power. We define brand power as brand specific coefficients to yield the
sales volume for each period. The modeling consists of multinomial log it model for each
product category, the brand-specific coefficients, mixture modeling and fuzzy clustering
algorithm. We apply our modeling to TV scanner data in Tianjin China. The results show 5
brands have 12 to 23 times change on their brand power in a year. The lasting time of brand
power spreads from 1 week to 12 weeks.
Keywords: Scanner Data, Brand Equity, China, Appliance

1. Introduction
The research about brand equity is not a new subject. The subject for primary brand equity is
mainly divided into three; study for definition of brand asset (Sung and Bong Woo, 2000;
Jung 1999), study for constitution (Chung and Bak, 2007;Choi and Rhee, 2004), study for
effect (Kim, 1996; Kim, 2008), and so on. If we analyze the previous studies for brand equity
measurement, we regard it as the measurement from the view of stock concept. In other
words, they are only divided into a big or low brand without specific period mentioned. As a
result, we are likely to find out in a newspaper that that Coca-coloa has the biggest brand
equity and Samsung has the biggest among Korean companies.
Have brand equity and brand power been changed when time goes by? If changed, what
should a marketer do? If a marketer knows the period that brand power is more vulnerable
and instead price sensitivity is high, price cut as a marketing tactic is likely to be employed
instead of advertising. On the other hand, if there is a period that brand power is strong and
price sensitivity is low, price cut is unnecessary. Accordingly, if a marketer can specifically
measure a certain term that brand power is high or low, it can a guideline for him/her to
practically implement effective marketing budgets.

This study is to gauge brand equity as flow concept rather than stock concept that previous
researchers have carried out, that is, we would like to show dynamic forms with evidence that
brand equity has been fluctuated during a certain period. As a consequence, this makes
academic contribution by presenting concrete evidence that brand power has been
dynamically changed. Besides, at practical fields marketing activities can be adjusted by
capturing brand power is high or small during the period.

29
International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

2. Modeling
This study takes following steps to measure lasting time of brand power. Firstly, related
variables influencing on sales volume are extracted by using multinomial logit model. In this
process it is proven that besides brand, other marketing variables influence revenue.
Accordingly, the extent which brand influences revenue separate from other marketing
variables is measured by calculating brand-specific coefficients which becomes brand power
influencing revenue during a certain period. Additionally, this modeling reflects competitive
condition by simultaneously considering both brand and price of competitors.

The second step is to adapt mixture regression modeling to measure brand power extracted
from multinomial logit model time to time. In case of adapting the equation of multinomial
logit model into mixture regression model, some sub-equations can be measured and it is
expected the highest and lowest brand equities are extracted during a research period.

The third step is to calculate post hoc probability to classify dynamic changes in brand-
specific coefficients of the second step. Through that brand equity is classified high and low
according to seasons, at which time researchers utilize fuzzy clustering algorithm.

2.1 step 1: multinomial logit model

The multinomial logit model has been used primarily to examine customer choice behavior,
although it has also been used to examine various market phenomena such as market share
forecasting, store selection behavior, price cut effect on a market, market segmentation,
choice set change, and brand switching (Bucklin, Gupta, & Han, 1995; Buckiln & Lattin,
1991; Hardie, Johnson, & Fader, 1993; Gupta, 1988). The majority of these studies have
utilized non-durables based upon scanner data.

If individual i confronted with a choice from a set, Ci, of alternatives, utility can be
expressed as follows, where alternative k is one of the alternatives Ci;

Uik = Vik + εik (1)

Vik = uik + βXik (2)

where Vik = a deterministic component of i’s utility

εik = a random component of i’s utility
uik = an intercept for brand k
= brand specific coeffient for brand k
β = a vector of coefficients for variables X.

Both marketing variables and evaluative criteria are included in the Xik vector. The εk are
independently distributed random variables with a double exponential (Gumbel type II
extreme value) distribution.

P(εk,≤ ε) = exp[-exp(-ε)]
- ∞ < ε <∞ (3)

We assume that individual i chooses the one with the highest utility among the alternatives.

30
 International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

pik = P{ Uik ≥ Uij, j ∈ Ci} (4)

Given assumptions (1)-(4), the conditional probability of choosing brand j can be expressed
by the multinomial logit model (k=1, 2,..., m) as follows;
m
pik = exp(Vik ) / ∑ exp(Vim ) (5)
k =1

where Pik = the probability of choosing brand k.

This expression is known as the multinomial logit (Ben-Akiva & Lerman, 1993; Guadagni
& Little, 1983).
Where, uik is an intercept for brand k or brand-specific coefficient for brand k. This is a
brand power for brand k for a given time horizon as a stock approach.

2.2. Step 2: mixture regression model

On Chusok (Korean Thanksgiving Day) and the Lunar New Year’s Day, customers feel
different about brands or promotion. So if we construct multinomial logit model that is based
on the sales database of the year and make just one price response function or sales promotion
response function on the basis of it, we are likely to neglect the different aspects of dull and
active seasons, promotion and non-promotion seasons, normal and abnormal seasons, and
high-demand and low demand seasons. Furthermore, we are likely to neglect the different
brand power for a given time horizon.
To reflect differences in customers’ responses to brand, price and promotion in these
different seasons, we applied mixture modeling to the multinomial logit model. This process
will generate the several brand-specific coefficients.
Mixture model is to unmix the sample, to identify the segments, and to estimate the
parameters of the density function underlying the observed data within each segment.
Therefore in the first stage of multinomial logit modeling, we find the most appropriate
brand-specific coefficient for brand k for whole time horizon. Then in the second stage, we
apply the mixture model within the multinomial logit model and divide it into several brand
specific coefficients that explain differences by time to time. Therefore the brand specific
coefficient of the first stage can be seen as an aggregated brand specific coefficient, and that
of the second stage, as a disaggregated brand specific coefficient.

2.3. Step 3: Post hoc Probability

Some brand-specific coefficients separate from the second step just shows that brand equity
the highest, lowest, or middle. Therefore, we adapt fuzzy clustering algorithm to capture
seasons at which brand brand is the highest, lowest and middle. Then, we are capable of
matching which brand specific coefficients of the second stage are related to a given time
period.

3. Application to Digital Industry

3.1 Sample: scanner data

31
International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

The weekly scanner data for 32-inch TVs from one of major retailers in China is used with
the period from Jan. 2008 to Dec. 2008 in Tianjin, China. Major five brands’ price and sales
volume data has been used.

3.2 Our model

The multinomial logit model will be employed to estimate beta coefficients for the model
with dependent variables equal to weekly sales volume and independent variables of brand
name and price.

Vik = uik +  pikpricek (5)

where Vik = utility assigned to brand k by consumer i

uik = the intrinsic utility/value of brand k for consumer i (brand-specific intercept)


pricek = the net available price of brand k for consumer i
pik = the parameters to be estimated for consumer i

4. Findings and Future Study

In the first place, the results of multinomial logit model for whole observations of this study
are shown as Table 1. For the study Brand 1 appears the highest brand power, and then Brand
3. Brand 5 has the lowest brand power.
Secondly, the results applied a mixture model are showed at Table 1 under condition that
entire observations are kept on the form of multinomial logit model. Brand 3 in segment 1 has
the highest brand power, Brand 1 in segment 2 has the highest, and again, Brand 3 in segment
3 shows the highest. We define brand power as brand specific coefficients to yield the sales
volume for each period.

Table 1. Aggregate and disaggregate brand-specific coefficients

The first Stage The second stage

Segment 1 Segment 2 Segment3
Intercept 332.8 72.6 205.2 1702.0
Brand 1 39.8 7.2 26.3 132.9
Brand 2 11.3 4.6 -4.6 89.4
Brand 3 34.9 11.9 17.8 276.8
Brand 4 -27.1 -3.1 8.5 -221.5
Brand 5 -58.8 -20.7 -48.1 -277.5
Price -226.5 -32.8 -106.2 -1237.4
R2 0.11 0.43 0.42 0.55
0.88

According to segments it is founded out that brand power is a different order. In segment 1
brand power is brand 3, brand 1, brand 2, brand 4, brand 5 in order, while in segment 2 brand
powers is brand 1, brand 3, brand 4, brand 2, brand 5 in order. Segment 3 has the same order
of brand power as segment 1, however the extent of that is different.
.

Table 2. Post hoc probabilities for Brand 1 for each week

32
 International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

week Segment 1 Segment 2 Segment 3 Classification to

segment
2008. 01 0.0% 0.0% 100.0% 3
2008. 02 0.0% 99.9% 0.1% 2
2008. 03 0.0% 99.4% 0.6% 2
2008. 04 0.0% 0.0% 100.0% 3
2008. 05 0.0% 0.1% 99.9% 1
2008. 06 0.0% 99.5% 0.5% 2
2008. 07 0.0% 98.6% 1.4% 2
2008. 08 0.0% 99.8% 0.2% 2
2008. 09 0.0% 99.7% 0.3% 2
2008. 10 0.0% 99.9% 0.1% 2
2008. 11 0.0% 99.7% 0.3% 2
2008. 12 48.0% 51.9% 0.0% 2
2008. 13 57.6% 42.3% 0.1% 1
2008. 14 56.5% 43.5% 0.0% 1
2008. 15 98.3% 1.7% 0.0% 1
2008. 16 97.2% 2.8% 0.1% 1
2008. 17 96.8% 3.1% 0.1% 1
2008. 18 0.0% 0.0% 100.0% 3

brand1

3.5

2.5

2
brand1
1.5

0.5

0
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52

Figure 1. Volatility of Brand 1 for a year (x axis: week, y axis: brand power
level)

In individual brands changes in brand power are appeared by segment. For instance, in case
of Brand 1 it has the highest brand power in segmentation market 3, yet has the most poor
brand power in segment market 1.

33
International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

Thirdly, it is process through distribute to know which equitation among each multinomial
logit model of the second stage is appropriate to which week. The following Table 2 shows
some results of Brand 1 in the first half of the year 2008
For example, in week 1 of 2008 brand 1 displays that revenue is calculated by equitation
of segment 3. The equitation of segment 3 is no included in week 2 of 2008, but explains
segment 2, which shows revenue should be calculated. In other words, brand power of Brand
1 in segment 3 has been just lasted 1 week, yet that of Brand 1 in the segment 2 has been just
maintained 2 week later on.

3.5

2.5

brand1
2 brand2
brand3
1.5 brand4
brand5

0.5

0
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52

Figure 2. Volatility for each brand ((x axis: week, y axis: brand power level)

If the process above is coupled with rest 4 brands, it is as Figure 2. At figure, periods brand
power is high or low are different in each brand. Consequently, marketing tools marketers of
each brand employ are totally different by time to time
Technical difference about lasting time of brand power of each brand is existed. At Table 3
Brand 1 is more frequent than others in terms of long-standing brand power. On the other
hand, it is rare to become the lowest brand power. In addition, with respect to changing
frequency of brand power Brand 3 is large, while that of Brand 5 is small, especially Brand 5
often lasts low brand power.
The outcomes earned by designing this dynamic brand equity chaser model can be
summarized as follows. First, different brand powers are derived by different aspects of high-
demand and low-demand seasons and active and dull seasons. Marketers can trace the brand
powers that appear differently in a low-demand season and in a certain period that all
promotion concentrates.
Secondly, different brand powers are derived by regions in China. Therefore, brand power
chaser model should be made for each region.
Thirdly, different brand powers are derived by size and type even in the same group of
product. For example, as for the sizes of TV like 32 inches or 40 inches, different degrees of
brand loyalty and different price and promotion elasticities are derived in the same market.
Therefore, brand power chaser model should be applied to product line as well as product
category.

34
 International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

Table 3. the brand dynamics of brand power

Brand 1 Brand 2 Brand 3 Brand 4 Brand 5

Low level (1) weeks 30 41 33 31 42
Middle level (2) weeks 17 8 15 19 8
High (3) level week 5 3 4 2 2
No. of brand power level 18 13 23 20 12
change

References
[1] Bucklin, R. E., Gupta, S., & Han, S. (1995) A Brand’s Eye View of Response Segmentation in Consumer
Choice Behavior. “Journal of Marketing Research”, 32(Feb.). pp. 66-74.
[2] Bucklin, R. E. & Lattin, J.M. (1991). A Two-State Model of Purchase Incidence and Brand Choice,
“Marketing Science”, 10(Winter). Pp. 24-40.
[3] Hardie, B.G., Johnson, E.J., & Fader, P.S. (1993). Modeling Loss Aversion and Reference Dependence Effects
on Brand Choice. “Marketing Science”, 12 (Fall). pp. 378-394.
[4] Kim, H. (2008). A Study on the Effects of Brands on Products Purchase and Choice. “The Korean Society of
Management Consulting”. pp.155-177
[5] Jung, H. (1999). Building Brand Equity Through Advertising and Evidence Utilization (Implications for Brand
Extension Strategy). “Industrial Relations Research”. pp.158-171
[6] Chung, K. and Bak, G. (2007). Antecedents and Consequence of Brand Personality and Brand Equity on
Private Brand. “Korea Research Academy of Distribution Information” (May). pp.97-120
[7] Kim, Y. and Park. Y. (1996). A Conjoint Measurement of Brand Equity with Internet Effect. “Korean
Academic Society of Business Administration” (May). pp.61-96
[8]Sung, Y. and Woo, S. (2000). Psychological Approach to Brand Equity: Interaction of Consumer and Brand,
“Korean Journal of Consumer and Advertising Psychology”. pp. 39-61
[9] Choi, Y. and Rhee, E. (2004). Dimensions of Brand Equity of Luxury Fashion Brands. “Journal of the Korean
Society of Clothing and Textiles” (March.). pp.1007-1018

35
International Journal of u- and e-
e- Service, Science and Technology
Vol. 2, No. 4, December,
December, 2009
2009

Authors

Dr. Yongsik Nam

Director at Samsung OpenTide China

Prof. Youngsik Kwak

Associate Professor at Dept. Venture and Administration in
Jinju National University

36