The impact of competitive interactions on category
penetration and purchase frequency of mature FMCG
categories
Laurent Rochat
Consultant, Rue de Grenus 12, 1201 Geneva, Laurent@ innovationatelier.com
Ann Van Ackere
Faculté des Hautes Etudes Commerciales (HEC), Université de Lausanne, 1015
Lausanne, Ann.V anA ckere@unil.ch
Many fast moving consumer goods (FMCG) categories like laundry detergents, diapers
or cereals are mature and rather stable, so the interest is high for manufacturers to
make diagnostics on the category and their brand to find growth opportunities. Attempts
to model mature category dynamics have mostly been based on probabilistic models.
The most famous one is the “NBD-Dirichlet” model, which was first presented in 1984
and has subsequently triggered significant research in the area. The model has
limitations as it assumes stable marketing and promotional activity and stable category
dynamics. This paper uses a system dynamics model to relax some of these stability
assumptions and explain how competitive dynamics impact the total category
penetration, purchase frequency and volume size over time.
Key words: FMCG category dynamics; competitive interactions; impact of promotions;
brand choice; buyer behavior; consumer purchasing; NBD- Dirichlet theory.
Background
Over the past decade, many fast moving consumer goods (FMCG) categories like
laundry detergents, diapers or cereals have reached a plateau; although the average
transaction size is increasing, the actual consumption has remained mostly unchanged,
penetration is stable, an total category volume size even eroding. The consumer
purchase dynamics explaining these observations from household panels (HHP) data are
still not very well understood. New insights in this area could tremendously help FMCG
business managers to shape brand strategy and identify growth opportunities.
FMCG category dynamics are well described through mathematical models like Fourt-
Woodlock (1960), NBD- Dirichlet (Goodhardt, Ehrenberg & Chatfield 1984), Eskin
(1973) or Kalwani & Silk (1980). These models are incredibly helpful to pose a
diagnostic on a specific category or brand, but offer limited insights to understand the
underlying factors influencing category dynamics without additional mathematical
complication. Their low applicability for brand strategy elaboration has been
highlighted by Ehrenberg, Barnard & Sharp (2000). System dynamics is better suited to
explaining the complexity of category dynamics. Though somewhat criticized with
regard to validity and validations (Forrester & Senge 1980, Martinez & Richardson
2001), system dynamics does not seem to carry lower objectivity than mathematics-
based sciences. The latter models actually make parameters interpretation more difficult
(Fader & Hardie 1999), while system dynamics models leverage simple bivariate
relationships, have the ability to produce a wider range of mathematical curves as
output and can be more accurate for forecasting purposes (Lyneis, 2000).
There are some system dynamics based purchase behavior models which help to
understand new product introduction dynamics, but they tend to be either too simple
(Sterman 2001, Morecroft 2007) or too broad for practical use for marketers (Maier
1998). Also, they focus on the diffusion of new products and do not explain the
evolution of mature categories over time.
The model presented here aims to provide a better representation of the market
dynamics of a mature consumer goods category. In particular, our model aims to
capture how competitive activity, market share objectives (Armstrong & Green 2007)
and promotion support (Bell, Chiang & Padmanabhan 1997) lead to this evolution. To
our knowledge, this model is the first one to link consumer consumption, category
penetration and purchase frequency to capture mid-term and long-term category
evolution.
Model description
The starting point to the model is the “NBD-Dirichlet” theory for mature, stable
categories. The “NBD-Dirichlet” model uses a negative binomial distribution (NBD) for
consumers’ purchase distribution of a given FMCG product category, and a Dirichlet
distribution to integrate brand choice to the purchase dynamic. The core assumptions of
the model can be summarized as follows:
(1) Each household is either a stable product category user or a non-user. The split
between both groups is given by category penetration (B). A glossary of
household panel metrics and concepts is available in Appendix 2.
(2) The probability of purchase by a category user follows a Poisson process with the
purchase frequency (wp) as parameter. The average purchase frequency for the
category across all category users is denoted by W.
(3) The distribution of wp across households follows a Gamma distribution, with
parameters K and A which can be computed from Household Panel (HHP) data. A
useful tool to find K and other “NBD-Dirichlet” parameters using VB and Excel
has been made available by Z. Kearns (2002). It can be downloaded from
http://marketing-bulletin.massey.ac.nz/ (Dirichlet VB.xls link)
(4) The choice of each brand at any purchase act follows a multinomial distribution,
with a probability assigned to each brand and the sum of probabilities totaling 1.
The conjunction of assumptions (2) and (3) is called a Gamma- Poisson mixture and
gives a negative binomial distribution (NBD).
While the he “NBD-Dirichlet” model is a good predictor of aggregated category
dynamics and brand purchase, two of its core hypotheses make it impossible to capture
the drivers of some dynamics observed in consumer behavior:
I. The category penetration is time-invariant in the “NBD-Dirichlet” model. This
hypothesis is not confirmed in practice; category penetration can evolve over time.
Il. The purchase frequency is modeled using a Poisson process, which is a discrete
process. Consequently, households who do purchase the product cannot have a
purchase frequency lower than once per year. This assumption may look trivial at
first, yet it fixes an unnecessary lower bound for light buyers and has an impact on
the evolution of category penetration.
Using a system dynamics model, we remove these hypotheses and are thus able to
explain the drivers of category penetration evolution and category purchase frequency
evolution. An overview of the full model structure can be found in Appendix 1.
Asa starting point to build the system dynamics model, we use three mathematical
properties of the “NBD-Dirichlet” model to tackle the limiting hypothesis as follows:
(i) The algorithm to determine A and K:
-
B=1- G+ayk (1)
This is based on Goodhardt et al. (1984), where A is the scale parameter and K the
shape parameter of the negative binomial distribution (NBD) for category purchases.
The parameters capture households’ purchasing heterogeneity. A more detailed
description of the parameters used in the “NBD- Dirichlet” model can be found in
Carl Driesener (2005).
(ii) The relationship between the parameters A and K:
A= ed (2)
A isa function of the annual buying rate per household in the total population
(W*B) and of K.
(iii) The linear relationship of the scale parameter A with time:
Ap=T*#A (3)
HHP data provides annual values for B and W, so K and A can be computed for the
category using equations (1) and (2). Changing A to Ar to reflect a longer time period as
per equation (3) and feeding the results back into equation (1) while keeping K constant,
we can determine the % evolution of the penetration over periods longer than 1 year.
As the penetration computed over a longer time period exceeds the 52-week penetration
level, we refer to the difference as the excess penetration. This excess penetration comes
from households who buy the category less than once a year. For instance, consumers
buying once every two years would only be included in the annual penetration every
other year. We refer to these consumers as light buyers and denote the excess
penetration by B,. This interpretation is compatible with the “NBD-Dirichlet”
hypothesis that the annual penetration is stable: the number of light buyers who
purchase during a 52 week period remains constant.
The dynamic model divides the light buyers representing the excess penetration B, into
two groups:
1. Stable light buyers. These are consumers that are committed to the category
(category users) but consume little and only need to purchase the category less than
once a year. These consumers are considered as stable users not driven by
promotions; they will buy the category regardless of promotions, though they
would buy on promotion if there is one in the store the day they shop the category.
2. Opportunistic new buyers. New buyers that are exclusively buying on promotion.
Ina stable category with stable promotional activity, the percentage of these
promotion/impulse driven consumers is assumed stable and total category
penetration (B) is stable. The dynamic model here considers this group of
consumers separately in order to understand how changes in the promotional
activity in the category affect category penetration and purchase frequency.
Looking at these two groups independently is key to assess the promotional impact on
overall category penetration. The overall dynamic of these two groups and their relation
to promotional activity is shown in Figure 1, which provides a simplified view of the
model for visual convenience.
The increase in the 52W Rolling User base is determined by the light buyers (buying
regardless of promotion) and the opportunistic new buyers (buying through
promotions). These will leave the user base 52 weeks later (Lapsed Users). The
proportion of lapsed users under stable conditions is calibrated using the “NBD-
Dirichlet” theory and is a fraction of yearly B, (split linearly over 52 weeks).
W Proportion of
opportunistic new Bae
buyers a = W Proportion on +———W Proportion of lapsed ~~
_ Es Sable ight ayers Acets fom NBD:
Users SN <fotal NBD-Ditichlet-
~ \ stable category user
ee Y
\ 52W Rollin
. °. nonusers” SW Rolin User base |g — pr base
\ Lapsed Users| W Increase in Lapse, Uses " |
over year User base
Pax \
~~ \
SN __ Total HHs 52W MAT
\ Total WD W Pew
= romo users
Penetration Ps
promotions ,
,
Proportion of promo
~~ purchases that are repeat.
Promotions meat purchase
\ Impact Index
A -~
W forward repeat
~ promoted
W replacement
repeat
Figure 1. The user base dynamic.
The impact of promotions on the W proportion of opportunistic new buyers is derived
from a promotions sales impact index which is considered as a known input (it can be
found from econometric trade promotions analysis done by IRI, Nielsen ...etc.). The
model splits the promotion sales impact index into transaction size driven volume, extra
forward purchase acts (W forward repeat — promoted from stable category users) and
extra trial acts (from opportunistic new buyers). The proportion of trial purchases from
promotions is negatively impacted by the level of category penetration; the less
available users there are, the less new buyers there can be. This is a stabilizing feedback
loop in the model (similar to diffusion curves).
Figure 2 shows the simplified dynamic of repeat purchases and promotions.
Total Product stock
nae stable category users Gea
—™ mption
W avg VPP~ epeal ee \ Wav " »~
stable can promote ted Total HHs / Consumption Cat of \
\ | users .
| Uninfluenced | < iHs> |
| VPP \ gy
Proportion of Promo VPP Product at home NBD-Dirichlet -
influenced purchases - 1 amongst stable «stable category user
stable users W repeat - 4— category users base
4 unpromoted
\ "Se, Buyers Purchase W Z Value
4 ae W opportunistic an one ©” Consumption
repeat - promoted «_
Proportion of |
Influenced Purchases Se —— N
re Se. ~W replacement
i ~ Total WD repeat
= Promotions sales promotions
‘W repeat - Impact Index i]
promoted ~_
__W forward repeat «—~
- promoted
Figure 2. The repeat purchases dynamic.
Every week, a certain proportion of category buyers go to the store and purchase the
category to keep a certain level of stock at home. These purchases are repeat purchases
as they occur amongst households that have already bought the category in the past.
Only stable category users are taken into account for repeat purchasing and their
penetration (B- B,) is invariant even under unstable category dynamics. The dynamics
of stable users’ repeat purchases (W replacement repeat, bottom right of figure 2) are
based on the average product they have left at home (Product at home amongst stable
category users), which depends on their consumption and the average transaction size
(W avg VPP stable users) of their purchases (figure 2).
Product at home amongst category users triggers repeat purchases when a certain
threshold is reached, under which they will purchase the category to keep an acceptable
stock of product at home and continue consuming. The proportion of consumers that
fall below the threshold and therefore make a category purchase in a given week is
assumed to follow a normal distribution for modeling simplicity. The variance of the
normal distribution is derived from the “NBD-Dirichlet” purchase acts distribution and
adjusted for average consumption using equations (4) and (5):
ao? (x) = K * M2 (4)
a(ax) = lal * (x) (5)
Where K is the “NBD-Dirichlet” parameter and M is given by W*B. Note that despite
this modeling approach, the model remains deterministic: distributions are used to
predict replacement repeat purchases without uncertainty.
In most mature categories, manufacturers impose larger sizes in promotions to load
consumers with their product (in our beer example, the promotion would be a 10-can
pack compared to the usual 8-can pack). This strategy works well for manufacturers as
they can offer a higher absolute deal in promotion (i.e. a bigger monetary discount,
which is more interesting for consumers) and the bigger size mechanically increases
their brand’s volume share of requirement.
We assume consumption to be stable. This implies that category users buying bigger
sizes build product stock and do not need to shop for the category as early as if they had
bought their regular size (it will take them longer to reach the threshold at which they
go purchase the category). In fact, there is evidence that consumption is influenced by
pack size (Wansink 1994), but the effect on purchase frequency is small enough to not
be taken into account in the current version of the model.
This is a second stabilizing feedback loop in the model: as promotion purchases
increase, stable users buy a bigger size (volume per purchase, V PP), which increases
their product stock, decreases their natural replacement purchases and consequently
their promotion purchases.
For conceptual convenience, the model limits the transaction size variability to two
options: the uninfluenced (usual) volume per purchase (uninfluenced VPP) and the
promotion volume per purchase (promo VPP).
Finally, the model includes a simple dynamic of competitive interactions that captures
variations of promotional intensity. Promotional dynamics do not reflect a store model
(analyzing where consumers shop) but a category model (analyzing when consumers
buy and how much). Two competitors are considered; Brand A and Brand B, equally
liked by consumers (50% preference share for each brand), and 100% substitutable.
The model assumes that competitors know the impact of promotions on their market
share and can adjust promotions to the exact level so as to reach their target market
share under current conditions — with a certain delay. Promotions are expressed as a %
of weighted distribution. The weighted distribution of promotions is defined as the % of
all stores having a promotion, weighted by store sales of the product category. Weighted
Distribution (WD) takes into account the stores’ importance for category sales, which is
not the case with Numerical Distribution (ND, the % of all stores).
These dynamics are shown in figure 3. Promotions (W Brand A WD promotions) are
modeled as a stock as they take time to be adapted. This assumption is also made for the
sake of consistency; high/low promotion activities over short time periods would be
more difficult to capture in terms of decision making and would induce unnecessary
fluctuations in the model which would confuse rather than help us in understanding the
link with category penetration and purchase frequency evolution.
WD promotions
_-— impact on sales ™~ Total WD
7 % consumers Promoting
Proportion of preferring brand B .
Influenced Purchases = ‘
% consumers ew ‘ime = promotions>
preferring brand A soiaine shares SSS.
| ——— —~. oo \
_—— Gap between targetand =~ \
1 fa Gap between target -~_ tual share Brand B \
W Brand __, and actual share Brand Sea ae \
volume share A Ses f WD promotions
f WD promotions Tenpt mole change target for
Target volume change target for share B mae i
Saw ne \ Wands
l W Brand A Brand B promotions i
& WD and B pi promotions
Brand A promotions |_promotions implementation
implementation a
Se 1p wee
ae
Figure 3. The competitive promotional dynamic
Figure 3 shows the competitive promotional dynamics. These promotional dynamics
induces an oscillation of market shares and levels of promotions (weighted distribution
of promotions) of Brands A and B. Equilibrium can be achieved if both competitors
have a volume share objective equal to the natural consumer preference and their
relative promotional pressure is equal to their relative market share.
The model admits that each competitor can increase his level of promotions to cover up
to 100% of the stores. A different upper limit could be imposed in the model to take into
account financial constraints or stores acceptance of extra promotions; however the
model would show the same dynamics, so we decided to keep 100% as limit for
promotions WD.
We assume that promotions of each brand are distributed independently across stores,
so they can be found in the same store at the same time, in which case consumers buy
brand A or B according to their natural preference ratio.
Model calibration and validation
The model is calibrated using publicly available household panel (HHP) data for the
Beers/Ale/A lcoholic Cider category from the IRI Marketing Data Set of 2008: category
penetration, purchase frequency and the volume sold on deal (VSOD), ie. the
proportion of total volume sales that is sold on promotion.
The promotion impact index has been set to deliver a VSOD in line with panel data. The
consumption rate and the purchase threshold have been calibrated to reach model
equilibrium (the steady state corresponding to the “NBD- Dirichlet” model).
In addition, the following numerical values are used for calibration:
1. The population of the theoretical case has been set to 10,000,000 households.
The proportion of trial purchases that come from non-promotion driven buyers
(stable “NBD-Dirichlet” base in the model) has been set to 80%.
3. The transaction sizes have been set to 2.64 liters for regular (unpromoted)
purchases and 3.3 liters for promoted purchases.
4. The promotions are initialized at 10% of the weighted distribution for each brand.
5. The brand preference ratio has been set to 50% for A and B respectively.
Weekly time periods are used, as they represent the shortest measurable period of time
from household panels. The simulation time-step 6t of 4 allows delays as short as 1
week if need be. We run the simulation for a 3-year period (156 weeks), which is the
minimum required to observe the patterns under investigation. Seasonality is excluded
from the model to make interpretations easier. The model has been implemented in
Vensim®.
The model has been validated using Sterman’s twelve model tests (Sterman 2000, 859-
61). Two comments are worth highlighting:
1. Structure assessment: the simplification of competitive dynamics is consistent with
the model objectives.
2. Dimensional consistency: purchases are assumed to be dimensionless as they are
interchangeable with consumers for trial purchases.
All tests provide satisfactory conclusions regarding model robustness.
Results
We consider two scenarii:
1. The equilibrium scenario, which is a more granular application of the “NBD-
Dirichlet” theory and is referred to as the base case.
2. Starting from the base case, we assume an increase in the market share target for
one of the brands (brand A) at time 0.
Scenario 2 shows the competitive dynamic as a competitor (brand A) increases its
market share target and relies on extra store promotions to reach his target. This creates
disequilibrium in the competitive environment leading to an escalation of promotional
activity of both competitors up to the maximum level (see Figure 4). As brand A gains
market share thanks to higher promotional pressure than brand B, the second competitor
increases his promotions to recover the loss, hindering the ability of brand A to build
share with promotions and prompting for more promotions. As promotions are only
constrained by the number of stores in the model, the escalation of promotions goes on
until the maximum of 100% of the stores have a promotion for each brand.
The rise in promotions (figure 4) induces an increase in the category user base, fueled
by the recruitment of promotion-driven buyers with no loyalty to the category. This
shows how the category user base and overall category penetration are inflated by
promotions (figure 5).
Evolution of total category promotions (% WD) Evolution of total user base (Mio households)
100% as
90% 344
80% 33
70% 32
60% 34
50% a
40% 29 f
20% 2a
20% 27
10% 26
0% | 25 |
° 32 104 156 ° 32 308 156
Total WO promotions Weeks 52W Rolling User base Weeks
Figure 4. Total promotions of the category Figure 5. Total user base
In addition, promotions increase the average transaction size of regular buyers, which
builds their product stock (see figure 6) and decreases the natural replacement repeat
rate. The increase in user base due to these “one-off buyers, combined with the drop in
repeat rate amongst stable category users, induce a significant decline of category
purchase frequency over time (figure 7).
Note that in the short term (up to week 6) we observe an increase in purchase frequency,
driven by forward purchases by stable category users due to the additional promotions.
However, this effect is quickly compensated by the increase in user base (“‘one-off”
buyers) and the lower natural replacement repeat purchases, as mentioned above.
Evolution of total product stock (Mio liters) Purchase Frequency (N. of acts per buyer)
a 52 104 156 a 52 104 156
Total Product stock stable category users Weeks Purchase Frequency Weeks
Figure 6. Total product stock Figure 7. Purchase Frequency
The gradual drop of category purchase frequency from week 6 onwards makes
competitors more dependent on promotions; as purchase acts occur less often in the
category, it is more important for competitors to keep a high promotion pressure over
time to divert purchases towards their brand. When promotions are on the rise, Brand A
market share is ahead of Brand B as it is moving first on promotional pressure and
Brand B is only reacting (figure 8). Note that the sum of the Brand A and Brand B
promotion is not equal to the total category promotion (figure 4) because when
computing the total, we subtract the duplicate promotions, i.e. the cases where both
brands are on promotion in the same store at the same time. Duplicate promotions are
close to 0 for low levels of promotion (time 0), and reach 100% when both brands
advertise in all stores (towards the end of the simulation period).
Evolution of brands promotions (% WD) Evolution of brands volume share
100% —— cow
90%
80%
70% om
60%
50% 50%
40%
30%
ee 459%
10%
o% | 40%
° 52 104 4156 ° 32 100 156
Wo promotions: W Brand volume share:
~ Brand A ~ Brand B mies Brand A Brand 6 Mosel:
Figure 8. Brand A and Brand B promotions Figure 9. Brand A and Brand B volume share
Brand A benefits from this first mover advantage until he gets close to the maximum
promotional pressure possible (in week 126, with 97% of weighted distribution). At that
point Brand B is catching up on his promotional disadvantage and gaining market share
back. As Brand B approaches the maximum promotional pressure, his promotional
pressure gets very close to Brand A levels. Both brands lean towards similar
competitive pressure, so promotions-based competitive advantage and market shares
tend towards consumers’ natural brand preference (50% in this case). Market shares
converge to their original levels as of week 126, but with almost all volume sold on
deal. These curves lead to a new equilibrium with equal maximum promotional
pressure, 100% volume sold on deal and 50% market shares. At that point, a drop of
promotional pressure by one of the competitors would result in him losing market share.
Hence, this state is a Nash equilibrium.
10
In addition to market shares variations (figure 9), total category volume shows an
interesting dynamic. As expected, category volume initially grows with the intensity of
promotions in the category. However, category size is not following the evolution of
promotions over the full period; category volume reaches a maximum at week 62 while
promotions keep increasing beyond week 120. From period 62 to period 120,
competitors continue to increase promotions, yet the yearly category volume (see figure
10) is going down in the simulation.
Total category volume sales (Mio liters)
Category Volume Weeks
Figure 10. Total category size (volume)
While this observation is a priori counter-intuitive, it makes perfect sense in a dynamic
context: category volume gains are slowed by the duplication of promotions and extra
brand promotions have a declining marginal impact at category level (a new promotion
for, say, Brand B is likely to be redundant in the store as there is already a promotion
for Brand A). Also recall that stable consumers have increased their product stock
(figure 6) and do not need to purchase the category as often, delaying some purchases.
Thus, looking at the evolution over a 52 weeks moving average, category volume is
decreasing while promotions for both competitors keep increasing.
As the first competitor increases his promotions, his actual market share remains below
his new target as the second competitor is reacting, yet his volume sales are
significantly increasing (see figure 11). However, competitors relying exclusively on
market shares objectives will keep on increasing their investment in promotions despite
increases in their volume sales — even though the actual volume sales get close to their
share objective applied to the initial category size. Market share objectives rather than
sales objectives fuel the escalation of promotions (Armstrong & Green 2007).
Evolution of brands volume sales (Mio liters)
° 82 108 156
W Brand volume sales:
-Brand A — Brand B Weeks
Figure 11. Brand A and Brand B volume sales
11
Interestingly, if the first competitor decreased his market share target back to the
original level, the level of promotions of both competitors in the new equilibrium would
remain well above the original level: the competitive escalation is not reversible if
competitors rely exclusively on market share objectives as in the model.
Conclusions
The proposed model enables us to explain how competitive interactions in store
promotions impact category size, penetration and purchase frequency evolution. Some
restrictive hypotheses of the “NBD-Dirichlet” model can be relaxed through the use of
system dynamics modeling. The model considers three groups of consumers that may
make a purchase in the category: the stable yearly user base, the committed light buyers,
and the opportunistic buyers. Under stability conditions (our base case), the dynamic
model shows the same pattems as the “NBD-Dirichlet” model. However as competitors
enter a promotional fight for market share, this stability condition is no longer fulfilled
and category metrics change: we observe a gain of penetration as more opportunistic
buyers enter the category and a reduced purchase frequency (due to the loading of stable
users with extra stock as well as the extra light buyers that do not repeat their purchase).
The model also explains why even if the promotion intensity is increasing in a category,
the total category volume size can be decreasing. And this shows how market share
objectives can lead to a higher reliance on promotions by competitors, which is healthy
neither for competitors’ financials nor for category growth in the long term.
The model described assumes stable consumption by stable category users; an
interesting extension would be to include the transaction size as well as the price impact
on consumption.
12
Appendix 1
Model description (simplified)
st —_ = ¥ Proportion of W Proportion of | : NBD-Dirichlet - stable
opportunistic new stable light buyers” --—-W proportion of lapsed __- category user base
buyers / ~ users from NBD wt
‘i vA ~~
|
|
52W Rolling 52W Rolling User
inet vas non-users base
over year
\
52W MAT
promo users
*
\\ <i Proportion
t opportunistic new
Penetration uyers> /
~™~
Total Product stock
stable category users 2
‘onsumption
| *
| W Avg Consumption
J - Cat of users
Proportion of'influenced __ȴ BV VPP Product at home
purchases stable users 3 omoted> Ne amongst stable category
{ users
Uninfluenced — | ey
4 VPP \
i Promo VPP
<W Proportion of
pportunistic new W opportunistic
buyers> <Total WD repeat - promoted
B promoti ons> bas
i \ W fora repeat
2 V forward repeat - =~
Wtral- W trial W repeat pomotsl
promoted unpromoted —_unpromoted ee
,/ purchases that are repeat
io Promotions sales « \
Proportion of Impact Index. —
Influenced Purchases © consis <Penetrati
\ preferring brand B
|
% consumers
a
“* W Brand B
prefeming | ;
brand A volume share. Gap between target
Gap between target SI and actual share
ahd gota hare | Brand B
Brand A 7 ‘ VD promotions
oN a
Target volume Brand A \l Tenge von Brand B - :
S\ SurA og J. W Brand A WD se s W Brand B WD
Brand A promotions} promotions Brand B promotions promotions
implementation implementation >
The complete list of equations is available from the author.
13
Appendix 2
Glossary
Penetration (B). The % of households in a particular area that have bought a brand or a
product category over a 52 week period.
Trial. The % of households in a particular area that have bought a brand or a product
category over a 52 week period for the first time.
Repeat / repeat rate. The % of households in a particular area that have bought a brand
ora product category more than once if they have bought at least once over a 52 weeks
period.
Purchase Frequency (W). The average number of purchases of a brand or a product
category that households make over a 52 week period amongst households that have
bought the brand or the product category at least once.
Promotions. A1l promotional activities (leaflets, displays, shelf material ...etc.) that
carry a price discount. The mix of promotional activities is assumed constant over time
in the model.
Weighted Distribution. The % of all stores weighted by store sales of the category.
Promotions sales impact index. Volume sales when and where a promotion takes
place divided by the normal sales level (i.e. without promotion).
Volume sold on deal. The volume sales that was made on promotion divided by total
volume sales, i.e. the % of volume sold on promotion.
Volume per purchase (VPP). The transaction size expressed in consumption units
(liters in the case of the Beers/Ale/A lcoholic Cider category).
Volume share of requirement. The average proportion of total category volume that a
brand represents amongst buyers of that brand over a certain period.
14
References
Armstrong, J. Scott and Green, Kesten C. “Competitor-Oriented Objectives: The Myth
of Market Share.” International J ournal of Business 12 (2007).
Dekimpe, Marnik G., Hanssens, Dominique M. and Silva-Riss Jorge M. “Long-run
effects of price promotions in scanner markets” J ournal of Econometrics 89 (1999):
269-291.
Barlas, Yaman “Formal aspects of model validity and validation in system dynamics”
System Dynamics Review 12 (1996): 183-21.
Bassi, Francesca. “The Dirichlet Model: Analysis of a Market and Comparison of
Estimation Procedures” Marketing Bulletin 22 (2011): Technical Note 1.
Bell D. R., Chiang J., Padmanabhan V. “The 84/14/2 Rule Revisited: What Drives
Choice, Incidence and Quantity Elasticities” INSEAD working paper 97/016/MKT
(1997)
Coyle, Geoff and Exelby, David. ‘The validation of commercial system dynamics
models”. System Dynamics Review 16 (2000): 27-41.
Driesener, Carl. “Empirical Generalisations in the Parameter Values of the Dirichlet
Model — An Examination Across 50 Categories.” PhD diss., University of South
Australia, 2005.
Ehrenberg, Andrew. S. C. “Repeat Buying - Facts, Theory and Applications”. Journal
of Empirical Generalisations in Marketing Science 5 (2000): 392-770.
East, Robert, Wright Malcolm and Colombo Richard. “The Association Between
Penetration and Frequency”. Paper presented at the ANZMAC 2004, Australian New
Zealand Marketing Academy, Victoria University, Wellington, New Zealand, 2004.
Fader, Peter S. and Hardie, Bruce G.S. "Investigating the Properties of the
Eskin/Kalwani & Silk Model of Repeat Buying for New Products," Paper presented in
the 28th EMAC Conference, Humboldt University, Berlin, May 11-14, 1999.
Forrester Jay W. and Senge Peter M. “Tests for building confidence in system dynamics
models” in Legasto Augusto A., Forrester Jay W. and Lyneis James M. Lyneis (eds.)
System Dynamics. TIMS Studies in the Management Sciences 14. New Y ork: North-
Holland: 209-228.
Fourt, L.A. and Woodlock, J. W. “Early prediction of market success for grocery
products”. Journal of Marketing 25 (1960): 31-38.
Goodhardt, Gerald J., Ehrenberg, Andrew. S. C. and Chatfield, Christopher. “The
Dirichlet: A Comprehensive Model of Buying Behaviour”. J ournal of the Royal
Statistical Society 147 (1984): 621-655.
15
Lyneis, James M. “System dynamics for market forecasting and structural analysis”.
System Dynamics Review 16 (2000): 3-25.
Maier, Frank H. “New product diffusion models in innovation management—a system
dynamics perspective” System dynamics review 14 (1998): 285-308.
Martinez, Ignacio J. and Richardson, George P. “Best Practices in System Dynamics
Modeling” Paper presented at the 19" International Conference of the System Dynamics
Society, Atlanta, Georgia, July 23-27, 2001.
Meade, Nigel and Islam, Towhidul. “Modelling and forecasting the diffusion of
innovation”. International J ournal of Forecasting 22 (2006): 519- 545.
Morecroft, John. Strategic Modelling and Business Dynamics: A Feedback Systems
Approach, J. Wiley & Sons Ltd, 2007.
Nijs, Vincent R., Dekimpe, Mamik G., Steenkamps, Jan- Benedict E.M. and Hanssens,
Dominique M. “The Category-Demand Effects of Price Promotions” Marketing Science
20 (2001): 1-22.
Sterman, John D. Business Dynamics, Systems thinking and modeling for a complex
world, McGraw-Hill, 2000.
Sterman, John D. “Systems dynamics modeling- tools for learning in a complex world”
California management review 43 (2001): 8-25.
Sterman, John D. Conference “Why Bad Things Happen to Good Technologies” Video
from January 30, 2008. http://mitworld.mit.edu/video/553/
16
