Transcript De Prijs als wapen - Universitat Pompeu Fabra

Department of Marketing
Faculty of Economics
Store location:
Evaluation and Selection based on
Geographical Information
Tammo H.A. Bijmolt
Joint project with:
Auke Hunneman and Paul Elhorst
Department of Marketing
Faculty of Economics
Importance of store location
For many customers, store location is a key
factor driving store choice.
 Store location determines the trade area.
 Store location can be a source of competitive
advantage.
 The decision is almost irreversible  costs of
mistakes are high.

Department of Marketing
Faculty of Economics
Situation:
Chain of stores with many outlets
Important issues:
1. Performance of current outlets
2. Site selection for new outlets
?
Department of Marketing
Faculty of Economics
Modeling framework
1. Current outlets: Determine impact of drivers
2.
of store performance (characteristics of
customers, outlet, and market/competition)
Copy relationships found in stage 1 to new
sites to determine potential performance.
Department of Marketing
Faculty of Economics
Store Characteristics, including:
Location
Size
Consumer Characteristics, including:
Geodemographics
Number of households
Competitor Characteristics, including:
Number of competitors
Retail activity
Main and
Interaction effects
Store Performance
Existing stores
New stores
Department of Marketing
Faculty of Economics
Which consumers?
• Trade area: geographical space from which
the store gets most of its sales.
• Trade area definition: based on travel
distance or travel time of the customers.
 Loyalty cards provide
information on purchase
behavior and residence location
(Zip code) of customers.
 Databases provide
demographic information per Zip
code.
Department of Marketing
Faculty of Economics
Definition of the trade area
Our approach:
1. Rank the ZIP codes on
2.
Store
3.
= Trade area
decreasing sales.
Determine which ZIP
codes yield 85% of the
total sales.
Trade area includes all
these ZIP codes and
those closer to the store.
Department of Marketing
Faculty of Economics
Store revenues
Sales to
members
Sales from members
+
outside trade area
Sales from
zip code j=1
+
+
Sales to
non-members
Sales from members
within trade area
Sales from
zip code j=2
+
Sales from
zip code j=3
+
Sales from
zip code j=4
Trade area
No of HHs
at j=3
x Penetration rate x Avg no of visits x Avg expenditures
at j=3
at j=3
at j=3
Department of Marketing
Faculty of Economics
Model (1)
Van Heerde and Bijmolt (JMR, 2005):
Total sales of a store i in period t can be
decomposed into:
• Sales to loyalty card holders SLit
• Sales to other customers SN
it
Sit  SLit  SNit
Department of Marketing
Faculty of Economics
Model (2)
Sales to loyalty card holders (within the trade
area) can be further decomposed into:
SLit   SLijt  NH jt  PRjt  NV jt  EPjt 
Ji
j 1
NH jt
Ji
j 1
= number of households in zip code area j
i: Store
j: Zip code
t: Time period
PRjt = penetration rate of the loyalty card in zip code area j
NV jt = avg number of visits of loyalty card holders in j
EPjt
= avg expenditures per visit of loyalty card holders in j
Department of Marketing
Faculty of Economics
Example
Households
LC holders
Avg number
of visits
Avg amount
spent
Penetration
Rate
ZIP Code 1
100
75
5
€100
0.75
(75/100)
ZIP Code 2
200
100
10
€75
0.50
(100/200)
Sales ZC 1 = NH*PR*NV*EP = 100*0.75*5*100 = €37,500
Sales ZC 2 = NH*PR*NV*EP = 200*0.50*10*75 = €75,000
Total sales to loyalty card holders = €37,500+ €75,000= €112,500
Department of Marketing
Faculty of Economics
Dependent variables
• Per Zip code:
 Penetration
•
•
of loyalty card (Logit)
 Average number of visits (Ln)
 Average purchase amount (Ln)
Percentage of sales to loyalty card holders
outside the trade area (Logit)
Percentage of total sales to other customers
(Logit)
Department of Marketing
Faculty of Economics
Explanatory variables
Components of the sales equation to be explained by
factors concerning characteristics of:
• Store
Zj predictors that vary
between zip code areas
• Consumer
Xi store specific
predictors
• Market/Competition
N NV
e.g.
ln NV j   NV ,0i   NV ,n Z NV , jn  RNV , j
n 1
K NV
 NV ,0i   NV ,00    NV ,0k X NV ,ik  U NV ,0i
k 1
Department of Marketing
Faculty of Economics
Spatial-lag Random-effects
Hierarchical model
• Relation between ZIP codes that are close to each
other.
• Here, spatial lag specification
R l  λ l WRl  ξ l
• Spatial weight matrix in the error term accounts for
spatial autocorrelation.
• Random-effects Hierarchical model: ZIP codes
nested within stores.
• GLS estimation based on Elhorst (2003)
Department of Marketing
Faculty of Economics
Empirical study
• Dutch chain of clothing retailer
• 28 stores throughout The Netherlands
• Trade area: about 60 to 200 ZIP codes per
•
•
store
3 years (2002-2004)
We have data for each store as well as data
about characteristics of their market areas
(consumer and competitor information).
Department of Marketing
Faculty of Economics
Average sales per store
Average sales per store
2500000
2000000
1500000
unscanned
scanned
1000000
500000
0
2002
2003
2004
Year
About 75% of the sales is by loyalty card holders.
Department of Marketing
Faculty of Economics
Department of Marketing
Faculty of Economics
The relationship between travel distance and
the penetration rate
0,35
0,3
0,25
0,2
0,15
0,1
0,05
0
0
1
2
3
4
5
6
travel distance
7
8
9
10
predicted penetration rate
0,4
Department of Marketing
Faculty of Economics
3,12
3,1
3,08
3,06
3,04
3,02
3
2,98
2,96
2,94
2,92
2,9
0
1
2
3
4
5
6
travel distance
7
8
9
10
predicted number of visits
The relationship between number of visits and
travel distance
Department of Marketing
Faculty of Economics
Department of Marketing
Faculty of Economics
Model predictions: steps
1. Model for explaining revenue components (LP
2.
3.
4.
5.
penetration, number of visits, etc.) based on data
from existing stores.
Model predictions of the revenue components per
ZIP code / store.
Per ZIP code: # households x LP penetration x #
visits x average basket size = predicted revenues.
Aggregate predicted revenues across ZIP codes,
add the percentage sales outside the trade area and
percentage sales to customers without a loyalty card
Final result: Prediction of sales per store, per year.
Department of Marketing
Faculty of Economics

!
Legend
!
Store
new.PENRATE
0.008 - 0.023
0.024 - 0.030
0.031 - 0.036
0.037 - 0.045
Department of Marketing
Faculty of Economics

!
Legend
!
Store
new.VISITS
2.00 - 3.00
3.01 - 4.00
4.01 - 5.00
5.01 - 7.00
Department of Marketing
Faculty of Economics

!
Legend
!
Store
new.EXPENDITUR
0.00
0.01 - 24.00
24.01 - 25.00
25.01 - 27.00
Department of Marketing
Faculty of Economics

!
Legend
!
Store
TOT.SALES
0 - 2062
2063 - 5101
5102 - 9067
9068 - 16055
Department of Marketing
Faculty of Economics
Conclusions
 New methodological tool based on geodemographic and purchase behaviour to assess
store performance.
 We explain a substantial amount of variance in
store performance.
 We identify important drivers of store
performance.
 Drivers differ between penetration, number of
visits and expenditures, e.g. distance and
household composition.
Department of Marketing
Faculty of Economics
Further research
• Predictive validity:
 Predict
•
sales for potential new locations
Comparison to benchmark models