SLIDE 1 Understanding Price Variation Across Stores and Supermarket Chains Stores and Supermarket Chains
Some Implications for CPI Aggregation Methods
Lorraine Ivancic Kevin J. Fox
SLIDE 2 Aggregation: Study Motivation
Availability of scanner data: highly detailed information
Price indexes constructed with scanner data volatile Price indexes constructed with scanner data volatile Reinsdorf (1999): Need some aggregation to dampen volatility Many different ways to aggregate Very little guidance in the literature about appropriate aggregation methods
SLIDE 3 Aggregation and Unit Values
Aggregation: calculate average price and total quantity across some unit
- eg. time, items, stores
- eg. time, items, stores
Aggregation of prices → ‘unit value price’ Unit value is appropriate when items within the aggregation unit are Question: ‘ when is a commodity (group) – that is, a set
- f economic transactions , sufficiently homogenous to
warrant the use of unit values? ’ (Balk, 1998)
SLIDE 4
Homogeneity
Focus of this work – test for homogeneity across stores within a chain and across supermarket chains. chains. In particular: If the same item is found in different stores which belong to the same supermarket chain should we consider the item to be homogenous? And… If the same item is found in a different supermarket chain should we consider the item to be homogenous?
SLIDE 5 Homogeneity
Economic theory: higher degree of competition and lower degree of item differentiation → equalisation of prices across sellers But, price dispersion may exist if different sellers offer But, price dispersion may exist if different sellers offer different range of auxiliary services to consumers
- eg. different opening hours, range of items, service
Price of the item now reflects a ‘bundle’ of attributes, including item and service attributes Consumer not only buying item also buying level of auxiliary service Same item is NOT homogenous across sellers if bundled with different level of service
SLIDE 6
Defining Homogeneity
Definition Definition The same item sold by different sellers is viewed as homogeneous if the price of the item is found to be consistently the same across sellers in the long term.
SLIDE 7 Scanner data set provided by ABS Data collected by A.C. Nielsen Period covered: 02/02/97 – 26/04/98 (65 weeks)
Data
Period covered: 02/02/97 – 26/04/98 (65 weeks) 110 stores, 4 supermarket chains Stores account for approx. 80% of supermarket sales in Brisbane Additional information: brand name, item weight, description, EANAPN (unique identifier for each item) Item category: coffee
- 436,103 weekly observations
SLIDE 8 Descriptive Statistics
Stores
%
items sold in each chain
weekly obs
chain Chain A 26 20 89 89,320 22,381 Chain B 9 4 101 29,155 8,063 Chain C 34 35 123 162,765 41,853 Chain D 41 41 88 154,863 38,953 TOTAL 110 100 157 436,103 111,250
SLIDE 9 Hedonic Regression Model
Hedonic time dummy regression model:
β β β + + + =
∑ ∑
Pti = price of item i in period t Dt = time dummy variables, 1…t Ztki = k characteristics of item i in period t WLS used (expenditure shares) Monthly observations used
β β β + + + =
∑ ∑
= =
SLIDE 10
Coffee Characteristics
Product Brand (25 brands, DV) Decaffeinated (DV) Additional flavouring (DV) Additional flavouring (DV) Bonus (DV) Espresso (DV) Freeze Dried (DV) Product weight (20 weights, spline) Piecewise linear continuous function 7 breakpoints allowing for changes in slope Stores in each supermarket chain (DV)
SLIDE 11
Testing for homogeneity across stores within a chain
Test hypothesis of equal prices across stores within a chain chain Separate regressions run for each chain using relevant store DV’s Tested store DV coefficients across pairs of stores
SLIDE 12 Results: Aggregation across stores within a chain
Chain A Chain B Chain C Chain D Chain A **** 0.0307 (0.115) 0.0374 (0.053) 0.0323 (0.115) (0.115) (0.053) (0.115) Chain B **** 0.0067 (0.711) 0.0017 (0.931) Chain C ****
(0.8005) Chain D **** Chain dummy variable coefficient estimates (P,values in brackets)
SLIDE 13 Results: No aggregation across stores within a chain
Chain A Chain B Chain C Chain D Chain A **** 0.0313* (<0.01) 0.0343* (<0.01) 0.0326* (<0.01) (<0.01) (<0.01) (<0.01) Chain B **** 0.0030 (0.585) 0.0013 (0.821) Chain C ****
(0.615) Chain D **** Chain dummy variable coefficient estimates (P,values in brackets) * Indicates significance at 1% level
SLIDE 14
Results: Store Differences within a Chain
Chain A (26 stores): No significant price differences found in the 325 store comparisons found in the 325 store comparisons Chain B (9 stores): 0 out of 36 comparisons Chain C (34 stores): 8 out of 561comparisons (1.4%) Chain D (41 stores): 61 out of 861 comparisons (7.0%)
SLIDE 15
Results: Implications
Seems reasonable to aggregate across stores in chains A, B and C chains A, B and C Only two stores in Chain D which seem to have different prices consistently – seems reasonable to aggregate across all other stores in D
SLIDE 16
Homogeneity Based on Hedonics
The same item is considered homogeneous if in stores in Chain A → unit values for same item in stores in Chain A → unit values for same item in stores in Chain A Unit values for same item in stores in chains B, C, and D, except for two stores in D. The two Chain D stores enter index number calculation separately
SLIDE 17
Index Numbers
4 different indexes used
Laspeyres, Paasche, Fisher, Törnqvist
4 different types of aggregation used
Aggregation of items across chains Aggregation based on hedonics No aggregation of items across chains No aggregation of item across stores
SLIDE 18 Index Number Estimates (Base = 1)
Homogeneity Over Chains Homogeneity Based on Hedonics No Homogeneity Over Chains No Homogeneity Over Stores Laspeyres 1.263 1.380 1.518 1.564 Paasche 0.985 0.903 0.808 0.793 Fisher 1.115 1.116 1.107 1.114 Törnqvist 1.115 1.118 1.110 1.116 Over 15 month period
SLIDE 19 Findings
Aggregation implications:
Hedonic regression may be useful to study the issue of homogeneity and associated issue of how to aggregate homogeneity and associated issue of how to aggregate This work shows fair amount of aggregation may be justified, leading to less volatile indexes with scanner data
Sampling implications:
Sampling from one store in Chain A or B may be enough to
- btain representative prices
Not the case for chains C and D
