Substitution between Mass-Produced
and High-End Beers
Daniel Toro-Gonzalez
Ph.D. candidate, School of Economic Sciences (SES)
Jill J. McCluskey
Visiting Professor, Cornell University and
Professor, SES, Washington State University and
Ron C. Mittelhammer
Regents Professor, SES & Dept. of Statistics
Presented at Beeronomics Symposium UC Davis
November 3, 2011
Macro Brews Dominate many U.S.
Markets
2
However, This is Changing
• Mass producers’ market share
still represents the vast majority
of sales, but their sales are flat
or declining.
• Trend of consumers switching
from mass to craft beers.
• Consistent with general shift in
food preferences:
 Increasing
desire for variety,
taste, and local products.
We know that consumers shift from
macro to craft brews. Does it go the
other way?
• “…consumers are very loyal to craft
beers and not shifting to macro
from craft. In economics terms the
cross-price elasticity of craft and
macro brews appears to be very
inelastic, or that beer drinker do
not think of macro lagers as a good
substitute for micro brews.”
-
“Beeronomics: Is Craft Beer Recession Proof After All ?” ,
The Oregon Economics Blog, Thursday, May 7, 2009.
Project Objectives
•Estimate demand for beer, which is a
differentiated product.
•Estimate the own-price, cross-price and income
elasticities.
Data
• Scanner data from 60
Dominick's supermarkets
in Chicago.
• Seven years of store-level
weekly sales data (1991
to 1997)
• 483UPCs for 343 brands.
• Product info and store
area sociodemographics
Market and Product Definition
• Oligopolistic differentiated product market.
• Each store is treated as an independent market.
• Each brand of beer is considered as a product.
Types of Beer
1. Mass produced beers are
defined as those with similar
characteristics of lightness,
same fermentation method
(bottom fermenting yeast) and
the use of adjuncts such as
corn or rice.
2. Import beers are those
produced abroad.
3. The rest of the beers are called
craft beers.
8
Number of Firms
• Long term secular decline in traditional breweries
• Rapid expansion in specialty breweries since 1980
Market Shares by Beer Type
Sample Averages for Dominick Stores
Type
Craft
Mass
Import
Share
5.3%
86.4%
8.2%
Price Per Bottle
0.80
0.54
0.95
Discrete Choice Model Issues
• Model weekly aggregate sales at each store, by
beer type
• Address dimensionality problem (large number
of underlying products) by projecting the
products onto a characteristics space.
• Market characterized by differentiated products.
• Prices may be correlated with unobserved
demand factors, causing endogeneity problem.
Discrete Choice Model
• Utility of consumer i for product j depends on
characteristics of both the product and the
consumer:
U ( x j ,  j , p j , vi , d )
 Observed
product characteristics, 𝑥𝑗 .
 Unobserved product characteristics, 𝜉𝑗 .
 Price, 𝑝𝑗 .
 Consumer characteristics,𝑣𝑖 .
 Demand parameters, 𝜃𝑑 .
Observable Variables
( x j , p j , j )
 Observed product characteristics:
– Size of the bottle
– Alcohol content
– Type (Mass, Craft, Import)
– Style (Ale, Fruit, Low Alcohol, Oktoberfest, Seasonal,
Smoked, Steam, Stout, Wheat)
 Price
 Observed consumer characteristics:
– Household income, home value, household size,
education (% college graduates), ethnicity (%
blacks+hispanics)
Discrete Choice Model
• Linear specification of utility
uij  z j    p j   j   ij
• where
z j   x j  j 
• j is interpreted as the mean of consumers’
valuations of unobserved product characteristics
(product quality).
• Error term encompasses the distribution of
consumer preferences around j .
• Errors are i.i.d. with “extreme value” distribution,
resulting in a multinomial logit formulation.
Mean Utility Representation
uij  d j   ij
• Simply using dj to represent the mean utility
for product j , which is defined as
everything other than the error term:
d j  z j  pj   j
Multinomial Logit
• The market share of product j is then
expressible in term of dj :
s j (d ) 
e
N
δj
e
k 0
δk
Multinomial Logit
• Assuming the relationship between
observed and predicted market shares is
invertible, with the mean utility of the
outside good (all other than beers)
normalized to zero,
ln( s j )  ln( s0 )  d j  z j    p j   j
• Prices and unobserved product attributes
are correlated  Endogeneity.
Instrument for Prices
• Prices in other markets? (Hausman, 1996).
 Prices
of brand j in two markets will be
correlated due to the common marginal
cost.
 But
prices in other markets uncorrelated
with the market-specific unobserved
product characteristics.
Variable \ Method
MNL
Price
-9.10E-06***
0.000
Size
9.11E-06***
0.000
Alcohol
-2.63E-06***
0.000
Craft
-1.77E-05***
0.000
Import
-1.74E-05***
0.000
Ethnic
8.22E-06
0.000
Education
-2.51E-05
0.000
Household Size
-7.90E-06
0.000
Incomes
6.85E-08
0.000
Observations
12066
R2
0.201
Legend: * p<.1; ** p<.05; *** p<.01.
MNL-IV
-0.283***
0.012
0.054***
0.002
0.029***
0.010
-0.319***
0.024
-0.202***
0.026
0.139***
0.047
0.217
0.155
-0.179***
0.030
0.002***
0.000
12066
0.438
MNL: Ignores
endogeneity
of prices.
MNL-IV: Prices in
other markets
as IV for
Price.
Problem with MNL
• Independence of Irrelevant Alternatives (IIA).
 Example,
if a consumer wants to try a beer that
is an American lager, he/she may consider
alternatives like Coors light or Bud Light, but
he will not consider any Stout type of beer.
Nested Logit Model
• The NL preserves the assumption that
consumer tastes are extreme value
distributed.
• Allows consumer tastes to be correlated
across products.
• More reasonable substitution patterns than
in the previous model (a priori).
Nested Logit Model
• We divide the products into g different
exhaustive and mutually exclusive
groups. u  d    (1   )
ij
j
jg
ij
d j  z j  pj   j
•  is common to all products in group g.
• (1-σ) is the average correlation in the
random utility across products of the
same group.
Nested Logit Model
• Berry (1994) shows that if the errors are
i.i.d. extreme value then:
 jg  (1   ) ij
it is also distributed as a extreme value.
Nested Logit Model
• We can represent the NL model as:
ln( s j )  ln( s0 )  d j  z j    p j   ln( s j / g )   j
where σ measures average similarity of
products within each group of beer types.
The new term is the log of the within group
share.
Variable / Method
MNL
Price
-9.10E-06***
0.000
Size
9.11E-06***
0.000
Alcohol
-2.63E-06***
0.000
Craft
-1.77E-05***
0.000
Import
-1.74E-05***
0.000
Ethnic
8.22E-06
0.000
Education
-2.51E-05
0.000
Household Size
-7.90E-06
0.000
Incomes
6.85E-08
0.000
σ(Average across g)
MNL-IV
-0.283***
0.012
0.054***
0.002
0.029***
0.010
-0.319***
0.024
-0.202***
0.026
0.139***
0.047
0.217
0.155
-0.179***
0.030
0.002***
0.000
Observations
R2
12066
0.438
12066
0.201
Legend: * p<.1; ** p<.05; *** p<.01.
NL-IV
-0.229***
0.011
0.006***
0.001
0.060***
0.008
-5.253***
0.040
-5.122***
0.040
0.090***
0.035
-0.130
0.110
-0.087***
0.022
0.002***
0.000
0.892***
0.000
12066
0.716
Price Elasticities
Mass
Craft
Import
Mass
-0.1223
0.0004
0.0002
Craft
0.0028
-0.3168
0.0013
Import
0.0004
0.0008
-0.1566
Over All
Source: Dominik’s dataset, calculations by the authors.
Over All
-0.1715
Compare with Other Findings
Source
Price Elasticity
Hogarty and Elzinga 1972
-0.889
Orstein and Hanssens 1985
-0.142
Tegene 1990
-0.768
Lee and Tremblay 1992
-0.583
Gallet and List 1998
-0.730
Nelson 1999
-0.200
Nelson 2003
-0.174
This study
-0.172
Source: Table 2.2. Tremblay and Tremblay (2005).
Income Elasticities
Elasticity
Mass
0.257
Craft
0.434
Import
0.460
Over All
0.260
Source: Dominik’s dataset,
calculations by the authors.
Price Elasticities: Other Findings
Source
Income Elasticity
Hogarty and Elzinga 1972
0.430
Orstein and Hanssens 1985
0.011
Tegene 1990
0.731
Lee and Tremblay 1992
0.135
Gallet and List 1998
-0.545
Nelson 1999
0.760
Nelson 2003
-0.032
This study
0.260
Source: Table 2.2. Tremblay and Tremblay (2005).
Conclusions
• Demand for beer is inelastic
with respect to prices.
• Cross-price elasticities are
very close to zero.
 Mass
and craft beers are not
close substitutes!
• From the income elasticities,
all of the types of beer (mass,
craft, and import) are normal
goods.
Next Steps
• Estimate the model using a random
coefficients specification for utility.
• Allow for consumer heterogeneity.
• Consumer characteristics can interact with
product attributes.
• Examine other formulations/instruments to
tackle endogeneity between price and
unobserved product characteristics.
Thank you and
Cheers!
Questions?
(pictures from the
Beeronomics Conference,
Belgium May 2009)