Substitution between Mass-Produced and High-End Beers Daniel Toro-Gonzalez Ph.D. candidate, School of Economic Sciences (SES) Jill J. McCluskey Visiting.

Substitution between Mass-Produced and High-End Beers

Daniel Toro-GonzalezPh.D. candidate, School of Economic Sciences (SES)

Jill J. McCluskeyVisiting Professor, Cornell University and

Professor, SES, Washington State University and

Ron C. MittelhammerRegents Professor, SES & Dept. of Statistics

Presented at Beeronomics Symposium UC DavisNovember 3, 2011

2

Macro Brews Dominate many U.S. Markets

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 Beer1. 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

Source: Calculations by the author. Beer Institute. Brewers Almanac, 2010.

• Long term secular decline in traditional breweries

• Rapid expansion in specialty breweries since 1980

Market Shares by Beer TypeSample Averages for Dominick Stores

Type Share Price Per BottleCraft 5.3% 0.80Mass 86.4% 0.54Import 8.2% 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:

Observed product characteristics, . Unobserved product characteristics, . Price, . Consumer characteristics,. Demand parameters, .

),,,,( dijjj vpxU

Observable Variables

Observed product characteristics:–Size of the bottle–Alcohol content–Type (Mass, Craft, Import)–Style (Ale, Fruit, Low Alcohol, Oktoberfest, Seasonal,

Smoked, Steam, Stout, Wheat)PriceObserved consumer characteristics:–Household income, home value, household size,

education (% college graduates), ethnicity (% blacks+hispanics)

( , , )j j jx p

Discrete Choice Model

ij j j j iju z p • Linear specification of utility

• where

• xj is interpreted as the mean of consumers’ valuations of unobserved product characteristics (product quality).

• Error term encompasses the distribution of consumer preferences around xj .

• Errors are i.i.d. with “extreme value” distribution, resulting in a multinomial logit formulation.

j j jz x

Mean Utility Representation

• Simply using dj to represent the mean utility for product j , which is defined as everything other than the error term:

j j j jz p

ij j iju

Multinomial Logit

• The market share of product j is then expressible in term of dj :

N

0k

δ

δ

jk

j

e

es )(

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,

• Prices and unobserved product attributes are correlated Endogeneity.

0ln( ) ln( )j j j j js s z p

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 MNL-IVPrice -9.10E-06*** -0.283***

0.000 0.012Size 9.11E-06*** 0.054***

  0.000  0.002 Alcohol -2.63E-06*** 0.029***

0.000 0.010Craft -1.77E-05*** -0.319***

  0.000  0.024 Import -1.74E-05*** -0.202***

0.000 0.026Ethnic 8.22E-06 0.139***

  0.000  0.047 Education -2.51E-05 0.217

0.000 0.155Household Size -7.90E-06 -0.179***

  0.000  0.030 Incomes 6.85E-08 0.002***

0.000 0.000Observations 12066 12066R2 0.201  0.438 

Legend: * p<.1; ** p<.05; *** p<.01.

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.

• is common to all products in group g.

• (1-σ) is the average correlation in the random utility across products of the same group.

ij j jg iju (1 )

j j j jz p

Nested Logit Model

• Berry (1994) shows that if the errors are i.i.d. extreme value then:

it is also distributed as a extreme value.

jg ij(1 )

Nested Logit Model

• We can represent the NL model as:

where σ measures average similarity of products within each group of beer types.

The new term is the log of the within group share.

0 /ln( ) ln( ) ln( )j j j j j g js s z p s

Variable / Method MNL MNL-IV NL-IVPrice -9.10E-06*** -0.283*** -0.229***

0.000 0.012 0.011Size 9.11E-06*** 0.054*** 0.006***

  0.000  0.002  0.001 Alcohol -2.63E-06*** 0.029*** 0.060***

0.000 0.010 0.008Craft -1.77E-05*** -0.319*** -5.253***

  0.000  0.024  0.040 Import -1.74E-05*** -0.202*** -5.122***

0.000 0.026 0.040Ethnic 8.22E-06 0.139*** 0.090***

  0.000  0.047  0.035 Education -2.51E-05 0.217 -0.130

0.000 0.155 0.110Household Size -7.90E-06 -0.179*** -0.087***

  0.000  0.030  0.022 Incomes 6.85E-08 0.002*** 0.002***

0.000 0.000 0.000σ(Average across g)       0.892***

         0.000 Observations 12066 12066 12066R2 0.201  0.438  0.716 Legend: * p<.1; ** p<.05; *** p<.01.

Price Elasticities

  Mass Craft Import Over All

Mass -0.1223 0.0004 0.0002

Craft 0.0028 -0.3168 0.0013

Import 0.0004 0.0008 -0.1566

Over All -0.1715Source: Dominik’s dataset, calculations by the authors.

Compare with Other Findings

Source: Table 2.2. Tremblay and Tremblay (2005).

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

Income Elasticities

Source: Dominik’s dataset, calculations by the authors.

  Elasticity

Mass 0.257

Craft 0.434

Import 0.460

Over All 0.260

Price Elasticities: Other Findings

Source: Table 2.2. Tremblay and Tremblay (2005).

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

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)