Complexity, Efficiency, and Fairness in Multi-Product Monopoly - - PowerPoint PPT Presentation

Intro. Preliminaries Model Counterfact. Summary

Complexity, Efficiency, and Fairness in Multi-Product Monopoly Pricing

Eugenio J. Miravete1 Jeff Thurk2 Katja Seim3

1University of Texas at Austin & CEPR 2University of Notre Dame 3The Wharton School, University of Pennsylvania

June 2, 2015

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Motivation

Wal-Mart collects 2.5 trillion megabytes of transaction data every hour. Business intelligence allows for very sophisticated pricing and marketing strategies We have enough information to implement the most sophisticated pricing models envisioned by economic theorist in their wildest dreams. Why pricing practices remain relatively simple though? How much is too much price discrimination?

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Uniform Pricing

We use data from Pennsylvania’s monopoly to study implications of pricing behavior for retail systems. Spirits, our subject of interest, are sold in state-run stores located across Pennsylvania. The PLCB uses uniform pricing policies for all products:

Each product (wine & spirits) sold at identical price statewide. Each product’s price based on a common mark-up percentage.

Uniform pricing is common in movie theaters, grocery store product varieties, nationwide retail chains. Zone pricing is common in the wholesale market for gasoline, convenience store chains, insurance, utilities, and telecommunications.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Empirical Approach

Estimate a discrete choice model of demand for horizontally differentiated products:

BLP demand estimation using prices and sales data for 2005. Estimation links heterogeneous price responses for different products to local demographics and product attributes. Estimation does not impose (supply) profit-maximization equilibrium conditions.

Policy evaluations:

Optimal uniform markup (1). Product-specific markups (234). Product- and market-specific markups (55,210).

Regression analysis of simulated outcome measures (elasticities, compensating variation, profit, etc.) on

• bservable, market-specific, socioeconomic information.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Goals of the Paper

1 Evaluate the welfare effects from using a uniform pricing

policy across heterogeneous products and markets.

How much profit does the PLCB forego relative to optimal monopoly pricing?

Provides implicit valuation of seller’s other considerations in pricing alcohol, such as public health concerns, etc.

Are the welfare gains/losses due to uniform pricing substantial...

...across products? ...across markets?

For 2DPD, gains to complex pricing are frequently small (Chu, Leslie, Sorensen; Rogerson; Wilson.)

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Goals of the Paper

2 How do profits and welfare of uniform pricing and 3DPD

compare to each other?

Robinson (1933) first addressed this question, which essentially cannot be addressed outside very restrictive theoretical models. Schmalensee (1981) proves that welfare increases with 3DPD if output increases relative to uniform pricing when demands are independent and monopolists enjoys CRS. Varian (1985) same result with interdependent demands and no IRS. Schwartz (1990) same result for any cost function that depend

• n total output alone.

Mauleg (1983) and Mauleg-Snyder (2006) explore demand shape restrictions to bound relative profits and welfare.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Goals of the Paper

3 Identify cross-subsidization induced by the current

• ne-size-fits-all pricing policy.

How does demand differ across products and markets? What are the distributional effects of uniform markup policy? Who benefits from and who suffers the consequences from the current uniform markup? How important are these rents? Will Pennsylvanians favor a different pricing policy and/or privatization of the PLCB?

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Motivation Approach Goals Outline

Agenda

Document Preference Heterogeneity Data and Background: PA Liquor Market Demand Model Estimates Analysis of Alternative Pricing Policies

Optimal Uniform Markup Product-specific Pricing Product- and Market-specific Pricing

Summary and Conclusions

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Basic Intuition & Approach

P P P MC

MR0

H+MRL

P0

H

P0

L

qH q0

H

q0

L

qL Q0

*

Q P0

(We estimate demand for 234 products in 484 markets and 20 pricing periods).

Profits increase with segmentation; welfare ambiguous & dependent on relative curvature of demands. Overall welfare effects depend on variation of markups across products and spatial 3DPD to account for consumers’ preference heterogeneity.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Preference Heterogeneity

4.6 6.0 19.3 31.7 38.4 5.4 7.7 18.0 37.9 31.0 4.6 10.8 20.6 33.0 30.9 4.1 12.3 25.0 30.9 27.8 3.8 13.8 24.9 30.1 27.3 4.4 13.2 23.5 32.2 26.7 4.2 18.6 27.9 27.2 22.2 3.6 15.4 29.4 28.5 23.1 2.6 20.2 33.3 25.7 18.3 2.7 23.5 35.3 22.4 16.1

10 20 30 40 50 60 70 80 90 100 Market Share (%) 10 20 30 40 50 60 70 80 90

(a) Share of Non-white Population

3.6 9.3 23.5 22.3 41.2 4.5 6.6 21.0 29.6 38.3 4.9 7.2 19.7 35.8 32.4 5.2 8.2 17.3 38.5 30.8 5.5 8.4 15.5 40.5 30.0 5.5 8.5 13.9 42.7 29.4 5.3 9.1 13.1 43.8 28.7 5.3 9.5 11.1 44.4 29.7

10 20 30 40 50 60 70 80 90 100 Market Share (%) 10 20 30 40 50 60 70

(b) Share of College Educated Population

3.0 11.8 24.0 25.4 35.8 3.6 12.8 26.9 26.3 30.4 4.0 7.7 21.1 26.1 41.1 4.7 6.4 20.8 31.7 36.3 4.9 7.0 19.7 36.2 32.2 5.5 7.3 16.3 39.8 31.0 5.9 7.5 13.5 44.0 29.1 6.1 7.4 12.9 46.1 27.5

10 20 30 40 50 60 70 80 90 100 Market Share (%) 10 20 30 40 50 60 70

Income

Whiskey Vodka Rum Gin Tequila

(c) Share of Population with Income > $50K

74.9 25.1 67.0 33.0 66.7 33.3 63.6 36.4 59.3 40.7 54.5 45.5 49.7 50.3 47.7 52.3

10 20 30 40 50 60 70 80 90 100 Market Share (%) 10 20 30 40 50 60 70

Income

Expensive Cheap

(d) Share of Population with Income > $50K Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Preference Heterogeneity

Table 3: Product Choice Sets in Markets with Different Income Levels

Bottom Income Decile Top Income Decile Alcohol Type Products Price Share Elasticity Products Price Share Elasticity gin 11.88 11.68 13.38

• 1.85

13.64 16.85 8.03

• 2.69

rum 23.27 12.51 26.10

• 1.98

24.69 12.76 14.13

• 2.03

tequila 9.33 20.25 4.05

• 3.23

15.18 21.85 6.38

• 3.48

vodka 38.88 15.54 27.69

• 2.47

48.40 16.61 43.62

• 2.64

whiskey 38.03 14.64 33.69

• 2.33

48.83 18.12 29.01

• 2.89

all products 117.90 14.07 22.03

• 2.24

148.80 16.85 25.06

• 2.68

Statistics based on markets with average annual income either less than $31,020 (Bottom) or more than $74,400 (Top). “Products” reports average number of products available of each category. Average price and elasticity statistics are sales-weighted. “Share” denotes the share of each liquor category based on total PLCB sales of 750ml bottles while “all products” corresponds to the share of total liquor sales in PA. Elasticity statistics based on multinomial logit demand system estimates and a price coefficient of -0.16.

Preferences are markedly different across income levels. Affluent consumers purchase more varieties of high quality products priced $3 higher on average. Affluent consumers concentrate their demand on vodka and whiskeys while poorest consumers favor gin and rum.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Variation in Market Demographics

Table 5: Demographics Across Stores

% of Population Statistic Population Minority Age ≥ 45 Income ≥ 50k Educ ≥ College mean 29,440.9 12.3 10.6 39.2 24.4 sd 17,708.5 18.0 3.7 12.7 13.8 max 111,964.0 98.9 33.4 72.0 72.7 min 2,574.0 0.2 5.2 9.4 3.0

Substantial variation in income, as well as other demographics, across the state of Pennsylvania. Combined with relationship between tastes and demographics, suggests that a uniform pricing rule should affect different consumers differently.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Background: PLCB Operations

Pennsylvania holds state monopoly for wine & spirits wholesale and retail. Motivations for public enterprise:

Significant source of tax revenue. “Sell liquor responsibly”. Potentially significant monopsony power.

Store network: 624 stores across Pennsylvania as of 1/2005.

Majority stand-alone retail stores. 7 outlet stores; 65 “premium” stores.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Background: PLCB Pricing (I)

Prices regulated by State Legislature. Use of uniform prices:

Uniform mark-up rule: each product’s retail price reflects 30% markup over wholesale cost and 18% liquor tax.

Temporary price reductions:

PLCB runs monthly sales, passing on temporary wholesale price cuts. For chosen selection of sale products, price reduced by on average $1.00. Same sale prices used in all stores, subject to product availability.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Background: PLCB Pricing (II)

PLCB negotiations with suppliers: A new product’s wholesale price remains fixed for one year following introduction. Wholesale price increases for established products negotiated

• n a quarterly basis.

Wholesale price decreases effective immediately. Adjustments translate into retail price changes effective in following 4-week reporting period. Together with sales, observe price changes approximately every two weeks.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Data

Store-level panel data obtained from PLCB for 2005. Daily retail & wholesale prices, bottles sold per store and product.

Aggregate to “pricing periods”: ∼14-day periods where prices remain constant (no change in wholesale prices or sales).

Analysis conducted at the store-market level.

Assign consumers of 10,000 PA block groups to their closest store in each period. If multiple stores within zip code, aggregate sales across stores. Consider resulting 484 store markets as separate markets.

Abstracting from store choice reflects identical retail prices and similar product availability across stores.

For 2005, 9,530 markets (zones × pricing periods) and 1,104,205 observations.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Geographic Markets

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Store Sales

Ambient population: 29,441 people. Average sales per pricing period: 5,786 750ml-bottles. Sales composition:

Focus on five types of spirits: Gin, Rum, Tequila, Vodka, and Whiskey (reference category). Five categories account for 29.9% of bottle sales. Median store carried 98% of the top 100 best selling products statewide.

We also control for holiday, specialty store, and border market.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Spirit Categories

Table 4: Product Characteristics by Alcohol Type

Number of Average Alcohol Type Products Share Price Proof Flavored Imported Score gin 21 14.61 14.32 84.93 0.00 41.13 64.02 rum 44 20.01 12.35 79.39 18.35 44.37 39.99 tequila 29 5.18 20.77 79.86 1.43 100.00 58.42 vodka 68 31.32 15.71 80.53 29.57 47.98 52.02 whiskey 72 28.87 15.72 81.50 0.00 30.52 53.01 all products 234 100.00 15.16 80.93 14.20 44.00 51.19 “Share” is defined as percentage with respect to total liquor sales (750ml bottles).

Whiskeys and Vodkas have larger market share. Rums are generally considered a low-quality product. Many of these products are imported and also flavored. Number of varieties available increases with category market share.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Price Patterns: Distribution within Categories

Table 2: Prices by Alcohol Type

Alcohol Type Avg Median Std Min Max gin 14.32 10.99 6.36 5.99 28.99 rum 12.35 12.49 2.74 5.49 21.99 tequila 20.77 18.99 7.71 11.99 52.99 vodka 15.71 13.99 5.54 5.69 30.99 whiskey 15.72 13.99 6.88 4.99 42.99 all products 15.16 13.49 6.07 4.99 52.99 Sales-weighted averages of 2005 liquor prices in Pennsylvania.

PLCB simply adds a uniform markup on top of wholesale price. Since 30% markup does not respond to changes in demand, retail liquor prices become endogenous as it reflects the effect of liquors’ unobserved characteristics and the nature of competition among distillers.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Price Patterns: Sales

12 13 14 15 16 17 18 19 20 Retail Price ($) 1 4 7 10 13 16 19 22 25 Pricing Period

Beefeater Gin ($18.99) Jim Beam Whiskey ($18.99) Absolut Vodka ($19.49) Sauza Tequila ($14.99) Bacardi Rum ($12.49)

Retail Price Across Time (Select Products)

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Intuition Background Data

Features: Sales

Table 6: Characteristics of Liquor Sales

Alcohol Type Sale Number Duration gin 61.9 3.2 1.6 rum 79.5 3.0 2.3 tequila 82.8 3.4 1.6 vodka 91.2 3.0 1.7 whiskey 77.8 2.9 1.6 cheap 79.2 3.0 1.9 expensive 83.3 3.0 1.6 all products 81.2 3.0 1.8

Table 7: Percent of Products on Sale in a Given Period

Alcohol Type gin rum tequila vodka whiskey spring 10.0 16.0 17.1 16.5 11.8 summer 18.3 19.5 28.4 23.5 21.4 fall 11.7 15.1 14.8 14.7 12.1 winter 19.1 20.6 20.7 21.5 20.6 year 15.5 17.9 20.2 19.0 16.6

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Demand Model

We estimate a standard discrete-choice demand model for differentiated products with aggregate data (BLP 1995). Each period, consumer chooses a product among offerings in the five spirits categories or the outside option of on-site consumption to maximize utility. Potential market: assume that every resident over age of 21 purchases state average per-capita quantity of spirits of 0.01 bottle per day 3.5 750ml bottles a year). Assumption of purchase of single bottle necessitated by lack

• f individual-level data, but rules out stockpiling, variety

seeking etc.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Utility function (I)

Consumer utility: uijlt = xjtβ∗

i + α∗ i pjt + ξjt + ǫijlt ,

i = 1, . . . , Ilt; j = 1, . . . , Jlt; l = 1, . . . , L; t = 1, . . . , T . where:

xjt: observed characteristics (proof, flavored, imported, spirit type, product quality score). pjt: price - common within product across geographic markets. ξjt: vector of unobserved characteristics.

decompose into product fixed effect and deviations thereof, capturing seasonal preference variation for different products.

ǫijlt: unobserved tastes of consumer i for product j, assumed distributed i.i.d. extreme value

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Utility function (II)

Random coefficients: α∗

i

β∗

i

• =

αi βi

• + ΠDi + Σνi ,

νi ∼ N(0, IK+1) , where:

Di: d × 1 vector of demographics. Simulate market consumers by drawing from:

Estimated market-specific distribution of income obtained by fitting generalized beta dist. of 2nd kind to observed discrete income distribution. Bernoulli dist. to determine discrete attributes: minority, highly educated

νi: vector of unobserved idiosyncratic taste components.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Predicted Market Shares

For consumer i, Logit assumption implies product choice probabilities of: sijlt = exp(xjtβ∗

i + α∗ i pjt + ξjt)

1 +

• k=1,...,J

exp(xktβ∗

i + α∗ i pkt + ξkt)

Integrating with respect to the distributions of observed and unobservable idiosyncratic taste parameters yields the predicted market share of product j, for a given value of the error term ξjt: sjlt =

• ν
• D

sijltdP ∗

D(D)dP ∗ ν (ν)

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Estimation

Demand error term is value of unobserved product characteristic ξjt for product j in period t. As in Nevo (2000,2001), the estimation includes two stages:

First: characterize demand as function of price, holiday dummy, and product fixed effect.

For each market l find common taste for product j, δjlt, that equalizes predicted and actual market shares of product j. Project δjlt onto prices and product fixed effects, assuming that individual markets have mean zero i.i.d. unobserved product shocks: ξ(θ)jt = 1 L

L

X

l=1

[δjlt(x·, p·t, S·lt; θ) − xjβ − αpjt]

Second: project fixed effects on product characteristics.

Interact with instruments to find parameters using GMM.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Instruments

As in unregulated industries, potential for price endogeneity.

Distillers can reduce wholesale prices up to four times a year; PLCB passes on price reductions. Distillers commit to price reductions five months prior to sale. If distillers aware of seasonal variation in their products’ valuations, expect wholesale prices to be correlated with unobserved ξjt.

Employ average retail prices in nearby control states in different pricing periods as instruments. Identifying assumption:

State-specific seasonal valuations of products are independent across states and time. Market-specific preference shocks unobserved at time when sale decision made (i.e. product placement on shelf, tastings...)

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Identification of Estimates

Variation of prices of each product over time identifies the

• wn-price elasticity.

Cross-price elasticities are identified because relative prices of different products do not remain constant over time. Across-market heterogeneity of the distribution of demographics helps identify the interaction coefficients. Notice again that the wholesale price is known and thus we

• nly estimate demand without the need to impose profit

maximization conditions.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Parameter Estimates - Summary

Parameters are in general significant. Demand is downward slopping and elastic. Demand elasticity increases with income.

Reflects availability of closer substitutes for product categories favored by high-income households.

Positive valuation of proof, imported, and score (reported quality) but negative effect of flavored liquors. Demand increases during the holidays.

Miravete, Thurk, Seim Alcohol Pricing

Table 8: Demand Estimates

OLS IV Full Model Mean Utility (β) price

• 0.1576 (0.0094) ***
• 0.2056 (0.0183) ***
• 0.1715 (0.0001) ***

holiday 0.5807 (0.0135) *** 0.5798 (0.0140) *** 0.7167 (0.0006) *** constant

• 8.8880 (0.5082) ***
• 9.1439 (0.6687) *** -10.1080 (0.6875) ***

proof 2.3658 (0.4796) *** 2.8882 (0.6388) *** 2.8280 (0.7375) *** product score 0.3380 (0.0706) *** 0.4301 (0.0908) *** 0.4150 (0.0707) *** flavored

• 0.6124 (0.1631) ***
• 0.6794 (0.1986) ***
• 0.6590 (0.1884) ***

imported 0.8712 (0.1347) *** 1.1753 (0.1655) *** 1.1240 (0.1423) *** gin

• 0.7713 (0.2626) ***
• 0.9568 (0.3117) ***
• 0.9060 (0.2577) ***

rum 0.1921 (0.1963) 0.1777 (0.2426) *** 0.2100 (0.2243) tequila 0.7283 (0.3568) ** 1.0983 (0.4615) ** 0.9830 (0.2374) *** vodka 0.3978 (0.2088) * 0.4148 (0.2547) 0.4230 (0.2007) ** Standard Dev. (σ) price 0.0017 (0.0015) constant 0.0173 (0.0423) Interactions (Π) income×constant 2.4698 (0.0009) *** price×income

• 0.0523 (0.0001) ***

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Estimated Own Price Elasticities - Full Model

Table 9: Price Elasticities by Alcohol Type

Alcohol Type Price Elast Median p25 p75 sd gin 14.3

• 2.9
• 2.3
• 4.1
• 1.9

1.3 rum 12.3

• 2.5
• 2.5
• 2.8
• 2.2

0.6 tequila 20.8

• 4.1
• 3.8
• 4.2
• 3.4

1.2 vodka 15.7

• 3.2
• 2.9
• 3.9
• 2.4

1.1 whiskey 15.7

• 3.1
• 2.9
• 4.0
• 2.1

1.3 all products 15.2

• 3.0
• 2.8
• 3.8
• 2.3

1.2 cheap 11.3

• 2.3
• 2.4
• 2.7
• 1.9

0.6 expensive 20.8

• 4.1
• 3.9
• 4.6
• 3.5

0.9

At a 30% markup, liquor elasticities are quite different. Rums are less elastic than any other category Demand for bottom 50% of least expensive spirits less elastic than for top.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary BLP Estimates

Estimated Own Price Elasticities - Full Model

Figure 2: Distribution of Demand Elasticities

Mean: -3.24 Median: -2.92 Std Dev: 1.30 N: 1.1e+06 1 2 3 4 5 6 7 8 9 Percent of Total Observations

• 10
• 8
• 6
• 4
• 2

Price Elasticity

(a) All Liquor

Mean: -2.98 Median: -2.38 Std Dev: 1.25 N: 98023.00 1 2 3 4 5 6 7 8 9 Percent of Total Observations

• 6
• 4
• 2

Price Elasticity

(b) Gin

Alcohol products are segmented by price.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Uniform vs. More Sophisticated Pricing

Prices under alternative objectives: A multiproduct profit maximizing firm f who offers products

• f a subset ℑf will price according to:

sj(p) +

• r∈ℑf

(pr − mcr) ∂sr(p) ∂pj = 0 . A system of J equations defines product j’s optimal markup. Given mcj, obtain optimal prices and use to evaluate welfare changes under different scenarios:

1

Uniform: Same percentage markup across products.

2

Product Specific: Different markup for each spirit although they are the same state-wide.

3

3DPD: Markups vary by product and by local market.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Counterfactuals: Summary

Table 10: Counterfactuals Summary Statistics

Scenario Profits CS Welfare Quantity Avg Price 30% uniform markup 53.33 58.28 111.61 8.89 15.16 Uniform Π-maximizing markup (32%) 53.58 56.20 109.78 8.60 15.28 product-specific 59.31 50.77 110.08 8.03 18.52 3rd degree price discrimination 61.00 50.01 111.00 8.27 18.52 perfect competition 40.69 96.01 136.71 13.37 13.40 Profits, consumer surplus, and welfare are denominated in millions of dollars. Profits include liquor taxes. Quantity is millions of 750ml bottles. Average price is sales-weighted.

PLCB vs. Competition: P ↑ $1.76(13%); Q ↓ 4.5m(44%); Π ↑ $12.6m(31%); CS ↓ $37.7m(40%); W m = 0.82W fb. Uniform: PLCB’s current 30% is very close to the optimal 32%; W m = 0.995W u. Product-Specific vs. 30%: P ↑ $3.36(22%); Q ↓ 0.8m(10%); Π ↑ $3.5m(11%); CS ↓ $7.5m(8%); W ps = 0.988W m ⇔ rent redistribution & no efficiency loss. 3DPD vs. Product-Specific: Solving for an additional 50,000 markups only increases profits by $1.7m (2.85%).

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Counterfactuals: Markups

Table 11: Product Percent Markups by Alcohol Type

Benchmark Uniform (Π-max) Product-Specific Type Average Std Average Std Average Std gin 31.1 6.1 32.0 0.0 50.6 41.0 rum 30.8 7.1 32.0 0.0 59.3 22.1 tequila 28.6 6.3 32.0 0.0 22.0 14.5 vodka 28.8 8.2 32.0 0.0 40.4 26.7 whiskey 29.8 7.2 32.0 0.0 39.8 30.8 all products 29.7 7.5 32.0 0.0 42.6 29.7 cheap 29.5 8.8 32.0 0.0 66.2 30.1 expensive 29.9 4.9 32.0 0.0 25.1 11.9 Average and standard deviation of the distribution of markups. Markups do not include the 18% liquor tax and a $1 bottling fee.

Uniform markups are not that far from current pricing practice. Product-Specific markups are higher, 42.6%. Increases are higher for cheap spirits, gins, and rums.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Counterfactuals: Prices

Table 12: Change in Prices Across Pricing Strategies

Benchmark Uniform (Π-max) Product-Specific Type

• Bench. Uniform Product Avg %∆ Std

%∆ > 0 Avg %∆ Std %∆ > 0 gin 14.3 14.4 17.9 0.5 4.1 40.1 12.2 25.1 66.9 rum 12.3 12.4 15.0 0.6 4.9 21.7 19.0 14.7 94.8 tequila 20.8 21.1 24.2 1.9 4.3 59.6

• 5.5

11.3 41.1 vodka 15.7 16.0 18.4 1.8 5.7 38.1 7.2 17.8 67.1 whiskey 15.7 15.8 19.1 1.2 4.9 40.9 6.0 20.6 65.0 all products 15.2 15.3 18.5 1.2 5.1 37.8 8.1 19.6 69.6 cheap 11.3 11.5 14.4 1.3 6.0 27.1 24.2 18.5 98.0 expensive 20.8 21.0 21.6 1.1 3.3 49.7

• 3.8

9.0 37.9

Prices virtually unchanged with the “Uniform” pricing rule. Product-specific pricing takes into account cross-product substitution effects, resulting in large heterogeneity in price changes. Prices of cheap liquors increase substantially when the monopolist has the ability to optimally price products separately. Notice that prices of most spirits will go up in both cases.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Distribution of Price Changes, Product-Specific Prices

.2 .4 .6 Density

• 50

50 100

All

.5 1 1.5 Density

• 50

50 100

Gin

.5 1 1.5 Density

• 50

50 100

Rum

.5 1 1.5 2 2.5 Density

• 50

50 100 Price Change (%)

Tequila

.2 .4 .6 .8 1 Density

• 50

50 100 Price Change (%)

Vodka

.2 .4 .6 .8 Density

• 50

50 100 Price Change (%)

Whiskey

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Counterfactuals: Sales and Profits

Table 13: Change in Sales and Profits Across Pricing Strategies

Benchmark %∆ Quantity Benchmark %∆ Profits Type Quantity Uniform Product Profits Uniform Product gin 0.8

• 1.1
• 14.9

4.4 0.8 10.5 rum 1.8

• 0.5
• 30.0

9.4 1.6 2.0 tequila 0.5

• 7.0

38.1 3.6

• 1.5

29.3 vodka 3.1

• 5.0
• 9.7

19.0 0.3 9.1 whiskey 2.7

• 3.1
• 2.4

17.0 0.3 15.0 all products 8.9

• 3.3
• 9.6

53.3 0.5 11.2 cheap 5.3

• 2.8
• 35.3

24.5 1.5 1.9 expensive 3.6

• 4.0

27.8 28.8

• 0.4

19.1

Product-specific pricing has more noticeable effects on demand as prices increase substantially. Demand for low quality products gets reduced significantly. Demand for expensive spirits increase leading to most of the profit increase of product-specific pricing.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Taxation by Liquor Pricing Regulation

Great heterogeneity of demand elasticities at a common 30% uniform markup. The current system distort pricing and leads to cross-subsidization among customers. If we move to product-specific pricing, high income and educated households with more elastic demands typically face less important price increases than low income minorities. Therefore, the current system acts as a tax on high income earners that transfer rents to minorites through “subsidized” alcohol pricing.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

Evaluating Consumer Welfare Implications of Price Changes

CVi(pz, pu), the compensating variation for individual i of moving from uniform pricing pu to an alternative pricing regime pz, is:

ln hPJ

j=0 exp

` xjtβ∗

i +α∗ i pz jt+ξjt

´i −ln hPJ

j=0 exp

` xjtβ∗

i +α∗ i pu jt+ξjt

´i α∗

i

.

The mean compensating variation for a market of size M is:

CV (pz, pu) = M Z CVi(pz, pu)dP ∗

D(D)dP ∗ ν (ν) ,

which can be numerically evaluated given the estimated parameters of the model.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness Table 15: Projecting Market-Level Aggregate Effects onto Market Characteristics

∆ Profits (%) %CV < 0 ($) CV ($) ∆ Quantity (%) ∆ Price (%) (1) (2) (1) (2) (1) (2) (1) (2) (1) (2) minority 4.3068** 4.5810**

• 6.2328
• 7.9586

0.1570*** 0.1556** 0.0302

• 0.7553

3.7062*** 4.2486*** (1.5730) (1.7608) (5.7171) (6.9082) (0.0431) (0.0487) (2.4047) (2.9222) (1.0876) (1.2586) high income

• 4.1552*
• 5.5287*

43.1788*** 39.5889***

• 0.1164
• 0.1288

14.0634*** 12.6202**

• 2.2609
• 2.1169

(1.7819) (2.3079) (8.6637) (10.1435) (0.0602) (0.0727) (3.4538) (4.4085) (1.7141) (2.0649) educ ≥ college 9.8505*** 9.7100*** 87.2331*** 87.9234***

• 0.0886
• 0.0901

25.8991*** 26.0711***

• 3.8252**
• 4.2001**

(1.4145) (1.6517) (7.1910) (8.1820) (0.0544) (0.0637) (2.5201) (2.9484) (1.1674) (1.3538) registered democrat 3.5131** 3.4006**

• 5.9609
• 5.9903

0.1105** 0.1166**

• 2.6179
• 2.4669

3.5986** 3.2456* (1.2074) (1.2790) (4.6813) (5.2827) (0.0392) (0.0427) (2.0465) (2.2222) (1.2429) (1.3341) churches per capita

• 1.1220
• 1.4890

4.9547 4.3741

• 0.2126*
• 0.1892
• 5.2543
• 5.8008

4.4332 5.7284 (2.5502) (2.7376) (13.1142) (14.8887) (0.1017) (0.1126) (5.5310) (6.2659) (3.4432) (4.0269) population density

• 0.0381
• 0.0502

0.7505** 0.8327**

• 0.0018
• 0.0024

0.2395*** 0.2709***

• 0.1814***
• 0.2079***

(0.0412) (0.0444) (0.2590) (0.2902) (0.0013) (0.0014) (0.0721) (0.0815) (0.0315) (0.0351) liquor related crimes

• 3.5573
• 2.2964
• 0.1309

2.3947

• 8.1499*

(4.7964) (10.2191) (0.0976) (4.8393) (3.6661) constant 6.9837*** 7.6949***

• 8.2970
• 7.7291

0.1582*** 0.1657***

• 20.3248***
• 20.0798***

20.9291*** 21.3076*** (1.1172) (1.2722) (5.5684) (6.4242) (0.0421) (0.0478) (2.2209) (2.5562) (1.3776) (1.5738) R2 0.1595 0.1489 0.6576 0.6492 0.1962 0.2013 0.3716 0.3142 0.1762 0.2067 N 484 371 484 371 484 371 484 371 484 371 Robust standard errors reported in between parentheses with p-values denoted by * p < 0.10, ** p < 0.05, and *** p < 0.01. minority is the percent

• f consumers which are minorities. high income is the share of consumers with annual income greater than $50K. educ ≥ college is share of

consumers with at least some college education. registered democrat is share of consumers registered as democrats. liquor related crimes refers to the per-capita number of reported incidences of domestic violence, drunk driving, drunkenness, vandalism over 2002-2003.

CV effects are robust to alcohol related crimes.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary Complexity Results Fairness

How Many Win with PD? How Much?

Table 16: Compensating Variation Conditional on Demographics

Demographic (Avg) (Mean) (Std) (%CV < 0) No College 2.58 0.58 6.60 26.6 College 2.80 0.64 7.12 25.4 Low Income 0.72 0.08 3.52 39.1 High Income 4.54 2.10 8.40 13.6 Minority 3.44 1.00 7.08 22.5 Non-Minority 2.54 0.56 6.66 26.8 All 2.64 0.60 6.72 26.3

CV effects are evidently heterogeneous within demographic groups. Only 26% of Pennsylvanians prefer a more flexible pricing system than the current one. Gains are very limited.

Miravete, Thurk, Seim Alcohol Pricing

Intro. Preliminaries Model Counterfact. Summary

Summary

Conclusions: PLCB’s current practice is not far from the optimal monopoly pricing when constrained to use a uniform markups. Very complex pricing strategies are neither socially preferable nor privately profitable. Rather than fair, the current system distort prices and taxes educated, high income households and transfer those rents to minorities.

Miravete, Thurk, Seim Alcohol Pricing