Retail Channel Price Discrimination Steven S. Cuellar, Ph.D. Department of Economics Sonoma State University Rohnert Park, CA USA (707) 664-2305 [email protected]

Marco Brunamonti Korbel Estates August 5, 2010

Abstract This paper examines price differentials of identical items across retail channels. Many consumer packaged goods are sold through both grocery and drug stores. Liquor is unique in that in much of the country there is a third retail channel of distribution, liquor stores. If consumers in each retail channel differ in their willingness to pay for certain items, then sellers can exploit those differences and charge different prices for the same items in each channel. We examine a unique data set of pooled cross sectional retail scanner data on wine to test whether sellers use retail channel to identify heterogeneous consumer market segments and engage in price discrimination. We begin by presenting a model of price discrimination by retail channel. Next we examine sales by retail channel and test for price differences across these channels. Controlling for sample selection bias and seasonality, we find persistent price differentials for the same item across retail channel. The extent of price differential, however, differs significantly with respect to price point and sales volume.

Keywords: Price Discrimination, Market Segmentation, Fixed Effects Models.

LITERATURE REVIEW This paper investigates the difference in price of identical items across retail channel. We argue that these retail channel price differentials are a form of third degree price discrimination in which consumers, who differ in their price elasticities of demand, self-select themselves to each retail channel. Modern concepts of price discrimination in non-competitive markets go back at least to Pigou (1920), whose categorization of price discrimination into first, second and third degree is still used today. Robinson (1933) elaborated on the conditions required for firms to engage in effective third degree price discrimination, namely that there exist identifiable market segments that differ in their price elasticities of demand. Using this background, Blattberg and Sen (1974 and 1976) and Blattberg, Buesing, Peacock and Sen (1978) show how market segmentation based on identifiable demographic characteristics can be effectively exploited. More recently, Hoch, Kim, Montgomery and Rossi (1995) use scanner data to show how demographic characteristics can be used to price discriminate by store location. Where differences in price elasticity are not easily identifiable, Moorthy (1984) provides a model where firms exploit differences in consumer preferences across market segments by offering product variants at different prices, allowing consumers to self-select among those products. More generalized models of price discrimination in contestable markets with differentiated products have been developed by Salop and Stiglitz (1977), Borenstein (1985) and Holmes (1989). The type of consumer behavior closest to that examined in this paper is that of Narasimhan (1984), who presents a model of coupon use as a form of price discrimination for identical goods. Specifically, Narasimhan presents a model in which consumers, who differ in their price elasticity of demand, self-select themselves into coupon

use based on comparing the savings associated with using coupons with the opportunity cost of using coupons. We also investigate aspects of shopping behavior as defined in Kahn and Schmittlein (1989) as well as shopping costs as modelled in Bell, Ho and Tang (1998). Finally, with respect to retail channel, Gerstner, Hess and Holthausen (1994) examine price discrimination by retail channel, however, their paper concentrates on the effect of retailer mark-up on the size of discount offered. Our paper, on the other hand, provides a unique perspective on the use of retail channel itself as a means of price discrimination.

A MODEL OF PRICE DISCRIMINATION In this paper, we model retail channel as a form of market segmentation. Just as coupons serve as a means of segmenting consumers distinguished by their price elasticities of demand (Narasimhan 1984), retail channel can serve a similar function. From the consumer’s perspective, purchasing some goods at a lower cost retail channel provides an alternative as long as the savings associated with shopping at that channel are greater than the costs. In this context, choice of retail channel is consistent with Narasimhan’s (1984) model of coupon usage on several dimensions. First, both coupon usage and retail channel are decisions of selfselection by rational utility maximizing consumers. Second, consumers will decide to purchase a specific product at a lower priced retail channel as long as the savings is greater than the opportunity cost required to search, travel to, and shop at that channel for the specific product. This is analogous to the model of coupons where usage depends on the savings being greater than the opportunity cost in time required to search, clip (print etc.), store, organize, retrieve and use coupons. Finally, coupon usage and choice of retail channel are both decisions

consumers make based on the potential savings on individual products and not overall savings on all products. Specifically, we do not assume that some retail channels are more or less expensive for all or even most products, but rather that some retail channels are more or less expensive for one particular product. From the firms’ perspective, a firm will engage in price discrimination as long as the marginal revenue from price discrimination is greater than the marginal cost. While price discrimination is generally associated with monopolistic or oligopolistic industries, models of price discrimination in differentiated product markets have been developed by Borenstein (1985) and others. In its simplest form, given i different market segments, which differ in their price elasticities of demand, and for whom arbitrage among the market segments is costly, we can investigate discriminatory behavior by examining the firms’ profit function, Π= ∑

i(Qi)

- TC(Q),

Where: i denotes the different market segments identified by the firm. TRi(Qi) represents total revenue in market i from unit sales Qi. TC(Q) represents total cost of production across all three channels so that Q = ∑

i.

Profit maximization results in the usual first order conditions:

If we assume costs are common to all markets ( condition as,

) we can rewrite the optimizing

which produces the familiar condition that MRi =MC. Since TRi = PiQi

Which can be rewritten to express in elasticity form,

For our three-market (channel) scenario, this results in . In this form we see that the price in each market is inversely proportionate to the absolute value of the price elasticity of demand in that market. That is, P1|ε3|. DATA We use scanner data of retail purchases of wine in the US to investigate price differentials across three retail channels: Drug stores, food or grocery stores and liquor stores. Scanner data, provided by proprietors such as Information Resources Incorporate (IRI) and the Nielsen Company, is increasingly becoming the primary source of data for analytics in the consumer packaged goods industry due to the ready availability of data at the item level on factors such as price, quantity, promotional activity and sales channel. In this paper, we use Nielsen Scantrack data to construct a pooled cross section of data on point of sale purchases of wines from major U.S. retail chains, for the years 2007-2010. The data consist of national sales of all wines, foreign and domestic, purchased from major retail chain stores, defined as those

with sales of over 2 million dollars per year. The data are aggregated for all markets and include the price paid, quantity sold, store keeping unit (SKU) and retail channel of each item. For uniformity, we concentrate on wine purchases of standard 750 ml glass bottles (approximately 84% of all purchases) and exclude boxed wine, and smaller or larger bottles. The benefit of scan data is that it represents actual purchases of wine by consumers and is thus more reflective of the consumer demand than manufacturers’ suggested retail price. The drawback of scan data is that it only records purchases in major U.S. retail chains and does not represent wine sold on premise at wineries, purchases through wine clubs or purchases at restaurants. Despite these limitations, the scan data works well for our analysis of pricing behavior across major retail channels. Furthermore, wine is unique in that it is sold through three retail channels, food, drug and liquor stores. Finally, the wine industry provides an ideal example of a differentiated product market characterized by a high degree of price and non-price competition.

SUMMARY STATISTICS The data used to examine retail channel price differentials consist of 44 four-week periods from 2007-2010. Each period contains approximately 14,000 unique items sold. However, not all items are sold through all three channels. Cursory examination of mean prices can lead to spurious price differentials based on sample selection bias and not discriminatory pricing behavior. For instance, if drug stores sell wines with lower average prices than grocery stores, and liquor stores sell wines with higher average prices than grocery stores, then differences in mean prices across channels would be due to different wines being carried across

the channels. This is illustrated in Table 1 and Figure 1, which shows summary statistics on prices for the full sample across channels.

Figure 1-Prices by Channel and Sample

Immediately evident from Figure 1 is the monotonic increase across all three channels in the range of prices with food stores selling a significantly greater range of higher priced wines than drug stores and liquor stores selling a greater range of high priced wines than food stores. Also clear from Figure 1 is the monotonic increase in the inter-quartile range of prices in the full sample. The difference in items sold across the three channels is further illustrated in Table 1. The maximum price at drug stores is approximately $43 compared with grocery and liquor

stores selling wines with a maximum price of $178 and over $200 respectively. Obviously differences of this magnitude are the result of differences in the composition of items sold in each channel. Further examination of the data shows that the highest priced wine sold through drug stores channel was a Conn Creek Napa Valley Cabernet Sauvignon blend which sold for $42.68, while the highest priced wine sold through the food channel was an Opus One Red blend that sold for $177.97 and the highest priced wine sold through the liquor channel was a Penfolds Shiraz that sold for $223.72. Clearly these are not the same wines sold through different channels at different prices. This is further illustrated by noting the difference in the number of observations in the full sample across channels in Table 1. Table 1-Prices by Channel and Sub Sample

Full Sample Mean Minimum Maximum Observations Sub Sample 1 Mean Minimum Maximum Observations Sub Sample 2 Mean Minimum Maximum Observations

DRUG $8.11 $0.57 $52.51 29,949

Channel FOOD $11.77 $0.15 $177.97 259,067

LIQUOR $13.22 $0.48 $223.72 226,454

$8.41 $1.18 $52.51 26,479

$9.07 $1.98 $58.14 26,664

$9.34 $2.69 $65.44 26,692

$7.66 $2.89 $24.02 9,396

$8.15 $3.11 $27.03 9,502

$8.47 $3.35 $27.35 9,537

To control for inherent sample selection bias caused by the difference in items sold across channels, we examine two subsets of the data. The first subset contains only those items sold across all three channels in any specific period. Thus, in each period, only items sold

in all three channels in that period are analysed. Sub Sample 1 is illustrated in Figure 1, and again shows a monotonically increasing relationship across the three channels, albeit, with significantly less variation in prices across each channel. Note also, that the highest prices are significantly lower than in the full sample and are more similar across channels. Table 1 confirms the graphical display, showing that the highest prices differ by less than twenty dollars across channels compared to a nearly two hundred dollar differential in the full sample. Table 1 also shows that the number of items sold across channels is nearly identical. However, even within this subset, not all items are sold in all three channels in all periods. While we do control for month sold in our model, we examine a final sub-sample of observations of only those items sold in all three channels in all 44 periods. This is shown in Figure 1 and Table 1 as Sub Sample 2. Figure 1 demonstrates two relevant characteristics about Sub Sample 2. First, the average price of items sold in all three channels in all periods is less than those in Sub Sample 1. Second, there is less variation in prices across channels in Sub Sample 2 than in Sub Sample 1.

RETAIL CHANNEL PRICE DIFFERENTIALS Figure 2 shows the difference between drug store prices and grocery store prices for each Figure 2-Drug Store-Grocery Store Price Differentials

-10

-5

0

5

Drug Store-Grocery Store Price Differential

bottle of wine in Sample 1. Differences greater than zero indicate wines at drug stores priced higher than the same wines sold at a grocery store, and differences less than zero indicate drug store prices less than the same wine sold at grocery stores. As can be seen from Figure 2, not all wines sold at drug stores are priced lower than the same wine at grocery stores. However, the majority of wines in the sample are priced lower at drug stores than the same wine at a grocery store. Figure 3 shows the price differential between liquor stores and grocery stores.

Figure 3-Liquor Store-Grocery Store Price Differential

-4

-2

0

2

4

6

Liquor Store Price-Grocery Store Price Differential

Once again, Figure 3 makes clear that not all wines sold at liquor stores are more expensive than the same wine sold at a grocery store. However, the majority of wines in the sample are priced higher at liquor stores than the same wine at a grocery store. While the summary statistics in Table 1 and Figures 2 and 3 indicate a consistent pattern of price differentials, we now take a closer look at retail channel price differentials by examining the following regression model: Priceijt =  0  1 Drugit   2 Liquorit    Month uijt The model specifies the price of each bottle of wine (i) as a function of channel (j) at time (t). Where: Drug is an indicator variable equal to one if the bottle is sold in a drug store. Liquor is an indicator variable equal to one if the bottle is sold in a liquor store. Month is a vector of monthly indicator variables.

The initial regression results for each sample are shown in columns (1) and (2) of Table 2. The results from Table 2 indicate that, on average, the price of a bottle of wine sold at a drug store is from 6.6% to 8.2% lower than the same bottle of wine sold at a grocery store. For wines sold at liquor stores, prices are 4.2% to 5% more, on average, than the same bottles of wine sold at a liquor store. Both results are statistically significant at a 1% level of significance. Price discrimination occurs when price differentials are based on differences in consumers’ willingness to pay and not on cost differences. A natural question to ask is, if the observed price differentials are driven by cost differentials due to quantity discounts on the part of wine producers or distributors. That is, what we may be observing is not third degree price discrimination based on market segmentation, but rather costs differences resulting from second degree price discrimination by producers or distributors. While we do not have data on costs, we can observe the volume sold through each channel.

Figure 4-Retail Sales Volume by Channel

Table 4 shows average monthly sales in cases by retail channel for Sample 1. As is evident from Figure 4, food stores sell by far the most wine followed by liquor stores with drug stores selling the least amount. If the price differentials were based on quantity discounts, then grocery stores should have the lowest prices followed by liquor stores and drug stores. Since drug stores have the lowest observed prices, followed by food and liquor stores, second degree price discrimination on the part of producers or distributors seems unlikely.

Table 2-Regression Results

Sample 1

Drug

Liquor

February

March

April

May

June

July

August

September

October

November

December

Sample 2 Corporate

Non Corporate

Sample 2

Corporate

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

-0.082

-0.066

-0.083

-0.079

-0.07

-0.042

-0.071

-0.084

-0.008

-0.01

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.29]

[0.68]

Adj. R

Corporate

Non Corporate

0.042

0.05

0.047

0.023

0.054

0.03

0.009

-0.038

0.018

-0.143

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.02]*

[0.05]

[0.00]**

[0.01]*

[0.00]**

-0.021

0.018

-0.013

-0.027

0.018

0.019

-0.016

-0.024

0.015

0.037

[0.01]**

[0.11]

[0.14]

[0.07]

[0.14]

[0.50]

[0.07]

[0.09]

[0.22]

[0.17]

-0.006

0.017

0.005

-0.025

0.017

0.021

0.003

-0.022

0.015

0.039

[0.46]

[0.13]

[0.57]

[0.09]

[0.17]

[0.46]

[0.76]

[0.13]

[0.23]

[0.14]

-0.021

0.007

-0.008

-0.051

0.007

0.01

-0.011

-0.05

0.005

0.026

[0.01]**

[0.52]

[0.34]

[0.00]**

[0.57]

[0.72]

[0.22]

[0.00]**

[0.70]

[0.33]

-0.023

0.011

-0.011

-0.053

0.011

0.009

-0.015

-0.052

0.006

0.026

[0.00]**

[0.27]

[0.13]

[0.00]**

[0.29]

[0.71]

[0.04]*

[0.00]**

[0.53]

[0.26]

-0.019

0.006

-0.008

-0.044

0.008

-0.004

-0.013

-0.044

0.001

0.009

[0.01]**

[0.54]

[0.31]

[0.00]**

[0.47]

[0.87]

[0.10]

[0.00]**

[0.91]

[0.70]

-0.028

0.01

-0.015

-0.055

0.01

0.008

-0.02

-0.054

0.004

0.019

[0.00]**

[0.36]

[0.06]

[0.00]**

[0.38]

[0.76]

[0.01]*

[0.00]**

[0.75]

[0.44]

-0.021

0.014

-0.009

-0.045

0.015

0.014

-0.013

-0.044

0.008

0.027

[0.00]**

[0.18]

[0.29]

[0.00]**

[0.21]

[0.60]

[0.12]

[0.00]**

[0.46]

[0.29]

-0.022

0.021

-0.01

-0.049

0.021

0.021

-0.016

-0.049

0.012

0.033

[0.01]**

[0.07]

[0.26]

[0.00]**

[0.09]

[0.44]

[0.07]

[0.00]**

[0.31]

[0.22]

-0.011

0.015

-0.002

-0.033

0.015

0.015

-0.007

-0.033

0.007

0.028

[0.16]

[0.20]

[0.86]

[0.02]*

[0.24]

[0.58]

[0.40]

[0.02]*

[0.55]

[0.29]

-0.009

-0.003

0

-0.029

-0.004

0.005

-0.003

-0.029

-0.007

0.017

[0.26]

[0.79]

[0.98]

[0.04]*

[0.72]

[0.86]

[0.72]

[0.04]*

[0.55]

[0.52]

-0.008

-0.014

-0.002

-0.02

-0.015

-0.008

-0.001

-0.019

-0.013

-0.003

[0.31]

[0.23]

[0.83]

[0.16]

[0.23]

[0.78]

[0.88]

[0.17]

[0.27]

[0.90]

Lxcases

2

Corporate

Non Corporate

[0.00]**

Dxcases

Obs

Sample 2

Sample 1

Cases

Constant

Sample 1

Non Corporate

0

0

0

0

[0.00]**

[0.00]**

[0.00]**

[0.01]*

0

0

0

0

[0.00]**

[0.00]**

[0.00]**

[0.00]**

0

0

0

0

[0.00]**

[0.00]**

[0.00]**

[0.00]**

2.133

2.02

2.078

2.3

1.994

2.173

2.115

2.33

2.017

2.202

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

[0.00]**

79835

28435

60949

18886

24211

4224

60949

18886

24211

4224

0.02

0.02

0.02

0.01

0.02

0.01

0.04

0.03

0.06

0.08

Absolute value of t-statistics in brackets * significant at 5% level; ** significant at 1% level

CORPORATE OWNERSHIP What is the source of the observed price differentials? To investigate the nature of retail channel price differentials, we pose a few questions. For instance, does corporate ownership of a wine affect the price differentials across retail channels? This would occur if, for example, large corporate wineries have greater marketing capabilities and/or influence over distributors and retailers to engage in effective price discrimination. We test this hypothesis by examining the price differentials between wines owned by large corporations with those that are not. We define corporate ownership as those wines owned by the largest wineries by case volume. Specifically, corporate wineries are those owned by corporations selling over one million cases of wine in 2010 among all wineries held by that corporation. Corporate owners include: E&J Gallo, The Wine Group, Constellation, Fosters/Treasury Wine Estates, Trinchero Family, Wj Deutsch, Banfi Vintners, Ste Michelle Estates, Diageo, Jackson Family Wines, Palm Bay International, Bronco Wine Company and Brown-Forman. Our regression model is modified to, Priceijt =  0  1 Drugit   2 Liquorit    Month  Corporate uijt Where “Corporate” is a vector including an indicator for corporate affiliation along with variables accounting for the individual effects of corporate affiliation on drug store and liquor store prices. The regression results analysing the retail channel price differential between corporate and non-corporately owned wineries are shown in columns (3) through (6) of Table 2. For Sample 1, the regression results show that corporately owned wines sold at drug stores were priced approximately 7.9% below grocery store prices, while non-corporately owned wines sold

at drug stores were approximately 7% below grocery store prices. While the differences between drug store prices and grocery store prices are statistically significant, the difference between corporate and non-corporate drug and grocery store prices is not statistically significant. For liquor stores, corporately owned wines were priced approximately 4.7% above grocery store prices, while for non-corporately owned wines, liquor store prices were approximately 2.3% above grocery store prices. For Sample 1, liquor store prices were statistically greater than grocery prices and corporate price differentials were statistically greater than and the non-corporate price differentials. For Sample 2, drug store prices were approximately 7% below grocery store prices for corporately owned wines and 4.2% below grocery store prices for non-corporately owned wines. These price differentials are statistically significant, and the difference between corporate and non-corporate differentials is significant at the five percent level. For liquor stores, corporately owned wines were priced are approximately 5.4% above grocery prices while non-corporately owned wines were priced 3% above grocery store prices. Once again, the price differentials between liquor stores and grocery stores, and corporate and noncorporate ownership differentials are all statistically significant. The regression results from both samples indicate that price differentials for corporately owned wines are greater than those for non-corporately owned wines. That is, corporately owned wines are priced lower at drug stores and greater at liquor stores.

SALES VOLUME Are price differentials greater for higher volume wines than lower volume wines? Just as firms will charge more to those with a greater willingness to pay, for those with a low willingness to pay, sellers reduce price and accept a lower profit on positive sales to these marginal consumers rather than lose those sales altogether. As a result, price discrimination effectively works to expand sales and market share. To account for sales volume, we modify the model. Priceijt =  0  1 Drugit   2 Liquorit    Month  Corporate  Volume  uijt

Where θ is a vector including a volume variable and interactions among volume and channel.

Figure 5-Sample 1

Figure 6-Sample 2

Regression results for the model including cases volume are shown in Table 2. For drug stores, average prices are still lower than grocery store prices for both corporate and noncorporately owned wines by 7.1% and 8.4% respectively. These results are statistically significant, but not statistically different from each other and are consistent with the earlier estimated price differentials. More importantly, the effect of case volume on drug store prices is negative and statistically significant. For liquor stores the results are mixed, with average prices for corporately owned wines above grocery store prices but prices for non-corporately owned wines less than grocery stores. Once again, however, case volume has the expected effect of increasing liquor store prices relative to grocery store prices. Figures 2 and 3 provide an unambiguous illustration of the relationship between case volume and price for Samples 1 and 2 respectively. Log prices are indexed to zero to more clearly show the relative effects of case volume on log price across retail channel. As is clear in the Figures, in both samples, as

monthly volume increases, drug store prices decrease, liquor store prices increase and grocery store prices decrease. As a reference, Table 3 shows a sample of prices for some of the top selling wines across all three channels. Most items are below $10 per bottle. Note that for Robert Mondavi Private Selection chardonnay, there is no difference in the drug-grocery store price, but there is a significant difference in liquor-grocery store price. Among the high volume wines, the Barefoot wines, produced by Gallo, do not appear to follow the pattern of price differentials observed by the others, and in fact are priced lowest at liquor stores. The higher priced wines include La Crema chardonnay, which is priced in the $17 range, and Chateau St. Michele, which is priced below $10 at drug stores, over $11 at grocery stores and over $12 at liquor stores. Table 3 Sample of Price Differentials for High Volume Wines

market Item Beringer California Collection Chardonnay Barefoot Cabernet Sauvignon Barefoot Chardonnay Barefoot Merlot Barefoot Pinot Grigio Chateau St. Michelle Riesling Columbia Crest Chardonnay La Crema Chardonnay Meridian Chardonnay Robert Mondavi Private Select Chardonnay

Drug

Food

Liquor

$5.32 $6.15 $6.15 $6.14 $6.29 $8.15 $6.28 $17.36 $7.27 $9.47

$5.86 $6.19 $6.17 $6.18 $6.22 $11.77 $7.00 $17.78 $7.45 $9.47

$6.31 $5.95 $5.84 $5.92 $5.94 $12.37 $7.32 $18.33 $7.49 $10.11

PRICE DIFFERENTIALS BY MARKET SEGMENT Because the wine sector is highly segmented, we examine channel price differentials by price segment. We construct three wine price segments: Under $10 per bottle, $10 to under $20 per bottle and over $20 per bottle. Figures 4 and 5 show the distribution of mean monthly cases sold through each channel by sample. Figure 7-Sample1: Case Volume by Price Segment

Under $10

0

1,000

2,000

3,000

$10 to under $20

DRUG

FOOD

LIQUOR

DRUG

FOOD

LIQUOR

0

1,000

2,000

3,000

Over $20

DRUG

FOOD

LIQUOR

Figure 4 clearly shows that food or grocery stores have the highest volume of sales across all price segments in Sample 1. However, moving up price segments, drug and food stores decline in sales volume while liquor stores increase sales volume. This trend is even more pronounced in Figure 5 showing the Sample 2 distribution of case volume.

Figure 8-Sample2: Case Volume by Price Segment

$10 to under $20

0

2,000

4,000

6,000

Under $10

DRUG

FOOD

LIQUOR

DRUG

FOOD

LIQUOR

0

2,000

4,000

6,000

Over $20

DRUG

FOOD

LIQUOR

In fact, in Sample 2, liquor store sales surpass grocery store sales in the over $20 per bottle price segment. The regression results by price segment are shown in Table 3. For the largest market segment, wines under $10, wines sold at drug stores were priced from 4.8% and 5.7% lower than the same wines sold at grocery stores. These results are statistically significant and consistent with our earlier results. In addition, corporate affiliation reduces drug store prices another 1% below grocery store prices, but those results are not statistically significant. For wines sold at liquor stores in the same under $10 market segment, prices were 3.4% to 4% more than the same wines sold at grocery stores. For large corporately owned wine, prices were another 1.8% more than grocery store prices but these results are only statistically significant for Sample 1 and not Sample 2.

Table 4-Price Differentials by Market Segment

Sample 1 Under $10

Drug

Sample 2 $10-Under $20

-0.057 -0.001 [0.00]** [0.83] Liquor 0.034 0.012 [0.00]** [0.02]* Corporate -0.104 0.003 [0.00]** [0.45] CorporatexDrug -0.007 0.021 [0.30] [0.00]** CorporatexLiquor 0.018 0.002 [0.01]* [0.73] February -0.004 -0.007 [0.53] [0.28] March 0.002 -0.008 [0.72] [0.22] April -0.006 -0.011 [0.32] [0.11] May -0.004 -0.006 [0.44] [0.27] June -0.002 0.007 [0.75] [0.28] July 0 -0.011 [0.98] [0.09] August -0.003 -0.007 [0.64] [0.30] September 0.001 -0.006 [0.90] [0.35] October 0.01 -0.009 [0.11] [0.21] November 0.001 -0.005 [0.87] [0.47] December 0.001 0 [0.84] [0.98] Constant 2.005 2.538 [0.00]** [0.00]** Observations 58121 19415 2 Adj. R 0.05 0 Absolute value of t-statistics in brackets * significant at 5% level; ** significant at 1% level

$20 & Over

Under $10

$10-Under $20

$20 & Over

-0.043 [0.00]** 0.002 [0.89] -0.059 [0.00]** 0.019 [0.28] -0.017 [0.31] -0.01 [0.53] -0.023 [0.17] -0.034 [0.04]* -0.028 [0.04]* -0.036 [0.02]* -0.029 [0.07] -0.007 [0.65] -0.029 [0.09] -0.02 [0.25] -0.019 [0.23] 0.003 [0.84] 3.217 [0.00]** 2299 0.04

-0.048 [0.00]** 0.04 [0.00]** -0.106 [0.00]** -0.01 [0.44] 0.017 [0.19] 0.008 [0.38] 0.012 [0.20] 0.003 [0.76] 0.008 [0.33] 0.007 [0.38] 0.008 [0.38] 0.01 [0.24] 0.016 [0.08] 0.013 [0.15] -0.001 [0.88] -0.006 [0.51] 1.984 [0.00]** 22823 0.05

-0.009 [0.45] -0.015 [0.18] -0.012 [0.19] 0.005 [0.72] 0.019 [0.13] 0.005 [0.70] 0.007 [0.58] 0.007 [0.59] 0.011 [0.28] 0.02 [0.07] 0.011 [0.32] 0.018 [0.12] 0.015 [0.22] 0.011 [0.38] 0.005 [0.68] 0.005 [0.67] 2.519 [0.00]** 5368 0

0.021 [0.07] -0.069 [0.00]** 0.086 [0.00]** 0.00 [.] 0.051 [0.00]** 0.027 [0.14] 0.027 [0.14] -0.003 [0.88] 0.008 [0.61] 0.023 [0.18] 0.017 [0.33] 0.032 [0.07] 0.025 [0.17] 0.036 [0.06] 0.018 [0.35] -0.011 [0.57] 3.067 [0.00]** 244 0.45

For wines priced $10 to under $20, the regression results show that drug store prices were lower than grocery store prices, but the results are small and not statistically significant in either sample. Moreover, corporate affiliation appeared to have little effect on the price of wines sold in drug stores in this market segment. For the $10 to under $20 per bottles, market segment, wines sold at liquor stores were priced 1.2% more than grocery stores in Sample 1. While these results were statistically significant, results for Sample 2 were not, nor were the effects of corporate affiliation. For the smallest market segment, wines priced $20 and over, the regression results for Sample 1 show that drug store prices were 4.3% lower than grocery store prices. These results are statistically significant, but the results for Sample 2 are not. For wines priced over $20 at liquor stores, the results are mixed. Sample 1 indicates that liquor store prices were greater than drug store prices, but the results are not statistically significant. For Sample 2, the regression results indicate that liquor store prices were 6.9% lower than grocery store prices. These results are statistically significant. Interestingly, for most of the price segments in both samples, the effect of corporate affiliation becomes statistically insignificant.

PRICE ELASTICITIES The data examined shows a clear and consistent pattern of price differentials across retail channel. Specifically, wines at drug stores are consistently priced lower than the same wines at grocery stores, while wines at liquor stores are consistently priced above those same wines at grocery store prices. Moreover, we have shown that these price differentials are greater for high volume wines than for lower volume wines.

Opportunities for price discrimination exist when market segments differ in their price elasticities of demand. Theoretically, if PDrug|εLiquor|. The previous section demonstrated that the price of a bottle of wine at a drug store is less than the same bottle of wine at a grocery store, which is less than the same bottle of wine at a liquor store. We turn now to estimating the price elasticity of demand for wine across all three channels. We use a standard double log demand specification of the form: Casesjt =  0  1 Pr ice tj   2 Incomei    Month u jt to estimates the price elasticity of demand for wine at time “t” for each of the three channels. The regression results are shown in Table 3. Table 5-Price Elasticities of Demand

Price Income February March April May June July August September

Drug -0.985 [0.00]** 3.525 [0.00]** -0.087 [0.09] -0.086 [0.09] -0.115 [0.02]* -0.19 [0.00]** -0.201 [0.00]** -0.237 [0.00]** -0.194 [0.00]** -0.227 [0.00]**

Sample 1 Grocery -0.577 [0.00]** 1.142 [0.02]* -0.125 [0.00]** -0.095 [0.02]* -0.088 [0.03]* -0.133 [0.00]** -0.153 [0.00]** -0.149 [0.00]** -0.118 [0.00]** -0.137 [0.00]**

Liquor -0.094 [0.00]** -2.444 [0.00]** -0.432 [0.00]** -0.414 [0.00]** -0.367 [0.00]** -0.45 [0.00]** -0.419 [0.00]** -0.446 [0.00]** -0.435 [0.00]** -0.38 [0.00]**

Drug -0.902 [0.00]** 0.871 [0.19] -0.189 [0.00]** -0.172 [0.00]** -0.181 [0.00]** -0.265 [0.00]** -0.32 [0.00]** -0.364 [0.00]** -0.342 [0.00]** -0.338 [0.00]**

Sample 2 Grocery -0.465 [0.00]** -1.303 [0.03]* -0.184 [0.00]** -0.17 [0.00]** -0.159 [0.00]** -0.207 [0.00]** -0.249 [0.00]** -0.24 [0.00]** -0.226 [0.00]** -0.235 [0.00]**

Liquor -0.283 [0.00]** -4.609 [0.00]** -0.489 [0.00]** -0.465 [0.00]** -0.45 [0.00]** -0.503 [0.00]** -0.487 [0.00]** -0.499 [0.00]** -0.504 [0.00]** -0.459 [0.00]**

October

-0.205 -0.156 -0.392 -0.295 -0.207 -0.438 [0.00]** [0.00]** [0.00]** [0.00]** [0.00]** [0.00]** November -0.193 -0.151 -0.381 -0.211 -0.173 -0.376 [0.00]** [0.00]** [0.00]** [0.00]** [0.00]** [0.00]** December -0.034 0.008 -0.118 -0.033 -0.006 -0.115 [0.49] [0.84] [0.01]** [0.57] [0.91] [0.07] Constant -26.261 -1.949 28.487 -0.489 21.347 48.896 [0.00]** [0.65] [0.00]** [0.94] [0.00]** [0.00]** Observations 26479 26664 26692 9396 9502 9537 2 Adj. R 0.07 0.04 0.01 0.1 0.04 0.03 Absolute value of t-statistics in brackets * significant at 5% level; ** significant at 1% level

Table 3 shows that consumers at drug stores are more price elastic than consumers at grocery store, who are more price elastic than consumers at liquor stores. The estimated price elasticities of demand are all statistically significant across both samples. Additionally, the relationship among the price elasticities is consistent with pattern of price differentials and is consistent with the condition required for price discrimination. That is, |εDrug|>|εGrocery| >|εLiquor|. DISCUSION We find that retail channel is used as an effective means of price discrimination. Just as coupons and rebates offer discounts to low willingness to pay consumers whose opportunity cost of time is less than the associated savings or rebate, retail channel provides a similar opportunity for discriminatory pricing based on consumer self-selection. We show that drug stores offer a selection and prices consistent with lower income and low opportunity cost shoppers. This is illustrated by Figures 4 and 5 which show that the sales of wine at drug stores decrease as you move up each price segment. Based on sales on prices, we show that drug

stores consist of a specific market segment, namely consumers of value wines. This is further illustrated by the type of shoppers at drug stores. Because drug stores carry fewer items in total than grocery stores, drug store shoppers may represent more frequent “fill in” or quick shopping trips consisting of smaller “shopping baskets” than grocery store shoppers. As Kahn and Schmittlein (1989) note, “quick [shoppers] have lower incomes, more older males and older females and more retired people” than regular shoppers. In addition to demographic characteristics, shopper intent may play a role in the lower prices observed at drug stores. If the primary purpose of shopping at a drug store is to purchase items other than wine, then offering a price lower than other retail outlets is an efficient means of enticing consumers. This is consistent with the behavior observed by Bucklin and Lattin (1991) who show that consumers are more price responsive when purchases are unplanned. The greater price elasticity of demand estimated for drug store shoppers supports both these explanations. Grocery stores are the largest of the three retail channels for wine but one whose market share falls as you move up the price segments. In contrast to drug store shoppers, wine, on average, can assumed to be part of a larger shopping list or basket of goods that are purchased on more infrequent shopping trips. Such shoppers, according to Kahn and Schmittlein (1989) tend to be from larger families with higher incomes. We show that these larger, high income families, who have a higher opportunity (or search) cost of shopping, pay more for an identical bottle of wine than drug store shoppers. While the typical grocery store shopper may not be aware of the lower prices of wine at a drug store, it is not clear that this knowledge would result in an additional shopping trip. Even if the typical grocery store shopper knows that an identical wine can be purchased at a drug store for less, they must weigh the

costs of an additional shopping trip against the associated savings. The costs associated with an additional shopping trip include the time needed to travel to a drug store including parking, the in-store shopping time and the check-out time. Travel time may or may not be a significant cost since many shopping centers have grocery, drug and liquor stores located in the same center or even adjacent to each other. Shoppers must still, however, weigh whatever costs they incur with the potential savings. Given that the potential savings would be approximately 4%to 8%, with the greatest savings being on wines less than $10 per bottle, the savings may not be enough to cover the costs of an additional shopping trip. This relative price inelasticity is consistent with our estimated price elasticity as well as the shopping behavior of those on planned trips observed by Bucklin and Lattin (1991). Finally, we examine liquor stores. In contrast to drug and grocery stores, liquor store sales increase as you move up the price segments. In contradistinction to drug store patrons, consumers at liquor stores are more likely to purchase more expensive wines. Additionally, liquor store patrons are ostensibly category specific shoppers with a specific shopping intent, namely to buy liquor. As a result, as Bell, Ho and Tang (1998) note, while category specific stores reduce the search costs of shopping for that category, this specific shopping intent increases insensitivity to price. This is confirmed by our estimated price elasticities of demand for liquor store shoppers as well as our results that show liquor store prices are on average greater than drug and grocery store prices for the same wine. SUMMARY Increased competition both domestically and internationally has led firms to seek new sources of revenue. Price discrimination across all dimensions, time, space, demographic

characteristics and offline versus online is one means of extracting surplus from consumers. The proliferation of social network coupon sites is an illustration of this. We show that use of retail channel is an effective means of price discrimination based on demographic self-selection and shopping intent. While the current research shows a clear and consistent pattern of discriminatory price differentials for a single category of goods, wine. There is no reason to believe that other categories are not following similar behavior. If not, we show that the opportunity to do so exists.

Works Cited Bell, David R., Ho, Teck-Hua and Christopher S. Tang (1998). “Determining Where to Shop: Fixed and Variable Costs of Shopping.” Journal of Marketing Research, Vol. 35, No. 3 (August) 352-369. Blattberg, Robert, Buesing, Tom, Peacock, Peter, and Subrata Sen (1978). "Identifying the Deal Prone Segment." Journal of Marketing Research, Vol. 15 (August), 369-77. Blattberg, Robert and Subrata Sen (1974). “Market Segmentation Using Models of Multidimensional Purchasing Behavior.” Journal of Marketing, Vol. 38 (October), 17-28. Blattberg, Robert and Subrata Sen (1976). “Stochastic Brand Choice Models.” Journal of Marketing Research, Vol. 13 (February), 34-45. Bucklin, Randolph E. and James M. Lattin (1987). “A Two-State Model of Purchase Incidence and Brand Choice.” Marketing Science, Vol. 10, No. 1 (Winter), 24-39. Chakravarthi Narasimhan (1984). “A Price Discrimination Theory of Coupons.” Marketing Science, Vol. 3, No. 2 (Spring), 128-147. Cort, Stanton G. and Luis V. Dominguez (1977). “Cross-Shopping and Retail Growth.” Journal of Marketing Research, Vol. 14 (May), 187-192. Gerstner, Eitan, Hess, James D. and Duncan M. Holthausen (1994). “Price Discrimination Through a Distribution Channel: Theory and Evidence, American Economic Review, Vol. 84, No. 5 (December) 1437-1445. Hoch, Stephen J., Kim, Byung-Do, Montgomery, Alan L., and Peter E. Rossi (1995). “Determinants of Store Level-Level Price Elasticity.” Journal of Marketing Research, Vol. XXXII (February), 17-29. Kahn, Barbara E. and David C. Schmittlein (1989). “Shopping Trip Behavior: An Empirical Investigation.” Marketing Letters, Vol. 1, No. 1, 55-69.

K. Sridhar Moorthy (1984). “Segmentation, Self-Selection, and Product Line Design.” Marketing Science, Vol. 3. No. 4 (Autumn) 288-307.

Park, Whan C., Iyer, Easwar S., and Daniel C. Smith (1989). “The Effects of Situational Factors on In-Store Grocery Shopping Behavior: The Role of Store Enviornment and Time Available for Shopping.” Journal of Consumer Research, Vol. 15 (March), 422-433.

Pigou, Arthur .C. The Economics of Welfare. London: MacMillan, 1920. Robinson, Joan. The Economics of Imperfect Competition. London: MacMillan, 1933. Salop, Steven and Joseph Stiglitz (1977). “Bargains and Ripoffs: A Model of Monopolisticallty Competitive Price Dispersion.” Review of Economic Studies, Vol. 44, No. 3 (October) 493510. Severin Borenstein (1985). “Price Discrimination in Free-Entry Markets.” Rand Journal of Economics, Vol. 16, No. 3 (Autumn) 380-397. Thomas J. Holmes (1989). “The Effects of Third-Degree Price Discrimination in Oligopoly.” American Economic Review, Vol. 79, No. 1(March) 244-250.