drinking cheaply: the demand for basic wine in italy -...
TRANSCRIPT
1
Drinking cheaply: the demand for basic wine in Italy 1
Luigi Cembalo, Francesco Caracciolo, Eugenio Pomarici 2
Department of Agricultural Economics and Policy - University of Naples Federico II, Italy 3
Corresponding author: Luigi Cembalo 4
Via Università 96, 5
80055 Portici (Naples), Italy 6
e-mail: [email protected] 7
Tel. +39 081 2539065 8
Fax. +39 081 7755143 9
10
Abstract. The wine market has evolved dramatically over the last three decades. The premium wine 11
segment has expanded significantly to the detriment of basic wines. Nevertheless, in traditional 12
wine producing and consuming countries, inexpensive wines still account for a large market share, 13
both in volume and value. Marketing strategies for such wines are changing in an attempt to tap this 14
increasingly crowded market segment. Despite its importance, the basic wine segment has not been 15
studied in depth, and is often assumed to have no product differentiation. This paper tried to 16
ascertain the existence of a possible degree of heterogeneity within non-premium wines and to 17
measure, by means of elasticity computation, the relationships among categories of wines 18
aggregated with criteria that go beyond price. A demand system (censored QAIDS) was estimated, 19
using a statistically representative panel of 6,773 Italian households, to see to what extent, if any, 20
substitution occurs in home consumption of basic wines, which is the main channel of distribution 21
of inexpensive wines in Italy. Although price is an important lever in supply policies, our results 22
also suggest the importance of packaging, such as carton as an alternative to glass. 23
24
Keywords: censored demand system; elasticity computation; Italian wine sector; basic wine; 25
QAIDS. 26
JEL: Q13; D12; C34 27
2
28
1. Introduction 29
In the past three decades or so, the world wine market has experienced dramatic upheavals. One 30
of the major changes has involved the structure of wine demand (Labys & Cohen, 2006; Hannin et 31
al., 2010; Anderson and Nelgen, 2011). In the larger traditional producing and consuming countries, 32
Italy and France, the annual per capita consumption of wine was, until the 1970s, higher than 100 33
litres (Corsi et al., 2004) while, in the last three decades, it has declined to between 50 and 40 litres 34
(OIV, 2012). During the same time span in the United States the increase in the number of wine 35
drinkers, combined with a higher per capita consumption, has pushed the aggregate national wine 36
consumption to the same level as Italy and France. Indeed, while during the ’70s wine consumed in 37
France and Italy accounted for 40% of world consumption, nowadays the two countries account for 38
about 25% (Anderson & Nelgen, 2011). 39
In traditional producing and consuming countries the dramatic change in wine quantity consumed 40
has been accompanied by major changes in consumption habits: from a necessary component in the 41
daily diet to non-essential to a meal (Mäkelä et al., 2006). Consumption habits in traditional and 42
new consuming countries are converging. Expensive wines have gained a considerable market 43
share, becoming an accessible good no longer confined to a restricted elite of consumers (Ritchie, 44
2009). At the same time, less expensive wines are reaching a place in the market similar to that of 45
other Fast-Moving Consumer Goods. 46
In order to have a clearer overview of the wine market, its characteristic high price range also 47
deserves mention. This peculiarity has called for the development of commercial terminology to 48
define different product categories based on specific price ranges. In the wine trade it is common to 49
distinguish wines into two broad classes according to price category: non-premium, less expensive 50
wines with basic quality characteristics; and premium, more expensive wines with complex quality 51
characteristics and a high-value image (Anderson & Nelgen, 2011). In light of the absolute increase 52
and relative importance of more expensive wines, a more detailed classification has been made in 53
3
the wine business, dividing premium wines into sub categoriesi (Ernst and Young Entrepreneurs, 54
1999; Rabobank, 2003; Costanigro et al., 2007). Nevertheless, a pan-national segmentation of the 55
wine market is complex and any estimation of the market share in each category, in aggregate terms 56
or per country, is no trivial task (Steenkamp & Ter Hofstede, 2002). However, estimation has been 57
attempted according to a three-category classification of wine supply for 2009: non-premium, 58
commercial premium and superpremium (Anderson & Nelgen, 2011)ii. Non-premium wines 59
account for only one-seventh of the global wine trade in value (US$7 billion) but almost half in 60
volume terms. It seems evident that non-premium winesiii represent a sizeable part in the world 61
wine market. Together with the search for increasing value for products, this is one of the main 62
reasons why suppliers are still hugely interested in the market for non-premium wines (Ritchie, 63
2009). 64
Despite the importance of non-premium wines, very few studies have specifically investigated this 65
segment (Torrisi et al., 2006). One possible reason explaining this lack of coverage could be the 66
presumption that, in a wine category with a narrow price range, there is almost perfect homogeneity 67
due to wines with simple intrinsic attributes, little quality complexity, and hence not much 68
differentiation. However, observation of the current supply shows a wide variety of products and 69
various supplier’s marketing styles. Therefore, in-depth investigation of basic wine is of paramount 70
importance to better understand how key parameters combine. A study is thus called for to explore 71
the needs and motivations sustaining the demand for non-premium wines. Beside a large number of 72
studies concerning the demand side on premium wines, or alcohol in general (Gallet, 2007; Davis et 73
al., 2008; Gil & Molina, 2009; Panzone, 2012), the literature on demand for basic wines appears to 74
be almost completely lacking. In this perspective, the objective of this paper is to ascertain the 75
existence of a possible degree of heterogeneity within non-premium wines and to measure, by 76
means of elasticity computation, the relationships among categories of wines aggregated with 77
criteria that go beyond price. 78
4
Studies of non-expensive wines have to be country-specific for two reasons. First, markets for non-79
premium wines have average selling prices that differ because of different import tariffs and 80
consumer taxes (Anderson & Nelgen, 2011); second, consumer product perception varies on a 81
country by country basis (Mäkelä et al., 2006). The focus of this study was Italyiv. Together with 82
France, Italy has the largest traditional wine consumption market, and the variety of strategies 83
adopted by suppliers is one of the widest: suppliers seek to capture customer interests by looking at 84
all intrinsic and extrinsic product attributes and working on communication and distribution 85
(Bernetti et al., 2006). Although all these aspects of supply are important (Fraser, 2005; Amadieu & 86
Viviani, 2011), pricing in a market with very small margins assumes great importance and the 87
availability of appropriate information to support price fine-tuning is therefore crucial. 88
The empirical strategy adopted was based on two steps. The first consisted in collecting information 89
directly interviewing key informants to delineate the driving factors of supply on Italian retail 90
shelves. Such factors allowed us to classify wines within the non-premium category using wine 91
business driven criteria (e.g. labelling, types of packaging, and price). The second step was to 92
estimate a demand system on a statistically representative sample of Italian households provided by 93
AC-Nielsen. Demand system estimates based on micro data may suffer several biases. The main 94
risks come from the violation of theoretical regularity restrictions (Barnett & Serletis, 2008) and 95
from sample selection bias due to only a fraction of the population that has nonzero consumption 96
for the items under study. Our model explicitly avoids the latter drawback, estimating a two-step 97
censored demand system (Shonkwiler & Yen, 1999) based on the quadratic AIDS (Banks et al., 98
1997). 99
100
2. The wine market in Italy 101
2.1 Essential data for the Italian wine sector 102
In order to illustrate the economic relevance of the overall Italian wine sector, a few essential 103
figures are provided. Total Italian wine output has a value of nearly 8.2 billion euros: as a producer, 104
5
exporter and consumer, Italy is top of the world ranking. In 2011, Italian wine production was 44 105
million hectolitres, about 23 million of which was exported, for a value of 4.5 billion euros. Imports 106
are rather small: in 2011 Italy imported about 2.3 million hectolitres, for a value of 300 million 107
euros (Mediobanca, 2012)v. 108
Italian wine production represents about 15% of world supply in volume and 13% in value. 109
Italian wine exports represent 24% of world exports in volume and 18% in value (Anderson & 110
Nelgen, 2011; OIV, 2012). Considering only non-premium wines Italy contributes, in value terms, 111
about 18% to global supply and 16% to world export; such percentages demonstrate to what extent 112
the Italian wine industry has invested in the non-premium wine business. 113
Domestic consumption accounts for about 23 million hectolitres consumed by a substantial 114
wine-drinking population, representing 54% of Italy’s total population (ISTAT, 2011; OIV, 2012). 115
Wine purchased for at-home drinking (WAH) accounts for the largest slice of the wine market: 116
about 70% of the total consumption in volume, and 30% in expenditure. WAH is supplied by many 117
channels but the most important, in terms of volume, are the supermarket chains which sell about 118
70% of wine (Mediobanca, 2011; Pomarici et al., 2012). The remaining percentage is purchased in 119
neighbourhood grocery stores and wine shops, while direct purchase at wineries occurs especially in 120
rural areas. The difference between the consumption figures in volume (70%) and in value (30%) 121
highlights the extent to which cheap wine consumption is concentrated at home. 122
123
2.2 Data description and categorization of WAH consumption of wine in Italy 124
Wine for at-home drinking can be investigated using scanner data from consumer panels. The 125
data used in this study were provided by the AC-Nielsen home scan, a statistically representative 126
panel of Italian households. Households involved in the Nielsen panel regularly record their 127
purchases through a scanner (ConsumerScan). The data analysed thus consist of household wine 128
expenditure at grocery stores (including neighbourhood and specialized grocery stores, convenience 129
stores, small and large retail outlets), and are representative of Italian consumption of WAHvi. The 130
6
information collected concerns value (euro), volume (litres) and the main extrinsic attributes of 131
WAH purchases made by 6,773 Italian households (for a total of 71,760 purchases of 6,251 132
different types of wines from 956 wine producers) during 2010 (from January to December). The 133
6,251 types of wines recorded in the AC-Nielsen database need to be aggregated in fewer 134
categories. However, any aggregation of wines belonging to the non-premium wines needs the 135
criteria to be identified as objectively as possible. This applies even more if one considers that this 136
segment is of great complexity due, for example, to a broad variety of brands, packaging, 137
production areas, possible origin labels and grape variety certifications. Nevertheless, the literature 138
contains no indication on non-premium wine categorization. Our study attempts to identify criteria 139
for aggregating non-premium wines which allow homogeneous categories to be obtained. The 140
empirical approach used was to proceed according to the same principles with which the supply of 141
more economical wines is structured on the shelves of large retailers, which is the main site of 142
purchase of packaged economical wines. Interviews were carried out with wine marketing experts. 143
The sales director of Gruppo Italiano Vini S.p.A. (top Italian group by sales) and the director of 144
commercial relations with large retailers at Marchesi dè Frescobaldi (one of the top 15 companies 145
in Italy by turnover) were involved in an in-depth semi-structured interview in autumn 2011. The 146
interviews, carried out individually with each of the two interviewees, were preceded by sending 147
out the subject of the interview. On the basis of the interview results we developed an aggregation 148
scheme of the single non-premium products sold in Italy. The results of this scheme were submitted 149
to a third expert for final verification, director of the Gruppo CEVICO of relations with large 150
distributors (Cevico is one of the top 15 Italian companies by turnover). The results can be 151
summarized as follows. Key informants indicated the threshold that defines non-premium wines as 152
3 euro/litrevii. Starting from this information, the distribution of WAH consumption by price, 153
divided into non-premium and premium, from AC-Nielsen home scan data, is shown in table 1. 154
Wine sold under 3 euro/litre represents nearly 60% of the total value and about two-thirds of the 155
total volumeviii. 156
7
Table 1. Premium and non-premium Italian distribution of wines 157
Price classes (€/litre)
Share on
expenditure (%)
Share on
volume (%)
Non-premium wines (<3 euro) 58 67
Premium wines (>3 euro) 42 33
Total 100 100
Source: authors’ calculations based on data from A.C Nielsen (2010) 158
In addition, the key informants all agreed that the critical factors behind the diversification of the 159
supply of non-premium wines on the shelves are as follows: (i) packaging; (ii) branding; (iii) price. 160
Packaging is currently an element of supply diversification as it is the field where many producers 161
have explored the way to add value to wine with significant innovations able to improve the service 162
delivered to the client and, at times, perception of sustainability. Such innovations have concerned 163
the material of containers, mainly introducing the carton, as an alternative to glass bottles. As for 164
the branding criterion, the key informants all agreed that a shelf must contain the market leader and 165
the private labels, together with a set of wines with only locally recognizable brands. Concerning 166
the third criterion, the shelf is formed to cover all price ranges in order to give any consumer the 167
possibility of choosing a product based on his/her own preferences/budget. In particular, the range 168
has to contain wines with a unit price ranging from 2 to 3 euro/litre, with characteristics that 169
emulate premium wines, and wines with a unit price below 2 euro/litre, which typically have the 170
profile of convenience products. The supply of non-premium wines on the shelves also appears 171
diversified for other elements such as wine colour and certification category (e.g. Denomination of 172
Controlled Origin or DOC, and Typical Geographic Indication or IGT). However, these elements 173
were not indicated by the experts as being major factors in forming the assortment of winesix. 174
The categorization scheme potentially yields 18 categories of wines: 2 packaging types (glass and 175
carton); 3 levels of branding (market leader, private label, and others); 3 price levels (below 2 176
€/litre, between 2 and 3 €/litre, above 3 €/litre). However, the information gathered by the experts 177
8
has to be matched with, or confirmed by, the actual consumption data available in the AC-Nielsen 178
dataset. Among the set of types/levels, the one that has to be explicitly identified in the AC-Nielsen 179
dataset is the market leader. According to the market share in volume, Caviro can doubtless be 180
considered the market leader with the highest market share in volume terms (12.5%); all other 181
competitors show a market share lower than 1.51% (Table 2). Caviro mainly promotes two brands, 182
Tavernello and Castellino, both sold in cartons, with an intense pull strategy, using TV and media 183
advertising extensively. The main theme in advertising is the integrity and safety of the wine as a 184
result of strict controls over every link in the supply chain, from vineyard management and 185
winemaking through packaging and marketing, ensuring consistent quality throughout the entire 186
process. 187
188
Table 2. Top four firms of the Italian wine-at-home market 189
Wine brand Market share
in volume (%)
Mean
price
(€/litre)
Caviro 12.48 1.42
Conad 1.51 1.39
Soldo 1.41 1.48
San Matteo 1.32 1.58
Source: authors’ calculations based on data from A.C Nielsen (2010) 190
191
Private labels are those supplied by using the names of big distribution chains. The number of 192
private labels available in the Nielsen database is 30, and the brands with the highest market share 193
are Conad, Carrefour, COOP, GS, and LD. The 30 private labelled wines account for 7.89% in 194
terms of market share (in volume terms) and are sold at an average price of 0.98 euro/litre, mainly 195
in carton packaging. A summary of statistics from the AC-Nielsen dataset, organized by criteria 196
9
suggested by experts, is shown in table 3. Some categories of wines are represented in small 197
percentages. The criterion used to aggregate products was to consider, as a single category, those 198
products represented in at least 2% in terms of market share (in volume terms). Those categories 199
below that threshold were aggregated into the neighbouring category. 200
201
Tab. 3 - Summary statistics of wines organized by experts’ 202
suggested criteria (volume %) 203
Price categories (€/litre)
< 2 2 < . < 3 > 3
CARTON
Caviro* 11.79 - -
Private Label 6.08 - -
Other 14.86 0.02 0.02
GLASS
Caviro* 0.69 - -
Private Label 1.81 0.20 0.60
Other 24.10 12.64 27.19
Total 59.33 12.86 27.81
Source: authors’ calculations based on data from AC-Nielsen (2010) 204
* Caviro is the market leader that sells wines almost exclusively in carton packaging 205
206
From the merging of information collected from experts and the Nielsen data we aggregated the 207
products into categories. The outcome is an ex-ante categorization of the non-premium wine 208
market. The whole set of wines contained in the AC-Nielsen dataset was therefore subdivided into 209
six categories. Thus the WAH market becomes divided into the following categories: 210
wines sold in carton at an average price below 2 euro/litre: 211
10
- Market Leader (ML): Caviro wines; 212
- Private Labels (PL): private label brand wines; 213
- Other wines in Cartons (OC): wines with no specific or recognisable brand; 214
wines sold in glass (bottle of 0.75 litre), one for each of the price categories: 215
- Other wines below 2 euro/litre (BG: Basic Glass); 216
- Other wines between 2 and 3 euro/litre (TB: Top Basic); 217
- Other wines over 3 euro/litre (PW: Premium wines). 218
This categorization permits analysis of brand-specific relationships among basic wines while jointly 219
evaluating the role of packaging and price. Table 4 provides volumes (litres) and expenditure (euro) 220
of wine as grouped according to the above market categorization. 221
Table 4. Consumption, expenditure share and purchase frequency by wine category 222
Mean
expenditure (%)
Household
purch (%)
Mean litre
consumption (%)
Market Leader (ML) 12 32 14
Private Label (PL) 6 22 8
Other wines carton (OC) 12 35 15
Other wines glass below 2 € (BG) 16 41 18
Other wines glass between 2 and 3 € (TB) 12 40 12
Premium wines over 3 € (PW) 42 61 33
Total 100
Source: authors’ calculations based on data from A.C Nielsen (2010) 223
224
3. Empirical model specification 225
In this section we present the empirical model adopted to estimate the demand system. A 226
Quadratic Almost Ideal Demand System (QAIDS) was implemented. The use of a model allowing a 227
more general Engel curve shape than the popular Almost Ideal Demand System (AIDS) of Deaton 228
& Muellbauer (1980) was required. The reasons are well rooted in the literature. In particular, 229
11
Banks et al. (1997) show that the demand for some goods, particularly alcohol and clothing, has a 230
quadratic relationship with the logarithm of total expenditure at higher income levels. For the above 231
reason, they propose a quadratic logarithmic expenditure share system (QUAIDS), directly derived 232
from the AIDS. This specification is spreading rapidly and was used, for example, by Moro and 233
Sckokai (2000) to estimate food expenditure in Italy, and by Blundell and Robin (1999) and Fisher 234
et al. (2001) to estimate food consumption in, respectively, the UK and USA. 235
The QUAIDS stochastic representation of the system of equations for the budget share wi of the ith 236
good (wine) is as follows: 237
∑=
+⎭⎬⎫
⎩⎨⎧
⎥⎦
⎤⎢⎣
⎡+⎥
⎦
⎤⎢⎣
⎡++=
J
ji
iijijii u
am
bamPw
1
2
* )(ln
)()(lnln
pppλ
βγα { Ii ,,1K=∀ 238
where *ia is the intercept of the model expressed as a linear function of some socio-economic and 239
demographic attributes (residence, age, income class). It serves to assess the potential differences in 240
household preferences and behaviour (Deaton and Muellbauer, 1980): 241
∑=
+=K
kkikii Daa
1* δ { Ii ,,1K=∀ 242
(2) 243
where δi,k represents the contribution of the k-th characteristics of the households on the intercept of 244
the i-th equation representing a category of wine. 245
γij are the parameters to be estimated of prices Pj related to the j-th good (wine). βi are the total 246
expenditure (m) parameters to be estimated, while λi represent the quadratic terms of expenditure. 247
ln a(p) has the translog form: 248
∑∑∑= ==
++=n
ij
n
jiij
I
iii PPPa
1 11*0 lnln
21ln)(ln γααp
(3) 249
b(p) is the Cobb-Douglas price aggregator: 250
∏=
=n
ii
iPb1
)( βp
(4) 251
12
Finally, ui are the error terms. 252
Information on prices paid by each household for the i-th wine is provided through unit values 253
(€/litre) in order to generate a volume-weighted average of the category price. As suggested by 254
Deaton and Zaidi (2002), the unavailable prices due to household non-consumption of the listed 255
category were imputed by the median of the reported price by the purchasers, differentiating the 256
value per region of residence and the household’s income class. 257
Due to incomplete information of household consumption and budget allocation in the data 258
generating process (DGP), a consumer multi-stage budgeting process was considered. It implies 259
home consumption of wine as strictly separable from the demand for other goods. This hypothesis 260
justifies the exclusion of expenditure on other goods for which data are not fully available. The 261
assumption is reasonable if the estimation objectives are restricted to understanding consumption 262
substitutability in the same food category: in this case it is non-premium wines. 263
From an empirical point of view, according to the established consensus, demand system estimates 264
based on household cross-sectional data can be cumbersome on several grounds. One of these is 265
that household consumption variables are censored at zero. In demand studies using cross-sectional 266
micro-data the zero-food consumption problem is a frequent issue; only a subset of households 267
shows a positive consumption for the i-th good (wine) during the selected observation period. In the 268
investigated wine demand system, percentages of zero-food consumption (censoring) are 269
substantial: Private Label wines are consumed by 22% of households, Other Wines over 3 €/litre by 270
61%. Since the seminal work of Heien and Wessels (1990), several empirical procedures for 271
censored data have been developed such as those suggested by Perali and Chavas (2000), Golan et 272
al. (2001) and Shonkwiler and Yen (1999). The latter approach is that commonly used in the 273
literature (Akbay et al., 2007). Shonkwiler and Yen (1999) modelled zero-food consumption for a 274
demand system of I equations as below: 275
276
13
iiiiiiiii emww ++Φ= )()();,(* zθzθΓp φτ { Ii ,,1K=∀ (5) 277
and 278
( )
⎥⎦⎤
⎢⎣⎡ −+
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡⎟⎠⎞⎜
⎝⎛−+=
)ˆ()(
ˆ);,(
iiiii
iiiiiii ΦΦmwue
θzzθ
θzzθΓp
φφτ { Ii ,,1K=∀ (6) 279
where Γi is a vector containing system demand parameters, z is a vector of exogenous variables, θi 280
is a conformable vector of parameters, )ˆ( ii θzφ and )ˆ( ii θzΦ are the probability density function 281
(PDF) and the cumulative distribution function (CDF), respectively. Estimation of equation 5 can 282
be performed in two steps, where in the first step the maximum-likelihood probit estimates of 283
)ˆ( ii θzφ and )ˆ( ii θzΦ are obtained using equation 7: 284
y*=(zθi +νi) { Ii ,,1K=∀ (7) 285
with νi the random errors and 286
⎪⎩
⎪⎨⎧
≤
>=
00
01*
*
i
ii yif
yify { Ii ,,1K=∀ (8) 287
and then estimating Γi and τi in the augmented system where 288
wi*=yi wi
{ Ii ,,1K=∀ . (9) 289
Equation 1 therefore becomes: 290
( ) iiiiii
J
j
iijijiii e
am
bamPwy ++Φ
⎪⎪⎭
⎪⎪⎬
⎫
⎪⎪⎩
⎪⎪⎨
⎧
⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
⎥⎥⎦
⎤
⎢⎢⎣
⎡+
⎥⎥⎦
⎤
⎢⎢⎣
⎡++= ∑
=
)()(
ln)()(
lnln1
2
* zθzθppp
φτλβγα { Ii ,,1K=∀
(10)
291
To estimate the above system of I equations, a non-linear Feasible Generalized Least Square 292
(FGLS) (Davidson & MacKinnon, 2004) estimating technique was followed, following the 293
recommendations of Tauchmann (2005): it efficiently manages the heteroskedasticity introduced in 294
the model estimate by the error term when using Shonkwiler and Yen’s (1999) approach. For the 295
14
same reason, as in the single-equation case due to the downward bias of the population variance, 296
standard errors were calculated by using the bootstrap method. Only socio-economic and 297
demographic characteristics were included among the explanatory variables z in the selection 298
equations (eq. 7), as suggested by Yen and Lin (2006). 299
The economic theory provides a set of restrictions that are known as homogeneity, adding-up and 300
symmetry. The symmetry and homogeneity conditions are respectively imposed by: 301
γij= γji ∀ i,j= 1,…, I; (11) 302
and 303
∑ =j
ij 0γ ∀ i,j= 1,…, I. (12) 304
Using the above specifications, the maximum-likelihood estimation of variance-covariance matrix 305
of the residuals is not singular (determinant not equal to zero). Therefore the adding-up condition is 306
not a-priori imposed by the share format of demand systems as usually happens. Estimation of the 307
complete system (without the deletion of an equation) was performed (Drichoutis et al., 2008). 308
309
4. Empirical results 310
The complete list of explanatory variables included in the demand system is reported in table 5. The 311
average prices of wine categories have a limited variability (SD), expecting a considerable degree 312
of substitutability or complementarity among wines of different categories, except for the Other 313
Wines category insofar as it contains wines at over 3 euros/litre. The presence of the latter category 314
in the model, though lying outside the specific objectives in this paper, ensures a higher degree of 315
completeness in representing the demand system. In the model the purchasers are characterised by 316
means of some socioeconomic variables (δi,k:1to9) directly obtained from information contained in 317
the database used (income classes, age classes, place of residence) and behavioural variables 318
constructed ad hoc (tendency to purchase in discount stores and to purchase red wines, wines from 319
the region of residency and certification of origin wines). These are important elements that directly 320
15
or indirectly reflect different consumption habits. Their introduction in the model thus allows the 321
heterogeneity of wine purchasers to be represented in the analysis. 322
323
Table 5. Description of variables used in the food demand system 324
Parameters Variables Mean SD
γi,j:ML Price of Market Leader (ML) 1.42 0.14
γi,j:PL Price of Private Label (PL) 0.98 0.16
γi,j:OC Price of Other Wines Carton (OC) 1.03 0.27
γi,j:BG Price of Other Wines basic glass (BG) 1.59 0.40
γi,j:TB Price of Other Wines top basic glass (TB) 2.47 0.36
γi,j:PW Price of Premium Wines over €3 (PW) 5.22 2.00
βi Total wine expenditure of the household (€/year) 67.36 112.36
δi,k:1 Share of household purchases in “discount” stores 0.10 0.25
δi,k:2 Share of household purchases for local wine 0.25 0.32
δi,k:3 Income class (1 low – 4 high) 2.51 0.98
δi,k:4 Age class of the household head (1 young – 5 old) 3.09 1.21
δi,k:5 1 if the household lives in North-East Italy 0.19 0.39
δi,k:6 1 if the household lives in Southern Italy 0.28 0.45
δi,k:7 1 if the household lives in North-West Italy 0.30 0.46
δi,k:8 Share of household purchases for “red wine” 0.50 0.37
δi,k:9 Share of household purchases for certificated wine 0.36 0.36
325
The demand model estimation results are shown in Table 6. Starting from the bottom of the table, 326
coefficients τi and λi are respectively the probability density function estimated parameter and 327
quadratic component of equation (10) estimated parameter. 328
329
16
Table 6. Demand system parameter estimation and t-ratios 330
Market
Leader
Private
Label
Other Wines
Carton
Other wines
Basic Glass
Other wines
Top B. Glass
Premium
wines over
3€
i,j: ML t-stat PL t-stat OC t-stat BG t-stat TB t-stat PW t-stat
αi 0.872 9.80 0.058 0.62 -1.784 -8.44 0.177 1.58 -0.388 -2.04 0.689 28.73
γi,j:ML -1.352 -4.99
γi,j:PL 0.231 2.87 -0.069 -3.75
γi,j:OC 0.338 12.36 -0.154 -5.11 0.038 1.72
γi,j:BG 0.369 2.81 0.127 8.00 -0.007 -0.41 -1.023 -1.13
γi,j:TB 0.588 4.59 0.017 0.81 -0.247 -9.71 0.046 2.26 -0.361 5.31
γi,j:PW -0.175 -25.89 -0.152 -1.72 0.032 -10.29 0.488 -3.85 -0.043 -4.01 -0.149 25.72
βi -0.028 -1.65 0.099 7.98 -0.222 -7.54 0.076 5.33 0.006 0.56 0.022 3.14
δi,k:1 -0.102 -1.81 0.101 2.37 -0.093 -1.93 -0.202 -2.47 0.176 4.26 -0.127 -5.98
δi,k:2 0.142 4.48 0.028 1.11 -1.176 -14.43 -0.088 -1.76 0.105 3.37 -0.063 -5.24
δi,k:3 0.030 3.18 -0.028 -3.35 -0.078 -5.69 -0.012 -1.37 -0.007 -1.00 0.003 0.79
δi,k:4 -0.044 -6.27 0.010 1.34 0.062 5.27 0.008 0.94 0.025 3.19 -0.010 -3.45
δi,k:5 0.151 5.10 -0.030 -0.95 -0.048 -1.29 0.115 4.43 -0.109 -3.32 0.061 6.53
δi,k:6 0.125 4.53 0.056 3.06 0.065 2.07 0.068 1.93 0.008 0.37 0.009 0.86
δi,k:7 -0.002 -0.11 -0.003 -0.17 -0.049 -1.51 0.075 3.34 -0.002 -0.14 0.008 0.95
δi,k:8 -0.101 -3.55 0.104 2.65 -0.170 -4.75 -0.219 -7.07 0.200 3.36 -0.085 -7.41
δi,k:9 0.223 2.28 -0.278 -8.14 -1.858 -12.22 -0.862 -8.32 0.120 2.70 0.139 8.25
λi 0.002 0.61 -0.013 -5.42 0.044 8.11 -0.017 -6.26 -0.002 -1.14 -0.003 -2.09
τi -0.081 -1.11 0.068 1.45 0.000 0.01 -0.216 -4.00 0.02 0.71 -0.019 -0.46
331
17
As for the latter, the statistical significance of the coefficient in five equations over six implemented 332
confirms the appropriateness of a quadratic form of the demand system. Coefficients δi,k represent 333
socio-demographic characteristics. They show the major importance of socio-demographics proved 334
by the level of significance reached for the most part. From the sign of the estimated parameters we 335
may also make an initial inference on the average consumer profile for each category of wine 336
analysed. As regards the Market Leader wine (coefficients δML,k:1to9 reported in the first column to 337
the left in table 6), it may be inferred that the typical consumer in this wine category is someone 338
who has a low purchasing share at discount stores (δML,k:1 -0.102), who prefers regional wines (δML,k:2 339
0.142), who tends to be in a higher income class (δML,k:3 0.030), is younger (δML,k:4 -0.044), resides in 340
north-east or southern Italy (δML,k:5 0.151 and δML,k:6 0.125), who prefers white wine (δML,k:8 -0.101), 341
and who prefers certified wines (δML,k:9 0.223). It is also worth pointing out the estimated 342
coefficients of the variables “income class” (δi,k:3) and “age class of the household head” (δi,k:4) for 343
the categories “Market Leader (ML)” and “Premium Wines over 3€ (PW)”. As regards the former 344
(δi,k:3) the sign is positive (δML,k:3=0.030 and δPW,k:3=0.003x) while it is negative for all the other wine 345
categories. For the latter (δi,k:4) the sign is negative (respectively δML,k:4=-0.044 and δPW,k:4=-0.010) 346
while it is positive for all the other categories. This allows us to state that the market leader 347
(Caviro), which invests significantly in marketing, and wines with a higher price than 3 euro/litre 348
(PW) attract the same consumer type, who can be depicted as younger with a higher disposable 349
income, geared to the purchase also of wines in a higher price bracket. 350
The effect of prices will be discussed after calculation of elasticities. From estimating the 351
parameters of prices (γi,j) and expenditure (βi) of equation (10) we obtained Marshallian elasticity 352
values for prices (direct ηi,i and cross ηi,j) using the procedure suggested by Yen et al. (2002): 353
(13) 354
355
ηi,i =∂wi
∂Pi
Φi zθ̂i( ) / wi −1
18
(14) 356
357
The elasticities thus calculated are reported in table 7, where along the row we may read the effect 358
on the quantity consumed of the i-th wine category of the change in price of the j-th wine 359
represented in the column. The directly estimated elasticities reported along the main diagonal of 360
the table show that demand for each of the six product categories considered in the analysis reacts 361
to price variations in a different way, albeit within a limited range for all categories varying from -362
1.18 to -1.50. The only exception is the “Other Wines Carton (OC)”, a category which, with a value 363
of -3.04, describes a particularly elastic behaviour of demand. At the same time, one product 364
category –Market Leader– has a fairly low elasticity. 365
These differences show the effectiveness of being a market leader, which ensures a lower degree of 366
vulnerability to price variations. The marketing implemented by Caviro is also effective with regard 367
to basic wines sold in glass bottles (both BG and TB), although the latter benefit from a higher 368
degree of market penetration (Tab. 4). The cross elasticities, in turn, show a very complex picture of 369
substitution trends with rare cases of reciprocity. Brand cartons (Market Leader and Private Label) 370
and non-brand cartons (Other Wine Cartons, OC) show a fairly differentiated demand behaviour in 371
terms of substitution. Brand cartons are substituted by wine in glass: the Market Leader is 372
substituted by more costly wines (PW), even if the opposite does not hold at the same magnitude; 373
Private Label wines tend to be substituted mainly by top basic glass bottles (TB), also in this case 374
without reciprocity. Other Wine Cartons (OC), whose demand is very price-sensitive, have various 375
(fairly high) coefficients of cross elasticity, although the most frequent substitution is with Private 376
Label wines. Basic Wines in glass bottles (BG) tend to be substituted mainly by Private Label wines 377
(PL) - in this case one of the few cases of reciprocity occurs – or by more costly wines in glass 378
bottles (PW). 379
380
ηi, j =∂wi
∂Pj
Φi zθ̂i( ) / wi
19
Table 7. Price elasticities 381
ML
PL
OC
BG
TB
PW
Market Leader (ML) -1.18** -0.05** -0.24*** 0.10* -0.29* 0.60**
Private Label (PL) -0.27** -1.44*** 0.18*** 0.31*** 0.94 -0.41
Other W. Carton (OC) 0.54*** 1.92*** -3.04* 0.79 1.78 -2.13***
Other W. Basic Glass (BG) 0.13* 0.30** -0.06 -1.38* -0.20*** 0.34***
Other W. Top Basic Glass (TB) 0.33* 0.04 0.39 0.18 -1.50** -0.57*
Premium Wines over 3€ (PW) 0.13*** 0.29* 0.02*** 0.05*** -0.06*** -1.37***
The elasticities in the jth column indicate the change in the demand for all i goods as the jth good's price changes. 382
Bootstrap estimate of the significance level: * p<0.10; **p<0.05; ***p<0.01 383
384
Top basic glass (TB) wines tend to be substituted by Market Leader (ML) or by non-brand cartons 385
(OC), there being another example of reciprocity. Lastly, Premium wines (PW) show a moderate 386
tendency to be substituted by all the other cheaper wines, with the exception of TB. Moreover, such 387
wines tend to substitute only two product categories and are overall a rather isolated category. 388
389
5. Concluding remarks 390
The aim of this paper was to investigate the domestic consumption of wine, with the focus on 391
non-premium wines. In order to do so, a censored demand system (QAIDS) was estimated using 392
AC-Nielsen homescan data, statistically representative of Italian households in year 2010. 393
Moreover, categorization of non-premium wines was necessary in order to synthesize a wide range 394
of wines in fewer categories. It emerged that the market for non-premium wines, apparently 395
showing thousands of products, can be aggregated into five categories. The results show a complex 396
picture in which there emerge important specific details in the demand behaviour towards the 397
various categories considered. Hence, some important considerations may also be inferred on the 398
strategies to be implemented with a view to consolidating demand-supply relations and on the 399
20
possible evolution of the market for basic wines. Another result of our analysis worth noting 400
concerns the categorization of basic wines proposed. As argued in section two, there is no single 401
categorization of non-premium wines available. 402
Our empirical strategy was to interview some key informants of the Italian wine sector who 403
described how the supply of wines takes place in the Italy’s big retail chains. Consumption data 404
provided by AC-Nielsen was managed to fit that picture on the consumption side. The statistically 405
significant level of almost all coefficients estimated by means of the demand system confirms that 406
the categorization implemented seems to capture the key elements of that particular market side. 407
We may thus confirm what was stated in the first part of the paper: the market for cheaper wines is 408
complex and products, categorized as proposed, show a significant degree of heterogeneity. 409
Besides confirming our underlying hypothesis, the results obtained show a complex picture of the 410
relations between the categories used in the analysis. Particular elements may be found which allow 411
hypotheses to be postulated on the functioning of the market in the period in question and on its 412
possible evolution. The most salient elements are low elasticities for carton wine of the market 413
leader (ML) and high elasticity of non-brand wine in cartons (OC). Albeit with the necessary 414
caution, on the basis of the results it may be stated that, also in the segment of more economical 415
wines, brand is an effective instrument of diversification. The brand effect, though to a lesser 416
extent, is also recognised in private label carton wines which, proposed with a similar price to that 417
of carton wines without a recognisable brand, have a much lower elasticity than the latter. The high 418
elasticity of products in cartons without a recognisable brand not only sustains the hypothesis of 419
brand importance but also envisages the evolutionary dynamics of the market for non-premium 420
wines which could be most accentuated, with a reduction in unbranded products and expansion of 421
the weight and number of branded products. Examination of the cross elasticities allows further 422
reflection on carton wine of the market leader and that with private label. The data show that as the 423
price of the market leader grows, purchase of this product declines in favour of the premium wine; 424
21
as the price of private label wine, purchases are shifted towards other basic products, except for the 425
market leader carton. The two branded wines in the carton are not substitutable. 426
Hypothesising on the evolution of the market size for more economical wines, it may be noted that 427
only for two categories, the carton wine of the market leader and basic wine in glass, is a tendency 428
shown to shift purchases towards premium wines. Indeed, most of the substitutions of purchases 429
seem to remain within the circle of non-premium wines. Taken together, the results of our 430
investigation clearly indicate that the market for non-premium wines is complex. Although non-431
premium wines may be considered convenience products, their supply shows well-defined 432
structuring. This encourages the launching of further studies, including those on the demand 433
structure, in line with that presented in this paper. New studies will require that the categorisation of 434
non-premium wines implemented herein be verified. In this sense, the experience of professionals 435
operating in the supply chain of non-premium wines is undoubtedly essential. This approach could, 436
however, be supplemented by surveys of consumers in order to analyse directly how the great 437
variety of the supply of non-premium wines is viewed and perceived. 438
439
References 440
Akbay, C., Boz, I. and Chern, W.S. (2007). Household food consumption in Turkey, European 441
Review of Agricultural Economics, 34, 209-231. 442
Amadieu, P. and Viviani, J.L. (2011). Intangible expenses: a solution to increase the French wine 443
industry performance?, European Review of Agricultural Economics, 38, 237-258. 444
Anderson, K. and Nelgen, S. (2011). Global wine markets, 1961 to 2009: a statistical compendium. 445
The University of Adelaide press, Adelaide. 446
Banks, J., Blundell, R. and Lewbel, A. (1997). Quadratic Engel curves and consumer demand, 447
Review of Economics and Statistics, 79, 527-539. 448
Barnett, W.A. and Serletis, A. (2008). Consumer preferences and demand systems, Journal of 449
Econometrics, 147, 210-224. 450
22
Bernetti, I., Casini, L., and Marinelli, N. (2006). Wine and globalisation: changes in the 451
international market structure and the position of Italy, British Food Journal, 108, 306–315. 452
Blundell, R. and Robin, J.M. (1990). Estimation in Large and Disaggregated Demand Systems: An 453
Estimator for Conditionally Linear Systems, Journal of Applied Econometrics, 14, 209-232. 454
Corsi, A., Pomarici, E. and Sardone, R. (2004). Italy, in K. Anderson (ed.), The World’s Wine 455
Markets: Globalization at work, Cheltenham: Edward Elgar, London. 456
Costanigro, M., McCluskey, J. J. and Mittelhammer, R.C. (2007). Segmenting the Wine Market 457
Based on Price: Hedonic Regression When Different Prices Mean Different Products, Journal 458
of Agricultural Economics, 58, 454-466. 459
Davidson, R. and MacKinnon, J.G. (2004). Econometric Theory and Methods, Oxford University 460
Press, New York. 461
Davis, T.R., Ahmadi-Esfahani, F.Z. and Iranzo, S. (2008), Demand under product differentiation: 462
an empirical analysis of the US wine market. Australian Journal of Agricultural and Resource 463
Economics, 52, 401–417. 464
Deaton, A. and Muellbauer, J. (1980). An almost ideal demand system, The American Economic 465
Review, 70, 312-326. 466
Deaton, A. and Zaidi, S. (2002). Guidelines for Constructing Consumption Aggregates for Welfare 467
Analysis, LSMS Working Paper No. 135, World Bank, Washington. 468
Drichoutis, A.C., Klonaris, S., Lazaridis, P. and Nayga Jr, R.M. (2008). Household food 469
consumption in Turkey: a comment, European Review of Agricultural Economics, 35, 93-98. 470
Einav, L., Ephraim L., and Aviv, N. (2008). On the Accuracy of Nielsen Homescan Data, ERR-69, 471
U.S. Dept. of Agriculture, Econ. Res. Serv. December, USA. 472
Ernst and Young Entrepreneurs Etude des Filières et des Stratégies de Développement des Pays 473
Producteurs de Vins dans le Monde: Analyse de la Filière Viticole Australienne, ONIVINS, 474
Office National Interprofessionnel des Vins, Paris. 475
23
Fisher, D., Fleissig, A.R. and Serletis, A. (2001). An Empirical Comparison of Flexible Demand 476
System Functional Forms, Journal of Applied Econometrics, 16, 59-80. 477
Fogarty, J.J. (2010). The demand for beer, wine and spirits: a survey of the literature, Journal of 478
Economic Surveys, 24, 428-478. 479
Fraser, I. (2005). Microeconometric analysis of wine grape supply contracts in Australia, Australian 480
Journal of Agricultural and Resource Economics, 49, 23–46. 481
Gallet, C.A. (2007). The demand for alcohol: a meta-analysis of elasticities, Australian Journal of 482
Agricultural and Resource Economics, 51, 121-136. 483
Gil, A. and Molina, J. (2009). Alcohol demand among young people in Spain: an addictive 484
QUAIDS, Empirical Economics, 36, 515-530. 485
Golan, A., Perloff, J.M. and Shen, E.Z. (2001). Estimating a demand system with non-negativity 486
constraints: Mexican meat demand, Review of Economics and Statistics, 83, 541-550. 487
Hannin, H., Couderc, J-P., d’Hauteville, F. and Montaigne, É. (2010). La vigne et le vin, La 488
documentation Française, Paris. 489
Heien, D. and Wessells, C.R. (1990). Demand Systems Estimation with Microdata: A Censored 490
Regression Approach, Journal of Business & Economic Statistics, 8, 365-371. 491
ISTAT, Italian National Institute of Statistics (2011), Consumption pattern of Italian population, 492
available at http://www.istat.it/en/ 493
Labys, W.C. and Cohen, B.C. (2006). Trends versus cycles in global wine export shares, Australian 494
Journal of Agricultural and Resource Economics, 50, 527–537. 495
Mäkelä, P., Gmel, G., Grittner, U., Kuendig, H., Kuntsche, S., Bloomfield, K. and Room, R. (2006). 496
Drinking Patterns and Their Gender Differences in Europe, Alcohol and Alcoholism, 41, 8-18. 497
Mediobanca (2011). Indagine sul settore vinicolo, MBRES – Ufficio studi Mediobanca, Milano. 498
Moro, D. and Sckokai, P. (2000). Heterogeneous preferences in household food consumption in 499
Italy, European Review of Agricultural Economics, 27, 305-323. 500
24
OIV: Organisation Internationale de la Vigne et du Vin (2012). Report of the Director General of 501
the OIV on the world vitiviniculture situation in 2011, International Organisation of Vine and 502
Wine, Paris, available at http://www.oiv.int/oiv/info/enconjoncture?lang=en 503
Panzone, L.A. (2012). Alcohol Tax, Price–Quality Proxy and Discounting: A Reason Why Alcohol 504
Taxes May Rebound, Journal of Agricultural Economics, 63, 715-736. 505
Perali, F. and Chavas, J.P. (2000). Estimation of Censored Demand Equations from Large Cross-506
Section Data, American Journal of Agricultural Economics, 82, 1022-1037. 507
Pomarici E., Boccia F., and Catapano D. (2012). The Wine Distribution Systems Over the World: 508
An Explorative Survey, New Medit, 4, 23-38. 509
Rabobank Wine is Business (Rabobank, Utrecht, 2003). 510
Reg. (EC) 607/2009, 14 July (2009) L193/60, available at http://eur-511
lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2009:193:0060:0139:EN:PDF. 512
Ritchie, C. (2009). The culture of wine buying in the UK off trade, International Journal of Wine 513
Business Research, 21 194-211. 514
Shonkwiler, J.S. and Yen, S.T. (1999). Two-step estimation of a censored system of equations, 515
American Journal of Agricultural Economics, 81, 972-982. 516
Steenkamp, J-B. and Ter Hofstede, F. (2002). International market segmentation: issues and 517
perspectives, International Journal of Research in Marketing, 19, 185-213. 518
Tauchmann, H. (2005). Efficiency of two-step estimators for censored systems of equations: 519
Shonkwiler and Yen reconsidered, Applied Economics, 37, 367-374. 520
Torrisi, F., Stefani, G. and Seghieri, C. (2006). Use of Scanner Data to Analyze the Table Wine 521
Demand in the Italian Major Retailing Trade, Agribusiness, 3, 391-404. 522
Yen, S.T., Kan, K. and Su, S. (2002). Household demand for fats and oils: two-step estimation of a 523
censored demand system, Applied Economics, 34, 1799-1806. 524
Yen, S.T. and Lin, B.H. (2006). A sample selection approach to censored demand systems, 525
American Journal of Agricultural Economics, 88, 742-749. 526
25
i Rabobank (2003), for instance, considers basic wines those sold at a price lower than 3 euro per standard 0.75 litre
bottle. Rabobank also proposes to classify premium wines into four sub-categories according to the price of a standard 0.75 litre bottle: popular premium (price between 3 and 5 euro); premium (price between 5 and 7 euro); super premium (price between 7 and 14 euro); ultra premium (price between 14 and 150 euro); icon (higher than 150 euro).
ii Such estimation considers wines subdivided into three categories using prices measured as wholesale pre-tax prices per litre at the national border: non-premium, lower than US$2.5; commercial premium, US$2.5 – US$7.5; superpremium, higher than US$7.5
iii From now on in this paper the authors have taken the liberty to use non-premium, less expensive and basic wines as synonymous.
iv The consumption and production pattern followed by Italy could be observed in all those wine producing and consuming countries showing the same structural characteristics.
v Data about import and export come from the Global Trade Information Services Database vi Nielsen ConsumerScan data have been questioned in terms of accuracy. For a complete reference see Einav et al.
(2008). vii The cut-off price between non-premium (basic) and premium wines confirms the Rabobank classification which is
substantially equivalent to that proposed by Anderson and Nelgen (2011). viii It is worth noting that the data reported refer to the distribution within WAH consumption of wines in Italy, which is
estimated to be 70% in volume and 30% in value terms by Anderson and Nelgen (2011) ix No categorization was implemented based on “red – white” since, as experts pointed out, criteria that make consumers
choose between red or white wines are the same within each of these categories. Put differently, consumers make a choice between red and white based on variables, or attributes, that is beyond the choice set of attributes available in a store shelf. As for legal category, this criterion is relevant to consumer choices almost exclusively for premium wines.
x Although the level of significance of this coefficient is estimated at 5%, it is worth holding it in due consideration for the purpose of describing purchasing behaviour.