The Importance of Regional and Local Origin in the Choice of Wine: Hedonic Models of Portuguese Wines in Portugal

Abstract

Having occupied a leading position as an export product, wine production and vine growing in Portugal are still highly relevant and deserve a dominant place in the Portuguese economic and social landscape. The objective of this work is to contribute to an analysis of the current state of development of Portuguese viticulture and viniculture, analysing the importance of regional origin in the composition of prices in the Portuguese market. Using a hedonic model and online data from a large retailer in Portugal, we are able to identify the regional premiums for Portuguese wines in the internal market. Furthermore, we assess AOC recognition and region-specific premiums, and we obtain evidence of differences in the price contribution of denomination of origin from different regions.

Introduction

Wine is historically one of the most important export-oriented products in the Portuguese agricultural economy. In Ricardo's example of comparative advantage (Ricardo, 1817) the model of international trade was built around the exchange of wine from Portugal and cloth from England. In national food culture wine has remained an important product for everyday local consumption. At the same time, wine production and vine growing continue to deserve a relevant place in the Portuguese rural economy, remaining one of the most important elements of the national agricultural landscape, with an economic, environmental, technological, and social impact (Simões, 2006).

Historically, viticulture was a national economic activity, and vineyards were grown in all the regions of the country. However, within each region, certain areas of production developed a stronger link with winemaking and these became an important contributor to the local economy (Simões, 2006). In order to protect this important asset of Portuguese agriculture, policies have tended to focus mostly on building collective reputations over every wine-producing area. This pattern was stimulated after the Portuguese accession to the former European Economic Community (EEC, later European Union), in 1986 (Simões, 2006). In fact, policies implemented in the sector focused on the improvement of the quality of wines in terms of production standards, and every region started regulating geographical origin and identifying areas or localities where the link between origin and wine production was particularly strong. The result was the increase in the number of products with Controlled Appellation of Origin (from the French ‘Appellation d'Origine Contrôlée’, or AOC henceforth) labels, increasing from only 11 regions with AOC in 1986, to 29 today, and covering practically the totality of the wine-growing areas in the country with a relevant wine density (Simões, 2006). Table 1 lists the different wine regions of Portugal and the different regional and AOC denominations. The focus on AOC labelling implied the creation of new specific institutions for the wine sector and the requalification of existing institutions, as also required by the adoption of the Common Market Organization (CMO) of the wine sector.

Table 1. List of Portuguese wine regions and their denominations of origin

Region	Regional denomination (geographical indication)	Region-specific AOCs
Alentejo	Alentejano	Alentejo AOC
Algarve	Algarve	Lagoa, Lagos, Portimão, Tavira
Beiras	Beiras	Bairrada, Dão, Lafões, Tavora-Varosa, Beira Interior
Estremadura	Estremadura	Alenquer, Arruda, Bucelas, Carcavelos, Colares, Encostas d'Aire, Lourinhã, Óbidos, Torres vedras
Ribatejo	Ribatejano	Ribatejo AOC
Minho	Minho	Verdes
Douro	Duriense	Porto,1 Douro
Terras do Sado	Terras do Sado	Palmela, Setúbal
Tras-os-Montes	Transmontano	Tras-os-montes
Madeira	Terras Madeirenses	Madeira,2 Madeirense
Açores	Açores	Biscoitos, Graciosa, Pico
Source: Portuguese Institute of Wine and Vine (IVV).

From a market point of view, wine consumption changed substantially, with quality wines reaching almost 50% of the total consumption of wine (Table 2), and the remainder shared among regional geographical indications (GI) and table wines. Specifically, wines from Alentejo are those with the highest market share, and the highest for regional wines, while the AOC with the highest consumption is the Vinho Verde from the Minho region (Table 3). It is interesting to notice that apart from the four major AOCs (Vinho Verde, Alentejo, Douro, and Dão), the remaining AOC wines have a market share in the internal market of less than 1%, suggesting a rather local consumption of those products.

Table 2. Shares in sales of wine per segment in the Portuguese market (excluding Açores and Madeira)

	Total consumption (2007)
AOC wines	45.6%
Regional wines	37.0%
Table wines	17.4%
Source: AC Nielsen (2006).

Table 3. Total consumption of wine by origin (excluding Açores and Madeira)

Segment	Origin	% total (2006)	% segment (2006)
AOC	Vinho Verde	17.3%	38.0%
	Alentejo	14.9%	32.6%
	Douro	6.6%	14.4%
	Dão	4.1%	9.0%
	Bairrada	0.9%	1.9%
	Palmela	0.6%	1.4%
	Lagoa	0.2%	0.5%
	Torres Vedras	0.2%	0.3%
	Bucelas	0.1%	0.3%
	Other AOC	0.7%	1.5%
Regional denomination	Alentejano	21.3%	57.5%
	Terras do Sado	6.5%	17.5%
	Estremadura	3.8%	10.4%
	Beiras	1.9%	5.0%
	Tras-os-Montes and Terras Durienses	1.7%	4.7%
	Ribatejano	1.6%	4.3%
	Minho	0.2%	0.5%
	Algarve	0.0%	0.1%
Table wine	National table wine	15.9%	91.3%
	Imported table wine	1.5%	8.7%
Source: our elaboration on AC Nielsen data.

The application of a regulation focusing on AOC labels also meant enlarging destination markets for producers, which expanded from local to national, and finally international. Following the increase in availability of geographically identified wines throughout Portugal, the introduction of a traceability mark aimed to protect consumers from fraud by providing clearly recognizable and indisputable labelling. This created a segmentation of the market in terms of production quality. Throughout the process, the Portuguese consumer maintained a clear preference for national wines, since import levels remained very low and generally associated with specific categories, that is, sparkling wines (Simões, 2001).

In our work, we aim to test for the importance of collective reputation (Tirole, 1996), that is the importance that every wine-producing region in Portugal has acquired in its internal market. From this perspective, we want to classify national production by premium per region. We allow for regional premiums being due to two different factors. First, certain regions may have inherited a strong reputation for wine making. At the same time, the quality improvements requested by AOC policies will generate a further premium, which we conjecture could be region-specific. In order to estimate region-specific premiums and discounts, our analysis applies the theory of hedonic pricing (Rosen, 1974), a technique widely used to understand implicit values in consumer markets, and also in wine economics. In the next section, we analyse previous studies applying hedonic modelling to the wine sector. A description of the data and discussion of the results of the econometric analysis will then follow, with our conclusions in the final section.

Theoretical Background of Hedonic Modelling

According to consumer theory (Nelson, 1970), consumers value products as a bundle of characteristics, each of which gives a utility increase depending on the level contained in the product sold in the market. As a result, any product i can be seen as a vector Zof n measurable characteristics, or

Rosen (1974) developed the approach, showing that the price of the product in a market is a function of its level of characteristic P(Z _i ) defined as

Coefficients, which are partial derivatives of the price function, ∂ P _i /∂ z _i , inform the marginal contribution of each characteristic to the final price in the market.

Empirical application of the hedonic theory in the wine sector started in the early 1990s, pioneered by Golan and Shalit (1993), on grape pricing, and by Oczkowski (1994), on Australian table wines. After the first papers were published, the technique was critically analysed by Unwin (1999), who highlighted weaknesses in the specification of variables and the reality of imperfect competition as a characteristic of the wine market. In our work, we consider Unwin's critical points identifying carefully our variables according to a previous consumer study and, in a second step, we present a segmentation of the market, as suggested by Thrane (2004), although this is not the primary focus of our work.

Hedonic modelling has often been used to capture the perceived value of reputation in the wine market. This includes reputation of the wine producer (Ling and Lockshin, 2003; Oczkowski, 1994; Landon and Smith, 1998; Schamel and Anderson, 2003; Haeger and Storchmann, 2006), which empirically appears to constitute an important component in the composition of price. Along with a measure of individual reputation, empirical analyses using this econometric technique have tested for the influence of the collective reputation of producers. In fact, certain wine producers, localised in specific geographic clusters, have developed through time and history an important reputation for a standard of production and the provision of quality not just as individual winemakers, but for all producers in that area. European countries have provided their products with a label signalling origin, recognising the uniqueness of wines coming from historically traditional areas. The so-called Denomination of Origin (DO) therefore indicates a wine whose name corresponds to the place where the wine is made, as that information is perceived as essential for consumers. Legally required information has allowed the price premium given to wines in the Old World producing countries to be tested (Steiner, 2004; Benfratello et al., 2008; Angulo et al., 2000; Landon and Smith, 1998). Nonetheless, certain New World regions have acquired more recently a collective reputation valued by consumers (Schamel, 2006; Schamel and Anderson, 2003; Oczkowski, 1994), although generally with lower premia compared to historical producers (Schamel, 2006).

The general hedonic model used in empirical works has the form:

where X _i can include objective characteristics both of the wine and of producers’ individual and collective reputations, and the dependent variable appears in a non-linear form (Ekeland, 2002), which we choose as semi-logarithmic (see e.g. Oczkowski, 2001). In order to capture differences in the retailer's behaviour and strategy for products coming from different segments, we start by estimating individual regressions for the three segments in the Portuguese market. Previous work has proposed segmentation by colour (Thrane, 2004; Noev, 2006; Ling and Lockshin, 2003), price (Angulo et al., 2000; Costanigro, 2007), climate regions (Ling and Lockshin, 2003) and varieties (Ling and Lockshin, 2003). We opt for segmenting by quality in accordance with the Portuguese legislation on wine (http://www.ivv.min-agricultura.pt/vinhos/index.html), which identifies:

1.	table wines, which are generic products;
2.	regional wines, which are wines with a generic geographical indication (GI) corresponding to the region they are produced in;
3.	AOC wines, which are wines with a very specific geographical origin, in general receiving the name of the locality from which they originate.

This implies running equation (1) for every segment j of Portuguese wine as

As a second step, we test for the presence of a premium in an AOC label. In order to capture the average premium, we run an additional model pooling the data of the three subsamples of wines, including a set of sample-specific dummies. This corresponds to a simpler regression in the form:

where AOC _i is 1 if the wine has a Denomination of Origin and 0 otherwise, and REG _i indicates if the wine has a general regional labelling instead.

As an alternative, we allow for a double effect of the AOC label, a shift of the constant term and a change in the slope of the regional premium/discount. This shows the different influence of an AOC label for specific regions. We identify with REG _i the region of origin of the wine (as in equation 3), as identified from the Portuguese institute of Vine and Wine (IVV). We therefore estimate the equation:

which corresponds to

and, if we call β₀=(α₀ + α₂ AOC _i ) and β₁=(α₁ + α₃ AOC _i ), equation (4) becomes

The assumption leading to model (3) is that α₃ = 0. In case of no AOC effect on wines, α₂ and α₃ both equal zero, therefore β₀ = α₀ and β₁ = α₁, and the premium associated with the product derives only from the inherited reputation of the region.

Data and Variables

The data used for the following exercise are a set of prices of products sold in the main Portuguese chain of retailers: Continente Hypermarket. This retailer accounts for around 40% of the total food sales in the mass market distribution channel in Portugal (APED, 2006), and is spread nationwide. Overall, hypermarkets are responsible for 35% of wine sales (ESB/UCP, 2003). Prices were obtained using the online version of Continente (http://www.continente.pt/HomePage.aspx) in July 2007, generating a dataset that is nationwide, that is, independent of regional sales influences. These prices are readily available online without a need to log into the system and, therefore, can be considered as recommended retail prices for those individuals who engage in a purchase, and comparison prices for those who browse the website. The data were taken directly from the website, reporting the description of the item (i.e. the name, and the observable characteristics, such as vintage year, colour, etc.) for every wine offered by the retailer. The characteristics of the products all appeared in a descriptive form and were converted into sets of dummy variables, assigning a value of 1 if a specific characteristic was present and 0 otherwise. All the variables will be described more precisely in the next paragraphs. Prices correspond to a 75 cl bottle (the volume of a standard wine bottle) and observations with different sizes were converted assuming linearity between prices and sizes. We then corrected the influence of scale on price, including dummy variables, for different original container sizes. Our sample is composed of 49.67% AOC wines, 25.82% Regional wines, and 24.51% table wines, which is similar to that found in Table 1 and to data from AC Nielsen (2006), although they suggest a lower representation of regional denominations to the advantage of table and AOC wines. Comparing origin and colour (Table 4), we notice that in the retailer's selection of products, white and rosé wines are slightly more represented than red when compared with actual market sales. Analogously, wines from Minho and Beiras are over-represented, while Alentejo and Douro are under-represented.

Table 4. Comparison between the sample and market share per selected categories

	Our sample	Market share
Red	63.2%	59.9%
Rose	3.9%	2.5%
White	32.9%	37.6%
Minho	8.9%	17.5%
Douro	15.6%	8.3%
Beiras	13.7%	7.1%
Estremadura	4.8%	4.2%
Ribatejo	3.8%	1.9%
Alentejo	21.9%	36.1%
Algarve	0.8%	0.3%
Terras do Sado	5.6%	7.1%
Others	NN	17.5%
Source: Own elaboration from the dataset (our sample) and AC Nielsen (market share).

Before we proceed, we need to stress the fact that our prices are not actual retail prices in actual hypermarkets, but online prices.3 Information on the market share of online wine sales in the Portuguese market relative to the retail channel seems to be not easily available, although we would imagine that it corresponds to a rather negligible part of the market. This is something to be taken into account when considering the results, since they may be affected by a sample bias, and therefore make generalisations difficult. An important step in assessing the problem would be to understand if wine prices actually differ when sold online rather than in a conventional shop. Current research on how actual online prices differ from actual retail prices gives no definite indication. General findings have shown that internet prices tend to be lower than in the physical shops of the same retailer (Brynjolfsson and Smith, 2000; Stylianou et al., 2005), but not every empirical analysis has successfully found significant differences (see e.g. Gan et al., 2007, who analysed grocery retailing). For heterogeneous products (i.e. books, Clay et al., 2001) the difference in price is not constant, therefore it is not possible to assume that the price difference, if present, is the same for all the products in our dataset. However, the price difference between online and conventional retail should converge as more consumers use them both (Xing, 2007; Gan et al., 2007). Convergence is easier when cross-comparison of product prices is easier (Gan et al., 2007) and, in the case of wine, only unique wines tend to be less likely to converge (Lynch and Ariely, 2000). In our case, the difference in prices is not assessable, and the same applies for the magnitude of the hypothetical difference. On the other hand, we do not have evidence to suppose the unreliability of data, and the impossibility of making a comparison in the market prevents us from giving a definite answer to this interesting research question. However, the retailer we survey is not the only one allowing online purchases, and the expectation of a low number of online wine transactions could reinforce the idea that those prices are used by consumers for comparison and are, therefore, kept as competitive and realistic as possible by the retailer.

Overall, the sample comprised 608 table and quality wines.4 The dataset included the main information found on the labels of those products. As mentioned earlier, being qualitative information, dummy variables were used, assigning the value of 1 when a particular characteristic was present on a product, 0 otherwise. The variables included relate to nine main pieces of information consistent with the buying factors of Portuguese consumers’ found in a previous study on Portuguese wines (ESB/UCP, 2003), which we now describe.

The first set of parameters refers to the technical characteristics of products: colour of the wine (red, white or rosé), packaging (bottle or box, plus the volume of the container), and indication of a specific quality (Reserve, Selection 5). Regional wine production has also been taken into account, focusing on two aspects. First, we included the region of production as it appears in the website. The variables introduced are Minho, Douro, Beiras and Alentejo, Terras do Sado, Algarve, Ribatejo, which contain both AOC and Regional wines. In our dataset, the Beiras region includes both the AOC Dão and Bairrada, but these denominations have been treated jointly for the purpose of comparability of models. The reference variable is No region specified (which corresponds to table wines). Second, for models (3) and (4), we included the market category, as found on the website of the National Institute of Vine and Wines (IVV):6 Quality wines (AOC), Regional wines, and table wines. Another piece of information inserted in the dataset was the vintage year, going from 1991 to 2006, or No vintage year. Due to the low number of observations, we grouped vintages from 1991 to 2000, representing long-aged products.

The importance of the producer was tested. Due to the absence of producer information for every observation, the hypothesis tested was whether or not a producer brand has a price premium compared to a private label (retailer brand). The only brands with a specific producer significance in our dataset are Mateus and Lancers, and we included a specific joint variable to capture the average premium associated with these names. Supermarket strategies were included in the dataset: exclusive, if the retailer sells the product exclusively; promotion, if sold at a promotional price; bargain if the website indicates that the product is cheap.

The final test was for the presence of keywords in the name of the product, which also give an indication of the producer of the specific wine. We tested the premium associated with two main words: Quinta/Herdade 7 (Estate), whose use in the name of a wine is defined by law, and Cooperative. According to ESB/UCP (2003), consumer choice can be influenced by these two keywords, with a preference for a product if it is made directly by the vine grower rather than a cooperative. The terms ‘Herdade’ and ‘Quinta’ both correspond to large farms with a large economic and social ‘prestige’. They differ in origin, since ‘Herdade’ is the term used in the North and Central part of Portugal, while ‘Quinta’ is used in the South of the country (i.e. Alentejo e Algarve).

External sources of information (i.e. ratings reported in specialised magazines, Oczkowski, 2001) were unavailable although, as a referee correctly highlighted, a summary variable on the sensory evaluation of wines would be important. The exclusion was due to missing information in the data collected, and the use of a different source would imply reducing consistently the number of observations. However, we think this is not a major drawback for two reasons. First, Portuguese consumers in ESB/UCP (2003) do not explicitly refer to taste as one of the most important buying factors. We do not expect taste to be irrelevant for wine drinkers, but probably consumers can estimate the sensory characteristics they are looking for on the base of the characteristics reported on the label. Moreover, Portuguese consumers prefer to use their own knowledge, and that of friends/relatives, as their main sources of information (ESB/UCP, 2003), which also supports the idea that the market may have a good degree of taste awareness. Second, sensory characteristics are region-specific, as they are the cause of a region-of-origin bias, and process specific. As a result, the use of variables specifying regions, colour, or Reserva/Selection should partially include the information, and a quality variable would increase the impact of multi-collinearity. This is not a perfect substitution of the taste variable, but it is a good approximation, and probably the best option available. A last important remark is that a previous empirical work by Horowitz and Lockshin (2002) found that quality ratings may depend on price, while price also depends on quality. This simultaneity between price and ratings would lead to endogeneity problems in the estimation, which would not be possible to correct using the present dataset.

Results

According to the assumptions of non-linearity, the dependent variable is the logarithm of price. We assume that the market is in equilibrium, or that stocks or shortages of the previous year do not influence current prices significantly. This is a credible assumption since, due to the presence of the wine CMO, there are various means of reaching this equilibrium (i.e. support for stocks and elimination of surpluses through distillation). Moreover, we assume the retailer offers competitive prices, which somehow try to capture the expected consumers’ evaluation of the products.

The logarithmic model expresses the final value as:

where y is price, β₀ is the constant term, x _i is the characteristic observed, and u is the error term. Accordingly, exp (β) indicates what coefficient corresponds to a variable, and {1-exp(β)} is the relative premium (a positive value will indicate a premium, while a negative value would imply a discount). Hence, if the coefficient is positive, its exponential is higher than 1 and the associated characteristic gives a premium; if it is negative, and its exponential lower than 1, it will give a discount. If the coefficient is not significantly different from 0, the variable will have no significant influence in the composition of price.

As mentioned above, in our analysis we first regress price on qualities on the single segments, then on the total sample. In this last case, we ran two different models: a simple regression (Total simple) and then allowing for a slope dummy for the AOC variable (Total complex). The test statistics of the different models are reported in Table 5. Overall, the model with the highest variability is the one including table wines, followed by the Total complex, as they have the highest R ². The AOC model appears to be the worst in explanatory power, with an adjusted R ² lower than 0.5. This relatively low level of R ² is probably due to two missing purchasing drivers mentioned in ESB/UCP (2003) that could not be included in the model thoroughly: the brand of the product and the producer's name. The absence of a variable including an evaluation of quality can be also the cause of a very low adjusted R ₂ for AOC wines, suggesting that quality could have more relative importance in the higher segment of the market compared to the lowest.

Table 5. Summary of the tests for the four models

	AOC	Regional	Table	Total simple	Total complex
Effective sample size	302	157	149	608	608
R-square	0.4263	0.6052	0.7098	0.6428	0.6541
Adjusted R-square	0.3766	0.5262	0.6794	0.621	0.6284
Overall F test	8.58***	7.66***	23.41***	29.41***	25.44***
Jarque-Bera test	102.0993***	40.0607***	0.4530	314.7638***	286.0965***
Breusch-Pagan test	59.3423***	60.4117***	16.8648	138.4645***	166.2435***
Source: own calculation. ***significant at 1%.

The Jarque-Bera tests show that, with the exclusion of the table wine model, the error terms are not normally distributed. However, the size of our sample should be sufficiently large to make the inference on the betas correct (Wooldridge, 2006: 787 – 794). Since all the models, with the exception of table wines, presented heteroskedasticity problems, as shown by the Breusch-Pagan tests, we calculated the robust standard errors using the White's heteroskedasticity consistent variance-covariance matrix.

The results of the regressions are shown in Tables 6 – 9. The analysis per segments allows an understating of the specific criteria used in the valuation of wines in different marketing categories. Results for the AOC and regional wines are jointly reported in Table 6, while table wines are reported in Table 7. Colour is an important factor in the choice of the wine. However, if red receives a premium compared to white wines in the AOC segments, consistent with ESB/UCP (2003), in the other two segments white and red wines are not valued differently, while rosé wines are valued more than reds and whites among table wines.

Table 6. Results of the hedonic model on the AOC and regional segments

	AOC		Regional
Parameters	Estimate	H.C. p-value	Estimate	H.C. p-value
Intercept	0.5164***	0.0036	0.2935***	0.0040
Red	0.2174**	0.0127	0.1866	0.2600
White	Reference		Reference
Minho	0.4049**	0.0189	−0.6569***	0.0000
Douro	0.4293***	0.0028	−0.6181***	0.0059
Beiras	0.0655	0.6476	−0.3943**	0.0389
Estremadura	0.5137**	0.0169	−0.2604	0.2833
Ribatejo	0.2112	0.2103	−0.5152*	0.0563
Alentejo	0.7446***	0.0000	−0.0404	0.7841
Algarve	−0.3552	0.1837	−0.2769	0.1637
Terras do Sado	Reference		Reference
Bottle	NA		1.6380***	0.0001
2006	0.6912*	0.0586	0.4532**	0.0165
2005	0.2212	0.1588	0.0604	0.7411
2004	0.1517	0.2465	0.2009	0.2340
2003	0.2982**	0.0268	0.7906***	0.0002
2002	0.6655***	0.0002	0.5155**	0.0135
2001	0.3868**	0.0249	0.8095**	0.0163
1991/2000	0.5914***	0.0042	0.7869***	0.0022
No year	Reference		NA
Reserve	0.6797***	0.0000	0.7168***	0.0024
Selection	0.5258**	0.0417	1.1461***	0.0000
Private label	−0.4937***	0.0000	−0.4300***	0.0008
Exclusive	−0.1775*	0.0940	−0.3777	0.1334
Discounted	−0.4875**	0.0118	−0.3146	0.2707
Bargain	−0.7673***	0.0000	−0.2067	0.2877
Estate wine—Quinta	0.4687***	0.0000	0.9759***	0.0001
Estate wine—Herdade	0.0205	0.9123	0.4695	0.1342
Cooperative wine	−0.5932***	0.0003	−0.5648***	0.0004
37.5 cl	Reference		Reference
75 cl bottle	NA		−1.1069***	0.0065
Source: own calculation. ***significance at 1%; **significance at 5%; *significant at 10%.

Table 7. Results of the hedonic model on the segment of table wines

Parameters	Estimate	p-value
Intercept	0.6434***	0.0000
Red	−0.0040	0.9379
Rosé	0.5646***	0.0000
White	Reference
Bottle	0.1166	0.1475
Box	Reference
Private label	−0.3464***	0.0005
Exclusive	−0.1920	0.2606
Discounted	−0.2361	0.1002
Bargain	−0.4004***	0.0034
Cooperative wine	−0.1066	0.4168
Mateus/Lancers	0.4076***	0.0038
Sparkling	0.4205***	0.0008
37.5 cl	Reference
75 cl	−0.5587***	0.0001
1lt	−0.7602***	0.0000
1 – 5 to 3 lt	−0.2723**	0.0879
5 lt	−0.8862***	0.0000
Source: own calculation. ***significance at 1%; **significant at 10%.

Unsurprisingly, bottled wines show a large price premium for regional wines, compared to wine in boxes, while in the case of table wines the packaging receives a non-significantly different valuation (AOC wines in the sample are all in bottle, therefore this variable is missing). On the other hand, coefficients of the volume variables show that the larger the packaging, the cheaper the wine, as expected from theory, due to a scale effect (also these variables do not appear in the AOC segment, where we only have 75 cl bottles).

From the class of manufacturers, it is possible to notice that an estate wine from a ‘Quinta’ is implicitly valued more than a normal wine, while the same is not the case for ‘Herdade’ wines, with an average premium to regional wines compared to AOC. On the other hand, cooperatives call for similar average discounts for AOC and Regional wines, while the price of table wines is not affected if wines come from cooperatives. Private labels receive a discount compared to producer brands, which is lower for table wines and higher for AOC wines. In the table wines segment, Mateus and Lancers are brands that receive a large premium, as they are clearly diversified products. Mateus and Lancers are two very popular brands of wines, and their producers have preferred to differentiate the products by brand rather than by origin. Lacking a geographical reference, they have been included in the table wines segment, in line with the classification by the retailer. Vintage years show a general trend of higher prices for older wines as compared with non-vintage wines in all segments. However, AOC wines show a high premium for 2006 wines, while aging tends to give higher price premiums to regional wines. Aged Reserve wines receive similar premiums for both AOC and regional wines, while selection wines tend to give more value to regional wines when compared to AOC. In the table wine segment, sparkling wines also call for a significant price premium.

When considering supermarket strategies, we notice that a wine sold exclusively in this supermarket chain receives a discount only if it belongs to the AOC segment (although the significance is only at 10%). Unsurprisingly, sales of discounted or bargain wines reduces the value of AOC wines, while results are not significant for the Regional segment. On the other hand, while discounted wines are not significantly cheaper in the table wine segment, bargains in this segment tend to have a significantly lower price. The significance of price-related variables could suggest that demand for AOC wines is more price-sensitive, and the retailer we are analysing could use wines in this segment as loss-leaders.

When considering the local origin of the AOC, AOC wines from the Alentejo were found to have the highest premium when compared to AOC wines from the Terras do Sado region, followed by AOCs from Estremadura, Douro, and Minho. AOCs from the Beiras region (which includes two important AOCs, Dão and Bairrada8), Ribatejo and Algarve, on the contrary, show no premium if compared to wines from Terras do Sado. These findings are consistent with Ribeiro and Santos (2008), who find that Alentejo, Douro and Vinho Verde are the regions preferred by consumers. The relatively strong position of the AOC from the Estremadura region is instead rather surprising. However, producers in this region tend to sell wines mostly in bulk, and the few manufacturers who entered the retail market had to conform to the standards required, supplying products that in our dataset appear to have a good quality level. Santos and Ribeiro9 (2005) found the highest premium for AOCs from Alentejo and Douro, followed by the Dão area. The lack of premium for AOC Dão and Bairrada can be justified by the lower development and slower progress towards conformity with national policies (Simões, 1994), which probably generated a sceptical response in retailers if compared with the remaining AOCs; alternatively, they may have been unable to gain bargaining power which has penalised them when supplying large retailers. When considering the regional wines segment, it is possible to see that only Algarve, Alentejo and Estremadura are not significantly different from wines from the Terras do Sado, while all the other areas attract a significant discount.

In the simple total pooled model (Table 8), it is possible to notice that regional origin per se attracts a price premium with the exception of products from the Algarve (however, in our sample these wines are not widely represented). The remaining coefficients are consistent with the previous models, and are not given further comment. The only different coefficient is that of ‘Herdade’ wines, which is now significantly different from 0. A more important and somewhat surprising finding is that the three segments in the market are valued, ceteris paribus, in the same way. In fact, an AOC or a regional wine on average does not attract any price premium as compared with table wines, which appears to be theoretically surprising. The only piece of evidence in recent research that finds unwillingness to pay for traceability labelling alone is Hobbs et al. (2005). However, Hobbs et al. refer to the case of meat, which is a much less diversified product than wine, from a demand perspective, and a comparison between the two cases is difficult. A possible explanation for our result may be that the premium the retailer pays for areas of origin already includes an implicit premium for AOC. In fact, in a wine producing country, buyers are very likely to attach more importance to the area of production, knowing a priori its fame for AOC/Regional wines.

Table 8. Results of the hedonic model on the simple pooled total sample

Parameters	Estimate	H.C. p-value
Intercept	0.5938***	0.0000
Red	0.0964*	0.0685
Rosé	0.5077***	0.0000
White	Reference
Minho	0.8417***	0.0001
Douro	0.9152***	0.0000
Beiras	0.5547**	0.0115
Estremadura	1.0343***	0.0000
Ribatejo	0.7343***	0.0014
Alentejo	1.1041***	0.0000
Algarve	0.1426	0.6132
Terras do Sado	0.9045***	0.0001
No regional origin	Reference
Bottle	0.1850**	0.0243
Box	Reference
AOC	−0.1815	0.3722
Regional	−0.2916	0.1167
Table wines	Reference
2006	0.4066**	0.0284
2005	0.2165*	0.0600
2004	0.2017**	0.0442
2003	0.5118***	0.0000
2002	0.6345***	0.0000
2001	0.6378***	0.0000
1991/2000	0.7663***	0.0000
No year	Reference
Reserve	0.6321***	0.0000
Selection	0.5496***	0.0090
Private label	−0.4056***	0.0000
Exclusive	−0.2037**	0.0301
Discounted	−0.3504***	0.0053
Bargain	−0.5541***	0.0000
Estate wine—Quinta	0.6532***	0.0000
Estate wine—Herdade	0.3385**	0.0315
Cooperative wine	−0.4203***	0.0000
Mateus/Lancers	0.4432***	0.0014
Sparkling	0.4090**	0.0216
37.5 cl	Reference
75 cl	−0.6162***	0.0000
1 lt	−0.7502***	0.0000
1 – 5 to 3 lt	−0.1581	0.3966
5 lt	−0.8946***	0.0000
Source: own calculation. ***significance at 1%; **significance at 5%; *significant at 10%.

However, in order further to investigate the issue, as on average the premium for AOC is not significantly different from zero, we review the assumption of a common AOC premium. In other words, having found that on average the AOC labelling gives no premium, we allow for the possibility of specific regions having different premiums or discounts. The results of the model allowing for a slope shift, as from equation (4), are reported in Table 9. The coefficients of the interaction variables are relative to the interaction between AOC labelling and the region Terras do Sado. Results highlight that in general regional origin supports findings of a positive premium compared to generic table wines, with the only exception being wines coming from the Minho region. In the case of the Minho, the premium is all contained in the presence of a denomination of origin within the region. The specific presence of AOC labelling, instead, becomes significant in discounting price, as if the retailer is selling AOC wines for less, ceteris paribus. A possible justification would come from the fact that higher quality AOC wines are sold in individual retail outlets (e.g. wine cellars), which can justify higher premiums, while in supermarkets AOC wines are not as highly priced. Another possibility could arise when considering whether a negative AOC decreases the value of the intercept. This would suggest that a basic white wine with neither year nor particular characteristic would cost less if it had a denomination of origin from Terras do Sado as compared with a table wine with the same label. This may suggest that costs of production are lower in those areas receiving discounts, the reason for which might be the lower level of development of some of those areas producing AOC (delivering wines with lower quality), as well as less effective marketing practices than producers of non-AOC wines. A final possibility could be that, due to the generally small size of wineries in AOC areas, the winemakers in those areas suffer from an unfavourable bargaining power when supplying large supermarkets.

Table 9. Results of the hedonic model on the pooled total sample with slope shift

Parameters	Estimate	H.C. p-value
Intercept	0.5946	0.0000
Red	0.1170**	0.0291
Rosé	0.5124***	0.0000
White	Reference
Minho	−0.0550	0.3024
Douro	0.2527*	0.0718
Beiras	0.4593***	0.0006
Estremadura	0.7775***	0.0001
Ribatejo	0.5762**	0.0229
Alentejo	0.7311***	0.0000
Algarve	0.3398**	0.0149
Terras do Sado	0.7569***	0.0000
Table wines	Reference
Bottle	0.1723**	0.0322
Box	Reference
AOC	−0.4371***	0.0121
2006	0.4581***	0.0095
2005	0.2196*	0.0560
2004	0.2133**	0.0301
2003	0.5156***	0.0000
2002	0.6621***	0.0000
2001	0.6374***	0.0000
1991−2000	0.7951***	0.0000
No year	Reference
Reserve	0.6361***	0.0000
Selection	0.5644***	0.0068
Private label	−0.4133***	0.0000
Exclusive	−0.2102**	0.0280
Discounted	−0.3719***	0.0040
Bargain	−0.5290***	0.0000
Estate wine—Quinta	0.6325***	0.0000
Estate wine—Herdade	0.3028*	0.0728
Cooperative	−0.4496***	0.0000
Mateus-Lancers	0.4601***	0.0017
Sparkling	0.4303**	0.0162
37.5 cl	Reference
75 cl	−0.6360***	0.0000
1 lt	−0.7509***	0.0000
1 – 5 to 3 lt	−0.2204	0.2096
5 lt	−0.8681***	0.0000
AOCxMinho	1.2050***	0.0000
AOCxDouro	0.9376***	0.0000
AOCxBeiras	0.2993	0.1754
AOCxEstremadura	0.4681	0.1408
AOCxRibatejo	0.3842	0.2207
AOCxAlentejo	0.7514***	0.0005
AOCxAlgarve	−0.0493	0.8487
AOCxTerras do Sado	Reference
Source: own calculation. ***significance at 1%; **significance at 5%; *significant at 10%.

In the description of the results of the interaction variables, it should be noted that all AOC wines are on average discounted (due to a negative coefficient for AOC), and the region-specific AOC premium has to add up to the initial discount given by the presence of a generic AOC label. The interaction between denomination and regions highlights that while Minho has a premium due to the denomination of origin and not specifically to the region, other regions such as Beiras, Algarve, Ribatejo and Estremadura have a regional premium, while the AOC label adds no other premium. On the other hand, the two most important regions of production, Douro and Alentejo, receive an individual regional premium as well as a specific AOC price premium, and in all cases the individual AOC premiums are higher than the average collective AOC discount. These results imply that only the premium for AOC labelling of Minho, Douro, and Alentejo compensates for the loss as a result of the implementation of an AOC policy, while in the remaining regions use of AOC labels will imply a lower price. For the remaining regions (Algarve, Beiras, Estremadura, Ribatejo, and Terras do Sado) it is their regional premiums which will offset (in part or in total) the discount due to the AOC labelling, not their region-specific AOC premiums.

The remaining coefficients overall are consistent with the previous models, as described in the previous paragraphs. The only difference is, again, the significant price premium attaching to estate wines bearing a ‘Herdade’ origin.

Final Considerations and Conclusions

Recent policies in the Portuguese wine sector have tried to focus on the development of the wine sector, implementing European directives oriented to an increase in the quality of wines. These policies focus primarily on local development of established wine districts, increasing the number of AOC regions and improving, through new institutions, the traceability of products coming from specific geographical areas. The assumption behind this strategy is that wine is valued for its regional origin, and for the reputation that a region has acquired over time. Through a hedonic model, we tested this assumption, obtaining some relevant information and a better understanding of the market.

In general, confidence in the name of certain regions is a key added value in the selection of wines, and consumers are willing to pay more for those wines coming from very reputable regions such as Alentejo, Estremandura, Douro, and Minho, which show a consistent competitive advantage for AOC wines. On the other hand, regional wines that receive the highest credit are from Terras do Sado, Algarve, Alentejo and Estremadura.

However, in our work we find that AOC labelling is not a factor attracting a price premium per se, but rather that it is the interaction between the AOC and the region of production that actually gives a premium. Our results show evidence of premium discrimination depending on the reputation of the region of origin. De facto, in our sample, the AOC labelling alone captures a significant discount, which suggests that ceteris paribus AOC wines are offered for sale at a lower price compared to the rest of the products. This casts doubts on the economic viability of the AOC label in certain regions, as the regional AOC premium does not always compensate for the loss of the overall discount given by the labelling.

It is not clear to what extent our results can be generalised across the whole market, due to limitations of the dataset. However, even if results are specific to one single retailer, and if the prices we use are not different from actual retail prices, our findings show the existence of a pricing scheme that disadvantages denominations of origin, and those producers coming from areas with a generally low regional reputation. Whether this is due to the low bargaining power of AOC producers, to a lower developmental level in certain AOC regions, to inappropriate marketing strategies or to demand-led pricing is not clear, and the issue calls for further investigation. What remains is that, in general, small AOC regions have higher production costs due to their small scale and production constraints, which are not compensated for in the market. These outcomes need further analysis and consideration: our results describe the market on a national scale, and preferences on a local/regional level may differ consistently in terms of premiums and discounts.

In conclusion, the use of hedonic modelling segmentation for quality level seems to give results consistent with the economic reality of the Portuguese wine market. The accurate selection of variables through consumer studies, and the use of different models for different quality segments, identified the strengths and weaknesses of different quality levels. The results of the reduced model also proved to be more insightful than the total model, allowing a deeper understanding of the sector, and finding positive feedback by the reference study we used. Hence, hedonic modelling can be a very powerful econometric tool, but we acknowledge that it needs a careful specification of variables, possibly with the support of consumer studies, in order to obtain coherent results.

Acknowledgements

We are grateful to Francisco Areal and Viviana Albani for their support in the preparation of this paper, and Steven Barlow for support in editing the paper and to two anonymous referees for useful comments that substantially improved this work.

Notes

This is a fortified wine.

This is a fortified wine.

We thank an anonymous referee for suggesting the need to include this section.

In our dataset, we have not included sparkling, foreign and fortified wines.

The observations included in this category are defined as:

Reserva: It is assigned to AOC and regional wines, in glass bottles, associated with the vintage year, giving to the wine unique sensory characteristics and a superior alcoholometry, at least 0.5% vol. to the minimum limit fixed by the legislation (www.vinhos.online.pt).

Colheita seleccionada: reserved for table wine with geographic indication and AOC, conditioned in glass bottles, that presents detached organoleptic characteristics, an acquired minimum alcohol strength superior, at least, by 1% vol., to the legally minimum limit, with indication of the year of harvest (http://www.vinhoverde.pt/pt/vinhoverde/glossario); this characteristic was modelled jointly with the next definition.

Grande escolha: “Mention reserved for a QWPSD and regional wines, in glass bottles, presenting unique sensory characteristic, having to consist of a specific accountability, which, associated to the vintage year, can be identified as Great selection (grande escolha)” (http://www.vinhoverde.pt/pt/vinhoverde/glossario).

http://www.ivv.min-agricultura.pt/vinhos/index.html

It is a wine made by grapes coming exclusively from a specific estate (Quinta or Herdade, in Portuguese) mentioned in the name of the wine. The use of the word is defined in law (Decree no. 1084/2003, 29 September 2003).

We also tried to include them separately, obtaining similar results and both regions had coefficients not significantly different from themselves, and the two AOC were kept together for a matter of consistency among regressions.

In their study, it is not clear whether the data they collect are actually transaction prices (i.e. at equilibrium), or proposed sales prices, as in our case, so it not clear how comparable results are.