Functional foods as differentiated products: the Italian yogurt market

1. Introduction

Thanks to the increased consumers' interest for health and wellness, the market for functional food products (claiming to provide health benefits beyond the traditional nutrients they contain)¹ has experienced strong growth. During the period 2004–2007, the sales of fortified and functional packaged goods in western Europe have experienced a rate of growth exceeding 10 per cent (The Economist, 2009), reaching a value of approximately USD 0.8 billion in 2006 (Datamonitor, 2007).

Academics have shown a strong interest for this phenomenon (see Siró et al., 2008 for a review of the literature on functional foods). Many studies have investigated aspects related to the demand for functional products, such as consumers' willingness to pay for food with health-enhancing features (e.g. West et al., 2002; Markosyan, McCluskey and Wahl, 2009) or consumers' attitude towards them,² using survey data rather than data on actual consumer purchases. Surprisingly, only few studies have tried to assess the demand for these products and to characterise their markets using actual sales data: Yuan, Capps and Nayga (2009), for example, investigate sales cannibalisation due to the presence of a functional alternative in a given category (orange juice), while Bonanno (2011) analyses how health-related socio-demographic characteristics impact the odds of success of functional versus conventional yogurts in Italy.

Furthermore, no study so far has assessed functional foods' market performance and the extent of their differentiation. Functional foods are used by food manufacturers to attract new customers and to revitalise mature segments (Heasman and Mellentin, 2001), and their higher margins necessary to recover (i) the large R&D costs incurred in the development of the functional attributes,³ (ii) high marketing costs, and (iii) diseconomies of scope which may arise from excessive product-line length (Draganska and Jain, 2005) and from the failure to support the already existing core products (Herath et al., 2008). If functional foods' manufacturers fail in their differentiation strategy, as prices increase consumers may be less likely to purchase the more pricey functional alternatives and more likely to switch to the conventional ones. Thus, understanding consumers' purchasing patterns for functional and conventional alternatives is key to understand the likelihood of success of these products and their profitability.

The lack of characterisation of the market of functional products (and of their performance) is surprising, particularly in the light of the rapid transformation it is experiencing. For example, in Europe, functional foods are under scrutiny due to the implementation of Regulation (EC) No. 1924/2006, regulating food products' health claims. The European Food Safety Authority (EFSA) has set up a panel of experts (i.e. the panel on Dietetic Products, Nutrition and Allergies – NDA) in charge of reviewing the validity of each product's health claim, whose truthfulness has to be supported by scientific evidence. In November 2009, the EFSA announced its first decisions on 523 claims – two-thirds of which were negative. Since then, food industry pundits have been concerned that the lack of transparency of EFSA's protocols could generate a climate of uncertainty jeopardising innovation and growth of the European food industry (Starling, 2009). Such uncertainty has not spared large companies: for example, Danone (which shared the support that the Yoghurt and Live Fermented Milks Association gave the legislation) withdrew in April 2009 two heath claim (Article 13.5) submissions: a digestive health claim for Activia and an immunity claim for Actimel (drinkable), seeking further guidance from EFSA about scientific requirements. In August 2009, the company submitted an Article 14 disease reduction claim for Actimel and, in November of the same year, an Article 13.5 (health claim) for Activia, which were both denied (Starling, 2010).

This study contributes to the understanding and characterisation of functional foods' markets by pursuing the following objectives: (i) to characterise the demand for functional and conventional products inside one product category; (ii) to investigate the patterns and the determinants of consumers' switching between conventional and functional alternatives; and (iii) to provide an empirical assessment of the profitability of these products. To achieve these goals, Deaton and Muellbauer's (1980) Linear Approximated–Almost Ideal Demand System (LA/AIDS) model is modified following Rojas and Peterson's (2008) adaptation of Pinkse, Slade and Brett's (2002) Distance Metric (DM) method, and the model applied to a scanner database of yogurt purchases in 16 Italian regions, encompassing 18 conventional and 12 functional alternatives.⁴ The DM method builds on the concept that products more distant in the attribute space are less likely to be substitutes to one another. This method allows for a flexible substitution pattern while keeping the analysis tractable (e.g. only one equation needs to be estimated, even when a large number of alternatives are considered, and the number of the estimated parameters is heavily reduced). Although other frameworks exist that allow for flexible substitution patterns and ease of tractability (e.g. the random parameter logit by Berry (1994) and Berry, Levinsohn and Pakes (1995)), combining the DM method with the LA/AIDS model permits to quantify the role of product attributes on price-driven switching patterns. The Italian yogurt market is chosen as a case study since large yogurt manufacturers operating in this market (Danone, Parmalat and Nestlé) have heavily invested in adding new lines of functional products.⁵

The results show that the demand for functional yogurts in Italy is on average less price-elastic than that for conventional ones, and that brand loyalty plays a major role in this market. Also, although price does not appear as a strong determinant of switching between conventional and functional yogurts, intra-brand switching between functional and conventional yogurts is more likely than inter-brand, suggesting that the different functional attributes existing across brands, as well as effective marketing campaigns, may lead to substantial switching costs. Lastly, the results suggest that, in most cases, functional yogurts generate higher short-run margins than their conventional counterparts.

2. The model

In the discussion that follows, the demand side of the model is treated first, highlighting the rationale and some of the technical details of incorporating the DM method into an LA/AIDS model. Then, illustrating the supply-side of the model, emphasis is given to the specific assumptions regarding the hypothesised market structures (Nash–Bertrand equilibrium) and on the characterisation of the derived measures of profitability.

2.1. The demand side

To circumvent this issue, each cross-price parameter is assumed to be a function of the distance in the attribute space between products j and k. This approach, the DM method, was originally developed by Pinkse, Slade and Brett (2002) to analyse spatial price competition in the US wholesale gasoline market. It has been used first in demand analysis by Pinkse and Slade (2004), and Slade (2004), who applied it to ‘representative consumer's’ linear demand systems based on a normalised-quadratic indirect-utility function, and then adapted to the LA/AIDS model by Rojas (2008), Rojas and Peterson (2008), and Pofahl and Richards (2009).⁷

The sign and magnitude of the estimated characterise the structure of consumers' switching motivated by a price increase. A positive would, for example, suggest that consumers are more likely to respond to a price increase by switching to another alternative produced by the same manufacturer, i.e. that brand (vendor) loyalty plays a role in this market. Similarly, if the coefficient associated with closeness in the functional attribute is positive, consumers are more likely to switch within either conventional or functional yogurts than between them. If instead consumers are more likely to switch from a functional to a conventional yogurt (or vice versa). The other coefficients have similar interpretations.

2.2. The supply side

For the illustration that follows, it is assumed that yogurt manufacturers in the Italian market play a Bertrand game, and that prices are the outcome of a Nash equilibrium of such game. Assuming the existence of a Nash–Bertrand equilibrium is fairly common in the analyses of profit margins for firms operating in differentiated product markets. Since seminal works such as Hausman, Leonard and Zona (1994), countless applications using this assumption can be found in the literature.¹⁰ Among the many studies that use models based (at least in part) on the Nash–Bertrand assumption are Berry, Levinshon and Pakes (1995), investigating profit margins in the US car market; Nevo (2001), analysing the performance of ready-to-eat breakfast cereals manufacturers in the US; Slade (2004), testing the existence of unilateral versus coordinated effects in UK brewing; Di Giacomo (2008), evaluating the effect of new product introductions in the Italian yogurt market, and Rojas (2008), considering Bertrand–Nash as a possible structure of the competition among beer manufacturers in the US.

The reader should notice that the estimated PCMs obtained from equation (8) will be short-run margins and therefore do not account for fixed costs. As such, the estimated margins will be upper bound estimates of the profitability of the yogurts sold in the Italian market. Given the larger R&D costs associated with functional foods' development (Menrad, 2003), it is conceivable that the PCMs obtained from equation (8) will overestimate the profitability of functional products more than that of conventional ones.

3. Data, model specification, and estimation

Equation (5) is estimated using primarily a scanner database provided by the Food Marketing Policy Center at the University of Connecticut, supplied originally by Information Resources, Inc. (IRI, 2004–2005). The data include 24 monthly observations of yogurt sales (quantities and values) for the period January 2004–December 2005 in hyper- and supermarkets located in 16 Italian IRI regions,¹¹ for a total of 384 market combinations. Thirty products¹² are identified by vendor (Danone, Granarolo, Nestlé, Müller, and Parmalat, referred below as brands), flavour (plain, fruit, and other flavours), fat content (skim and whole),¹³ drinkable versus non-drinkable, and by the presence of a functional attribute, for a total of 11,520 observations. Volume and value of sales are used to calculate prices in EUR/kg.

Attributes information was collected from the manufacturers' websites or, when not available, from www.ciao.it, a website where Italian consumers share opinions on purchase experiences, often reporting the nutritional content of food products as they appear on the nutritional labels.¹⁴ The continuous product characteristics used in the analysis are protein, sugar, fat content and calories, referred to 100 g of product.¹⁵ Table 1 presents a summary statistics of the data for the 30 products included in the analysis, including product characteristics, price and expenditure shares. Product characteristics from the IRI data include average volume per unit (kg/unit) and a proxy for market coverage (average number of items per store). Additionally, monthly and regional dummies are included in the estimation to capture seasonal variation in yogurt consumption and unobservables across regions, respectively.

One challenge in estimating equation (5) was to maintain a flexible model specification while trying to mitigate the risk of multi-collinearity resulting from using the same product characteristics to shift the model's parameters. In the first place, average volume per unit and market coverage were chosen as intercept shifters (⁠⁠) so that one could limit the use of physical product characteristics as own-price and expenditure shifters (⁠ and ⁠, respectively).¹⁶ Fat content and calories were used in a mutually exclusive way as part of either or ⁠, because of the large correlation of these two variables (0.87).

Thirty-two model specifications were estimated: 2 ‘full’ specifications and 30 restricted ones. In the two ‘full’ specifications, the vectors of discrete product characteristics brand (i.e. vendor: Danone, Granarolo, Müller, and Parmalat) and flavour indicators (fruit, other flavours, and plain)¹⁷ are part of both and ⁠: while in the first (second) specification fat (protein and sugar) content shifts the log-price (expenditure) parameter, protein and sugar (fat) content shifts the category expenditure (log-price) parameter. The 30 restricted specifications were obtained excluding sequentially brand and flavour indicators from and ⁠. Model selection was performed by monitoring the average VIF¹⁸ to reduce the presence of multi-collinearity, the significance and sign of the key estimated parameters, and magnitude, sign, and significance of estimated elasticities.

Estimates of the parameters of equation (5)  can be biased if prices are correlated with demand shocks unaccounted for by the other variables in the model. Endogeneity was detected in the ‘full’ models using C statistics, obtained as the difference of two Sargan statistics (Hayashi, 2000: 232). To ensure unbiasedness of the estimates, an instrumental variable estimation method (Generalised Method of Moments – GMM) was adopted using variables related to yogurt manufacturing and retail costs as instruments for price. Such instruments are: farm-level milk price (national, monthly, EUR/l) price of cream at the origin (national, monthly, EUR/kg) interacted with fat content, farm-level national price of fruit (national, monthly, EUR/kg), taken all from the DATIMA database by ISMEA (Institute for the Study of Agricultural Markets); the producer price index for the dairy industry (national, monthly) by ISTAT (the National Institute of Statistics); the European import price (CIF) of sugar (monthly, USD/lb) by Index Mundi; retail workers' per capita earnings (regional, annual, thousand EUR) by OIC (Italian Observatory of Commerce – Ministry of Economic Development); the industrial price of heating oil (national, monthly, EUR/hl) by DGERM (Office of the Director General of Energy and Mineral Resources – Ministry of Economic Development); and the commercial price of electricity at the source (regional, monthly, EUR/Mw) by the Manager of the Energy Market (GME).¹⁹

The instruments' orthogonality was evaluated using Hansen's (1982),J-statistic, while problems of weak instruments were ruled out using Staiger and Stock's (1997) ‘rule of thumb’ (the value of the F-statistics for the joint significance of the instruments' parameters in the first-stage equation exceeds 10 in all models). Following Blundell and Robin (2000) and Dhar, Chavas and Gould (2003), category expenditure is also treated as endogenous, and it is instrumented by regressing it on median household income from ISTAT, its squared term, a (monthly) time trend, and region dummies.²⁰ The different specifications of equation (5) were estimated in STATA v. 11.²¹

4. Empirical results

The results summarised below, and presented in Table 2, are for three specifications of equation (5); in those specifications, fat content shifts the own-price parameter, while protein and sugar content shift the expenditure parameter. Besides the results of a ‘full’ model specification (left columns), the results of two restricted specification are reported: one where brand indicators shift the own-price parameter and flavour indicators shift expenditure's (Rest1) and one where no discrete characteristics shift the own-price parameter and only flavour indicators shift expenditure parameter (Rest2). The restricted models produce VIF much smaller than the full model; the average VIFs are 106.24 for the full model and 26.19 and 17.11, respectively, for the two restricted models.²²

As discussed below, the estimated elasticities from the two restricted models are relatively close to one another; most restricted models whose average VIF is below 30 presented elasticities in the same range as those that will be discussed below, while models showing larger VIF presented highly unstable elasticities.²³ Similar specifications where protein content and sugar content shift own-price parameters while fat content shifts the expenditure parameter are qualitatively similar to those in Table 2, although showing larger values of VIF, and produce in most cases non-statistically significant elasticity estimates.²⁴

In terms of model performance, the full model's R-squared is 0.7871; the restricted models do not show considerable loss in explanatory power: the R-squared are, respectively, 0.7580 and 0.7440. The results in Table 2 are GMM estimates obtained using the instruments described above; OLS results are omitted for brevity and available upon request. The p-values associated with Hansen's (1982)J-statistic for the orthogonality of the over-identifying instruments vary from 0.19 for the full model to 0.36 for the specification Rest1, suggesting that the instruments used are orthogonal to the error terms. Furthermore, the large values of the F-statistics for the joint significance of the instruments in the first-stage regression indicate that the results are free from weak instrument problems.

4.1. Estimated coefficients

The own-price coefficient, not statistically significant in the full model, is instead negative and significant at the 1 per cent level in the restricted models, showing a difference in magnitude across specifications Rest1 and Rest2 (−0.0326 and −0.0465, respectively).²⁵ The coefficients associated with the interaction of log-price with fat content are negative and significant at the 1 per cent level in all models, varying from −0.0020 to −0.0032, suggesting that, on average, Italian yogurt consumers are likely to show lower willingness to pay for products with higher fat content. The interaction of the functional indicator with log price generates positive coefficients in all specifications; although the coefficient is not statistically significant in the full model, it shows significance at the 1 per cent level in the other specifications (the estimates ranging from 0.0146 to 0.0204). This result indicates that, everything else constant, Italian consumers are less price-sensitive for functional yogurts than for conventional ones, corroborating previous findings that, on average, consumers tend to show a higher willingness to pay for products that carry a functional attribute (for example, West et al., 2002; Markosyan, McCluskey and Wahl, 2009).

The role of flavours as shifters of the own-price parameters is weak, the only statistically significant coefficient in the full specification being that of plain yogurts; similar patterns are observed in other model specifications (all showing higher VIFs than those where brand indicators are interacted with log price), and are therefore not shown. The interactions of log price with vendor indicators suggest that Italian yogurt consumers are more price-sensitive for Granarolo's products (Granarolo's coefficients is negative and significant in the full model and Rest1); however, they tend not to show different price sensitiveness for Danone's and Müller's compared with Nestlé's (the excluded vendor) since the coefficients are not statistically significant in either specification. The direction of consumers' price sensitivity with respect to Parmalat is unclear.

The behaviour of the cross-price closeness measures (i.e. the weighted sum of the log prices of the other products in the market) is similar across specifications, although with some differences.

The estimated coefficient associated with is positive and significant at the 1 per cent level in two out of three specifications, varying from 0.0018 in the full model to 0.0027 (Rest1), providing evidence that consumers respond to price increases by switching to products with similar nutritional profiles. Among discrete closeness measures, closeness in brands emerges as the strongest determinant of substitution, suggesting that, when motivated by a price change, Italian yogurt consumers tend to switch within products of the same manufacturer, or in other words, that this market is characterised by a substantial level of brand loyalty (the estimated coefficients are all significant at the 1 per cent level and vary from 0.0056 of the full model to 0.0031 of Rest1). The role of closeness in flavour is unclear: its coefficient is negative and significant in the full model and positive and significant in the restricted ones. Closeness in functional attribute does not appear to be a strong determinant of substitution, as its coefficients are not statistically significant, suggesting that price changes may not be a large motivator of switching between functional and conventional yogurts. Lastly, closeness in drinkable attribute does affect the substitution across yogurts; while the coefficient is positive but marginally significant in the full specification, its parameter is negative and statistically significant in the restricted ones, although small in magnitude (−0.0010 to −0.0008), suggesting that, as price of a non-drinkable yogurt increases, Italian consumers will likely switch to a drinkable one (and vice versa).

Concluding the illustration of the estimated parameters, the unshifted expenditure coefficient is negative in all models. Most of the product characteristics used as expenditure shifters generate statistically significant parameters whose signs are consistent across specifications. In particular, the positive and statistically significant coefficients for the interactions of protein and sugar (respectively) indicate that, as the total expenditure for the yogurt category increases, those products having larger (smaller) protein (sugar) content will tend to show larger expenditure shares. In other words, expenditure elasticities (not calculated for brevity) will be larger (smaller) for yogurts containing more protein (sugar). The parameters associated with discrete product characteristics interacted with the deflated log expenditure have similar interpretation. The coefficients of the demand intercept's shifters (average volume per unit and coverage) are positive, significant at the 1 per cent level and show similar magnitude across model specification, indicating that the yogurts sold in larger sizes and whose distribution is more intense tend to show larger expenditure shares.

4.2. Own-price elasticities

Estimates of own-price elasticities obtained using the estimated parameters of the restricted models and equation (6) are reported in Table 3; as the elasticities obtained from the parameters of those model specifications showing large VIFs show unreliably wide ranges, no elasticities for the full model are discussed. The range of elasticities presented in Table 3 varies from −1.22 to −6.86 (average value −3.14), and −1.27 to −9.38 (average value −3.76) for the Rest1 and Rest2 specifications, respectively. In all cases, Nestlé/conventional/other flavours/skim is the product showing the largest elasticity, while Danone/functional/drinkable/whole shows the smallest. The magnitudes of the elasticities are consistent with other studies where consumers can substitute between numerous products (see, for example, Villas-Boas, 2007; Pofahl and Richards, 2009); furthermore, the estimated values from specification Rest1 appear close to those of other brand-level demand analysis for yogurts. Di Giacomo (2008), who presents values of brand-level elasticities of demand for yogurt in Italy ranging from −0.88 to −2.66, reports also that an alternative specification of her model produced elasticities whose average was −3.17, close to the average values obtained here. Bonanno's (2011) estimated own-price elasticity of demand for yogurt subcategories in Italy (a higher level of aggregation than that used here) varies between −2 and −4.74. In the US yogurt market, Draganska and Jain (2006) estimate own-price elasticities in a range between −2.45 and −6.25, Villas-Boas (2007) presents an average value of −5.64, while Richards et al. (2011) estimate a narrower range between −1.26 and −4.73.

Overall, five patterns emerge:

1. Functional versus conventional. On average, functional alternatives show lower own-price elasticities than their conventional counterparts, although the opposite emerges for fruit-flavoured yogurts. This pattern does not necessarily hold if one considers values across brands: for example, Müller/plain/whole elasticities are smaller than Danone's plain/functional ones.

2. Drinkable. Functional drinkable yogurts show own-price elasticities of demand below the average values, with the exception of Granarolo's.

3. Brand (vendor). The demand for Danone's yogurts tends to be less elastic than that for other brands, across flavours, fat content, and functional properties, with the exception of plain/conventional yogurts, where the ‘whole’ alternative by Müller shows the lowest magnitude of elasticity among plain yogurts.

4. Flavours. The demand for fruit-flavoured yogurts shows (on average) higher values of elasticity than that for other flavours and plain, for both conventional and functional yogurts alike.

5. Fat content. No unique trend emerges with respect to fat content and elasticities; for plain yogurts, the values are similar, while for fruit-flavoured and drinkable yogurts, whole alternatives' demand is less elastic than that for skim ones.

4.3. Cross-price elasticities

The discussion of the cross-price elasticities will consider the estimated parameters of the Rest1 specification only due to brevity, as those obtained from specification Rest2 are similar. The illustration is divided in two parts: cross-price elasticities between conventional and functional yogurts are discussed first (both intra- and inter-brand), whereas those among functional yogurts are discussed later.

Negative signs emerge for cross-price elasticities between functional and conventional yogurts (and some between functional ones) with different flavours and fat content, produced by different manufacturers (in many instances not statistically different from 0). This suggests that products located far from one another in the characteristic space are not likely to be seen as substitutes, as suggested by Kadiyali, Vilcassim and Chintagunta (1996). An alternative explanation, as suggested by Betancourt (2006), is that as households purchase multiple items during each shopping trip, different household members may prefer different alternatives belonging to the same category (i.e. different types of yogurt), resulting in some alternatives behaving as complements instead of substitutes.

Intra-brand cross-price elasticities are illustrated using Danone as an example: the values are reported in Table 4. The notation (⁠⁠) indicates cross-price elasticity of demand for a functional (conventional) yogurt to a conventional (functional) one. The estimated cross-price elasticities for Danone's yogurts are positive and statistically significant at the 1 per cent level, with and showing similar magnitude for plain yogurts; for fruit-flavoured yogurts, the are smaller than while the opposite is observed for other flavours, suggesting that Danone's plain functional yogurts may not be seen as different enough from the conventional ones to justify asymmetric cross-price elasticities. For the fruit-flavoured alternatives instead, differentiation is more marked, while for other flavours it appears weak, with consumers being more likely to purchase conventional yogurts if the price of a functional one increases than vice versa. Interestingly, the cross-price elasticities between functional drinkable yogurts and non-drinkable ones (both conventional and functional) are asymmetric, showing that, if motivated by a price increase, Italian consumers are more likely to switch from drinkable to non-drinkable yogurts.

A detailed discussion of inter-brand cross-price elasticities is omitted for brevity. Overall, three trends emerge: first, on average, cross-price elasticities among plain yogurts are larger than those among ‘other flavours’ and fruit-flavoured yogurts, whose average values are 0.42, 0.31, and 0.24, respectively. Second, values of cross-price elasticities are rather small for yogurts produced by different manufacturers and with different fat content (particularly and ⁠), some of them being negative. Third, cross-price elasticities for yogurts of the same flavour produced by the same manufacturers tend to be large (for example, the cross-price elasticity of Granarolo conventional/plain/skim to whole is 0.85, while the of Parmalat fruit/whole to skim is 0.90).

The values of cross-price elasticities among functional products, reported in Table 5, follow trends similar to those discussed above. In particular, functional yogurts of different flavours produced by different manufacturers show some negative (although not statistically different from zero) cross-price elasticities. Those produced by the same manufacturer or of the same flavour show large, positive, and statistically significant cross-price elasticities. Also, on average, cross-price elasticities for non-drinkable yogurts to drinkable ones have larger values than their counterparts.

4.4. Price–cost margins

The short-run PCMs calculated under the single-product Nash–Bertrand and the multi-product portfolio pricing for specification Rest1 are reported in Table 6. On average, PCMs increase from conventional to non-drinkable functional to drinkable ones; also, as expected, margins calculated under single-product Nash–Bertrand are lower than the portfolio pricing ones.