Parc Mediterrani de la Tecnologia Edifici ESAB Carrer Esteve Terradas, 8 08860 Castelldefels, Barcelona CONSUMERS’ WINE PREFERENCES IN A CHANGING SCENARIO: A GENERALIZED MULTINOMIAL LOGIT APPROACH Cristina ESCOBAR, Zein KALLAS & José María GIL CREDA – UPC – IRTA Centre for Agro-food Economy and Development Castelldefels, Barcelona (SPAIN). X CONGRESO NACIONAL DE ECONOMÍA AGRARIA “Alimentación y territorios sostenibles desde el sur de Europa”. September 9th to 11th, 2015. Córdoba, Spain. Table of contents 1. INTRODUCTION AND OBJECTIVES 1.1. Catalonia as a wine region 1.2. Socio-economic context in Catalonia 2. CONSUMERS’ PREFERENCES TOWARDS WINE 3. METHODOLOGY 4. EMPIRICAL APPLICATION 4.1. Sample 4.2. Attributes and levels 5. RESULTS AND DISCUSSION 6. CONCLUSIONS 1. INTRODUCTION: Catalonia as a wine region Wine production in Spain > 42 millions Hectoliters Catalonia 3,7 millions Hectoliters 300.000 ITALY; 44.900 250.000 200.000 Rest of Spain 1000 hl Catalonia FRANCE; 42.016 150.000 100.000 50.000 Source: OIV & DAAM, 2014 SPAIN; 42.700 REST OF THE WORLD; 148.984 0 2013 Source: OIV, 2014 1. INTRODUCTION: Catalonia as a wine region WINE SECTOR KEY FACTORS (1): CONSUMPTION DECREASE 12 LITRES/CAPITA 10 8 6 4 2 0 2005 2006 TOTAL VINS (vins+escumosos) 2007 2.008 2.009 VI TRANQUIL (DO) 2.010 V.TAULA 2.011 2.012 ESCUMOSOS 2.013 ALTRES VINS Source: MAGRAMA 2014 1. INTRODUCTION: Catalonia as a wine region WINE SECTOR KEY FACTORS (2): LOW MARKET SHARE OF CATALAN DO WINES IN CATALONIA respecto al vino con DO % in volume Catalan quality wines Market share (11 DO) 2011 2012 2013 28,5% 28,9% 29,6% Source: Nielsen panel, 2014 1. INTRODUCTION: Catalonia as a wine region WINE SECTOR KEY FACTORS (3): EXPORTATION INCREASE 700 Milions d'euros 601 600 500 542 527 543 2010 2011 597 605 2012 2013 481 450 441 2005 2006 400 300 200 100 0 2007 2008 2009 Source: DATACOMEX 2014 1. INTRODUCTION: Socio economic context in Catalonia Market INSTABILITY (since 2007) and CRISIS (Ortega & Peñalosa, 2012) DEVASTATING IMPACT on the EMPLOYMENT IN SPAIN: 6.2 million people unemployed in 2011 (INE, 2014) SHARP DROP in consumption and in fixed capital investment (Carballo-Cruz, 2011) Also consequences for the AGRO FOOD MARKET (Nielsen market trends) 1. INTRODUCTION: Socio economic context in Catalonia POLITICAL CHANGES have also occurred in Catalonia Members of Parliament strongly in favor of an INDEPENDENT CATALONIA increased 7,4% (elections of 2012) Massive POPULAR MOBILIZATIONS in Catalonia after the long awaited decision of the Constitutional Court about the new Statute of Autonomy Plus, the MAIN NATIONALIST PARTY in Catalonia (CIU), has SHIFTED from nationalism to (Guibernau, M., 2013, amongst others) Catalan independentism 1. INTRODUCTION Thus, the OBJECTIVE of this paper is: To determine CONSUMERS’ RED WINE PREFERENCES for a special occasion and, More specifically, their CHANGES regarding the newer economic and political scenario We applied 2 DISCRETE CHOICE EXPERIMENTS (DCE), BEFORE and DURING the economic crisis (1TBefore vs.3TDuring) 2. CONSUMERS- PREFERENCES TOWARDS WINE Wine is a DIFFICULT AND CONFUSING product for consumers to choose IMMENSE NUMBER OF CUES COO White Red ROO DO BRAND Rosé Liquored TYPE PRICE AWARDS Sparkling Others... PACKAGING TASTE GRAPES VINTAGE ALCOHOL CONTENT 2. CONSUMERS- PREFERENCES TOWARDS WINE BRAND Are capable of acting as a SURROGATE for a number of attributes (including quality) and might help address RISK while providing product cues Consumers may use a SMALL REPERTOIRE, which may well be a collection of true brands and GENERIC TYPES ROO DO GRAPE GENERIC TYPES VARIETY 2. CONSUMERS- PREFERENCES TOWARDS WINE COUNTRY OF ORIGIN Plays a KEY ROLE in the consumers’ decision making process, specially in wine producing countries When consumers do not have information about the PRICE product, it generally performs as a PROXY to infer the quality of the product when: The product cannot be evaluated The RISK of making a wrong choice is high 2. CONSUMERS- PREFERENCES TOWARDS WINE RISK REDUCTION STRATEGIES Compilation of the main RRS in wine choice (Rawbone-Viljoen, 2012) INFORMATION SEARCH [assistants, waiters, wine editorials, tasting notes, packaging, word-of-mouth, family and friends and opinion leaders] SEEKING REASSURANCE [mainly through tastings, information seeking behaviour and prior experience] BRAND loyalty and Well-known brands PRICE STORE IMAGE 3. METHODOLOGY: The DCE: Econometric model U jn V jn ( X j , Sn ) jn The basic model is THE MULTINOMIAL LOGIT MODEL (MNL). U njt xnjt njt n n 1, , N j 1, , J t 1, , T It imposes homogeneity in preferences for observed attribute The IIA property seldom hold. 3. METHODOLOGY: The DCE: Econometric model The Mixed Logit Model (MIXL) U njt n xnjt njt n n 1, , N j 1, , J t 1, , T Extend the MNL introducing for unobserved heterogeneity by allowing random coefficients on attributes . Recent studies argued that much of the PREFERENCE HETEROGENEITY captured by random parameters in MIXL can be better captured by the scale term; and thus known as “SCALE HETEROGENEITY”. The MIXL turns to be likely a POOR APPROXIMATION to stated data if scale heterogeneity is not accounted for 3. METHODOLOGY: The DCE: Econometric model The Generalized Multinomial Logit Model (GMNL) U njt [ n γn (1 γ) nn ] X njt njt Known also as Generalized Mixed Logit Model (G-MXL). n is a scaling factor that proportionately scales the up or down for each individual n. is a mixing parameter, and its value determines the level of mixing or interaction between the scale heterogeneity coefficient and the parameter heterogeneity coefficient . 4. EMPIRICAL APPLICATION: Sample 2 IDENTICAL SURVEYS performed in 2 DIFFERENT TIMES: BEFORE & DURING the ECONOMIC CRISIS. Before Population Sample Design Field Sample Size Confidence interval Confidence level Control measure During Consumers over 18 years who purchase regularly food and are residents in the metropolitan area of Barcelona. Stratified sample by age and postal districts using proportional affixation to the number of persons by stratum. Metropolitan area of Barcelona 400 401 4.9 ± 4,9% 95.5% (k=2) 95.5% (k=2) Pilot survey (25 questionnaires) 4. EMPIRICAL APPLICATION: Attributes and levels To reduce wine choice complexity we delimited our wine selection by focusing on a RED WINE purchased for a SPECIAL OCCASION such as Christmas. Based on the literature and discussion groups we identified the following attributes and levels: ORIGIN: Catalonia (regional), Spain (national), Imported (international) WINE REFERENCES: Own Experience, Recommendation, Prestige GRAPE VARIETY: Cabernet Sauvignon, Grenache, Merlot PRICE: €6, €10, €14 4. EMPIRICAL APPLICATION: Attributes and levels We followed the DUAL RESPONSE CHOICE EXPERIMENT design. From the FULL FACTORIAL DESIGN using the total number of attributes and levels which led to a total of 81 hypothetical products. In a choice set of 2 alternatives we have 6,561 possible combinations. The orthogonal fractional factorial design with only 9 CHOICE SETS. 4. EMPIRICAL APPLICATION 4. EMPIRICAL APPLICATION: Attributes and levels All attributes, including the price, were coded with EFFECT CODING as discrete variables To avoid the base levels being CONFOUNDED WITH THE INTERCEPT Effects of all levels can be estimated All models were estimated by using 500 Halton draws. Both models are statistically significant and exhibited a good fit with highly significant likelihood ratios. 5. RESULTS AND DISCUSSION Before During Random parameter estimates 2.03293*** .27705 Spanish 3.80717*** .73884*** Catalan -.80762** -.20509*** Recommended -1.06844*** .09827 Prestigious -1.52072*** -.25125 Grenache 1.58843*** .29349** Cabernet sauvignon .90290** .01346 Price-10€ -1.96268*** -1.19805*** Price-14€ Practically all -2.52462*** 2.86293*** No choice SIGNIFICANT -3955.00 -3964.89 Log-Likelihood (0) 947.73 (0.000) 3650.00 (.000) LL ratio test Consumers’ preferences are higher for the CATALAN ORIGIN of .1198152 .4602901 Pseudo R2 1.965 1.217 and, AIC/N for the grape variety CABERNET product, SAUVIGNON Variance parameter tau in scale 2.05073*** .05930 parameter (τ) been PREVIOUSLY EXPERIENCED wines that have .02924* .10052 Weighting parameter Gamma (γ) the for 5. RESULTS AND DISCUSSION Before During SomeRandom levelsparameter estimates 2.03293*** .27705 Spanish turn into NON3.80717*** .73884*** Catalan SIGNIFICANT -.80762** -.20509*** Recommended -1.06844*** .09827 Prestigious -1.52072*** -.25125 Grenache 1.58843*** .29349** Cabernet sauvignon .90290** .01346 Price-10€ -1.96268*** -1.19805*** Price-14€ -2.52462*** 2.86293*** No choice A change into positive utility for the No choice intercept might -3964.89 Log-Likelihood (0) explain it-3955.00 947.73 (0.000) 3650.00 (.000) LL ratio test .1198152 Pseudo R2 In During consumers show a greater preference for not.4602901 taking the 1.217 AIC/N product, indicating PERSISTENCE 1.965 IN THE UNOBSERVED Variance parameter tau in scale 2.05073*** .05930 ATTRIBUTES parameter (τ) .02924* .10052 Weighting parameter Gamma (γ) 5. RESULTS AND DISCUSSION Before During Random parameter estimates In “During” the VARIATION OF 2.03293*** .27705 Spanish model it becomes NON-SIGNIFICANT: 3.80717*** .73884*** Catalan THE DEGREE of RANDOMNESS in their final decision and hence their -.80762** -.20509*** Recommended -1.06844*** significantly. .09827 Prestigious degree of UNCERTAINTY DECREASED -1.52072*** -.25125 Grenache 1.58843*** .29349** Cabernetsauvignon Tau parameter (KEY PARAMETER) .90290** .01346 Price-10€ captures the SCALE HETEROGENEITY -1.96268*** -1.19805*** Price-14€ -2.52462*** 2.86293*** No choice -3955.00 -3964.89 Log-Likelihood (0) Common circumstances may have had a 947.73 (0.000) 3650.00 (.000) LL ratio test HOMOGENISING INFLUENCE .1198152 .4602901 Pseudo R2 1.965 1.217 AIC/N Variance parameter tau in scale 2.05073*** .05930 parameter (τ) .02924* .10052 Weighting parameter Gamma (γ) 5. RESULTS AND DISCUSSION Before During GAMMA in both models is SIGNIFICANTLY DIFFERENT from ZERO. Random parameter estimates 2.03293*** .27705 Spanish TASTE heterogeneity is PARTIALLY CONDITIONED to SCALE Catalan heterogeneity. Recommended Prestigious In “During” model, however, Grenache Cabernet sauvignonfrom 0.10 to (Gamma increases Price-10€ Price-14€ No choice Log-Likelihood (0) LL ratio test Pseudo R2 AIC/N Variance parameter tau in scale parameter (τ) Weighting parameter Gamma (γ) 3.80717*** -.80762** -1.06844*** they-1.52072*** become 1.58843*** 0.57) .90290** -1.96268*** -2.52462*** -3955.00 947.73 (0.000) .1198152 1.965 .73884*** -.20509*** .09827 more INDEPENDENT -.25125 .29349** .01346 -1.19805*** 2.86293*** -3964.89 3650.00 (.000) .4602901 1.217 2.05073*** .05930 .02924* .10052 5. RESULTS AND DISCUSSION Before Spanish 3.12735*** .25613 Catalan 5.09622*** .59020** .23246 .08569 Prestigious 1.65080*** .11011 Grenache 1.81875*** .67845 Cabernet sauvignon 2.74603*** .21026 Price-10€ Practically all Price-14€ SIGNIFICANT 3.05442*** .38428 4.86483*** .97643** No choice 7.34722*** .52561 Recommended Standard deviations of parameter distributions Regarding During the UNOBSERVED TASTE (PREFERENCE) HETEROGENEITY, it is captured by the standard deviation of the random parameters. 5. RESULTS AND DISCUSSION Before Spanish 3.12735*** .25613 Catalan 5.09622*** .59020** .23246 .08569 Prestigious 1.65080*** .11011 Grenache 1.81875*** .67845 Cabernet sauvignon 2.74603*** .21026 Price-10€ 3.05442*** .38428 Price-14€ 4.86483*** .97643** No choice 7.34722*** .52561 Recommended Standard deviations of parameter distributions During Some levels turn into NONSIGNIFICANT 5. RESULTS AND DISCUSSION In the During model, Spanish wine’s show a NOT SIGNIFICANT UTILITY but its TASTE HETEROGENEITY TURNS SIGNIFICANT Catalan wines show again a SIGNIFICANTLY POSITIVE UTILITY and its TASTE HETEROGENEITY TURNS NOT SIGNIFICANT Furthermore, the SCALE HETEROGENEITY = ZERO CONSUMERS’ PREFERENCES TOWARDS CATALAN wines have become CLEAR 6. CONCLUSIONS Consumers’ preferences are higher for the CATALAN ORIGIN of the product, for the grape variety CABERNET SAUVIGNON and for wines that have been PREVIOUSLY EXPERIENCED The CATALAN ORIGIN of the wine shows a SIGNIFICANTLY POSITIVE UTILITY IN BOTH SURVEYS, which reveals the importance of the Catalonian identity in the consumer behaviour. However, During the crisis, this quality is HOMOGENEOUS ACROSS CONSUMERS (does not show any unobservable heterogeneity). This finding ENVIRONMENT is in ACCORDANCE with the POLITICAL 6. CONCLUSIONS SPANISH WINES shows a NON-SIGNIFICANT UTILITY During the crisis, this occurs in spite of gathering the highest market share in Catalonia, which suggests an influence of the political changes 6. CONCLUSIONS The GMNL model has shown to be appropriated to DECOUPLE both UNOBSERVED HETEROGENEITIES Has provided us with more information about the SOURCE OF CONSUMERS’ HETEROGENEITY In the “During” model: the results for the SCALE HETEROGENEITY indicate that the degree of uncertainty in the decision-making process has DECREASED SIGNIFICANTLY. This finding might show an IMPACT OF SOCIO ECONOMIC CHANGES in the environment of consumers’ decision-making towards wine. In this sense, external common circumstances may have had a HOMOGENIZING INFLUENCE.