Chapter II Determinants of Sheep and Goat Meat Consumption in Switzerland

Research Collection

Doctoral Thesis

Analysis of final demand for food and beverages in Switzerland

Author(s): Aepli, Matteo

Publication Date: 2014

Permanent Link: https://doi.org/10.3929/ethz-a-010251132

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.

ETH Library DISS. ETH NO. 22088

Analysis of final demand for food and beverages in Switzerland

A dissertation submitted to ETH ZURICH

for the degree of DOCTOR OF SCIENCES (Dr. sc. ETH Zürich)

presented by

Matteo Federico Aepli MSc ETH Born on May 9, 1988 Citizen of Lucerne and Niederhelfenschwil

accepted on the recommendation of Prof. Dr. Michael Siegrist, examiner Dr. Michael Weber, co-examiner Dr. Christian Kuhlgatz, co-examiner

2014

This thesis was funded by the Swiss Federal Office for Agriculture FOAG within the project ―Food demand analysis for Switzerland‖.

Table of contents

Summary I Zusammenfassung V

Chapter I: General introduction 1

Chapter II: Determinants of sheep and goat meat consumption 32 in Switzerland

Chapter III: Meat and milk demand elasticities for Switzerland: 54 A three-stage budgeting Quadratic Almost Ideal Demand System

Chapter IV: Consumer demand for alcoholic beverages in Switzerland: 102 A two-stage Quadratic Almost Ideal Demand System for low, moderate and heavy drinking households

Chapter V: Endogeneity in Censored Demand Systems 164

Chapter VI: General discussion 188

Acknowledgements 209 Curriculum vitae 212

Summary

Agricultural policy decisions are based on knowledge about consumers‘ responses to income and price changes for foods and beverages. Despite its importance, this field of research has been neglected in Switzerland. The results of this thesis will provide the basis to estimate the effects of price reduction on the final demand for food and beverages, for example, due to a free trade agreement between Switzerland and the European Union or another economic area, or in the context of a tax increase for some food products or alcoholic beverages.

First, several demand models were evaluated based on their theoretical properties and empirical performance in the literature. We employed the Quadratic Almost Ideal Demand System (QUAIDS) due to its ability to approximate non-linear Engel curves and its allowance for general income responses. Aside from the base model decision, we were confronted with various methodological challenges, such as zero consumption, which leads to censored variables, and missing price information. We present econometric solutions in Chapters 2 to 5.

Price and income elasticity were estimated for meat and milk; bread and cereal products; and beverages based on the Swiss Federal Statistical Office‘s household expenditure survey. The survey consists of repeated cross-sectional data with a one- month periodicity, including information about expenditures, quantities purchased, household income, and various household characteristics. Data were available from 2000 to 2009 for approximately 3,000 households per year.

In the first study, a quality adjusted price approach was developed using the example of sheep and goat meat. This method allowed us to overcome the issues of missing price information in the household expenditure survey, and it is the basis for the other studies.

Summary

In the second study, meat and milk demand elasticities were estimated. Due to high shares of zero consumption for some product categories, the basic QUAIDS was expanded to a two-step estimation procedure where the first step allows for QUAIDS correction for censoring in the second step. The results indicated that almost all product groups are necessary goods and most of the milk and meat product categories are substitutes.

In the third study, price and income elasticity for beverages for low, moderate, and heavy alcohol drinking households were estimated applying the same two-step estimation procedure as in the second study. This study is of particular interest in the context of a tax because it allows us to quantify to what extent heavy drinking households react differently to price and income changes, as compared to moderate or low alcohol drinking households. We found that heavy drinking households are much less price elastic with respect to wine and beer in comparison to moderate or light drinking households. A tax increase on alcoholic beverages would, therefore, have little effect on public health, but it would have a negative effect on welfare, especially in low and moderate drinking households.

In the last study, we developed and tested a method that deals with endogeneity in the QUAIDS arising from a disturbed variance component. The variance component varies because different households were sampled during each section of the survey. We compared the performance of different model specifications to deal with this issue. We found that fixed effects for each section of the survey sample result in the most significant improvement to the model's fit and parameter estimates. We illustrated this with household data for bread and cereal products.

We concluded first that the QUAIDS, combined with several procedures to overcome econometric problems, is an appropriate model to estimate demand elasticities for Switzerland. Second, we discovered that own-price elasticities, especially for meat and milk products, have risen over the last few years compared to the findings in the literature for Switzerland. Third, prices and income, in addition to other determinants ii

Summary

(e.g., household characteristics), still play a major role in explaining demand. Fourth, a new tax on alcoholic beverages will not achieve the intended health effect and may even cause negative effects on welfare. Finally, the month-fixed-effects specification deals very well with endogeneity.

iii

Summary

Zusammenfassung

Um abschätzen zu können, wie Angebot und Nachfrage auf agrarpolitische Entscheide reagieren, braucht es die entsprechenden Grundlagen. Während bei der Angebotsseite in der Schweiz in den letzten Jahren viel Forschung betrieben wurde, blieb die Nachfrageseite eher vernachlässigt. Preis- und Einkommenselastizitäten für Nahrungsmittel und Getränke sind von grosser Bedeutung, wenn es zum Beispiel darum geht, den Effekt einer Preissenkung auf die Nachfrage zu quantifizieren. Das ist vor allem im Zusammenhang mit einer anstehenden Grenzöffnung des Agrarmarktes (und dem damit verbundenem Preisrückgang) oder einer Steuer auf gewisse Nahrungsmittel und Getränke von grossem Interesse. Unsere Resultate bilden die Grundlagen, um den Effekt solcher politischer Entscheide abzuschätzen.

Im Vorfeld der Schätzungen wurde eine Methodenevaluation durchgeführt, die sich vor allem auf die theoretischen Eigenschaften der zur Auswahl stehenden Modelle und den empirischen Anwendungen fokussiert. Es stellte sich heraus, dass das Quadratic Almost Ideal Demand System (QUAIDS) aufgrund seiner guten Flexibilität bei der Abbildung von Engelkurven besonders geeignet ist. Zusätzlich zur Wahl des Basismodells mussten verschiedene ökonometrische Herausforderungen bewältigt werden, so zum Beispiel der hohe Nullkonsum auf disaggregierter Ebene oder die fehlenden Preisangaben in den Daten. Ökonometrische Ansätze zur Lösung dieser Probleme werden in den Kapiteln 2 bis 5 erläutert.

Im Rahmen der vier durchgeführten Studien wurden Nachfrageelastizitäten für Fleisch- und Milchprodukte, für Brot- und Getreideprodukte wie auch für Getränke geschätzt basierend auf der Schweizer Haushaltsbudgeterhebung des Bundesamtes für Statistik. Der Datensatz besteht aus Querschnittsdaten mit einer Periodizität vom einem Monat und enthält detaillierte Ausgaben für verschiedene Produktkategorien, das Einkommen jedes Haushalts, die gekauften Mengen, wie auch verschiedene

Zusammenfassung

Haushaltscharakteristika. Die Daten lagen für die Jahre 2000 bis 2009 vor mit einer jährlichen Anzahl von 3‗000 Haushalten.

In der ersten Studie wurde ein Verfahren entwickelt zur Generierung von qualitäts- angepassten Preisen und auf die Nachfrage nach Schaf- und Ziegenfleisch angewandt. Diese Methode ersetzt die fehlenden Preisangaben in der Haushaltsbudgeterhebung und ist deshalb eine wichtige Grundlage für die weiteren Analysen im Rahmen dieser Dissertation.

In der zweiten Studie wurden Elastizitäten für Fleisch- und Milchprodukte geschätzt. Aufgrund des hohen Nullkonsums bei einigen Produktkategorien wurde das QUAIDS zu einem zweistufigen Verfahren erweitert, wobei mithilfe der Schätzung der ersten Stufe die zweite Stufe (QUAIDS) korrigiert wurde (two-step estimation). Die Elastizitäten zeigen, dass die meisten Produktgruppen Grundgüter und die einzelnen Fleischprodukte wie auch Milchprodukte untereinander Substitute sind.

In der dritten Studie wurden Preis- und Einkommenselastizitäten für (alkoholische) Getränke geschätzt und zwar getrennt nach drei Haushaltstypen: wenig-, mittel- und viel-trinkende Haushalte (bezieht sich jeweils auf den totalen Alkoholkonsum). Diese Analyse ist insofern interessant, als sie aufzeigt, welchen Effekt eine Steuererhöhung für alkoholische Getränke hat. Aus den Schätzungen wird ersichtlich, dass viel- trinkende Haushalte besonders bei Bier und Wein am wenigsten preissensibel sind. Preisänderungen wirken sich vor allem bei wenig- und mittel-trinkenden Haushalten negativ auf den Konsum auf. Mit einer Steuererhöhung würde deshalb der beabsichtigte Gesundheitseffekt kaum erreicht werden. Zudem kommt es zu grossen Wohlfahrtsverlusten besonders bei wenig- und mittel-trinkenden Haushalten.

In der letzten Studie geht es um die Frage der Modellspezifikation. Aufgrund der Struktur des Datensatzes (Querschnittsdaten) besteht das Risiko einer über die Zeit sich verändernde Varianz. Das kann zu Endogenitätsproblemen führen. Um dieses Problem zu erläutern, wurden verschiedene Spezifikationen getestet anhand der vi

Zusammenfassung

Produktgruppe Brot- und Getreideprodukte, wobei sich herausstellte, dass die Spezifikation mit einem fixen Effekt für jeden Monat die geeignetste ist.

Im Rahmen der Studien konnte festgestellt werden, dass erstens das QUAIDS mit den ökonometrischen Ergänzungen ein geeignetes Modell zu Schätzung von Nachfrageelastizitäten für die Schweiz ist, zweitens der durchschnittliche Konsument bei Fleisch- und Milchprodukten in den letzten Jahren tendenziell etwas preissensibler wurde, drittens Preis- und Einkommen beim Kaufentscheid neben anderen Faktoren wie z.B. Haushaltscharakteristika eine immer noch wichtige Rolle spielen, viertens eine höhere Alkoholsteuer kaum den gewünschten Gesundheitseffekt erwirken würde und fünftens die Modellspezifikation mit den fixen Effekten tendenziell am geeignetsten ist.

vii

Zusammenfassung

viii

General introduction

Chapter I General introduction

Chapter I

1 Introduction

1.1 Food and beverage expenditure of Swiss households

In comparison to other countries in Europe, Swiss households spend a low proportion of their total expenditure on food and beverages. Only 6.5% of total expenditure is spent on food, while 1.4% is spent on beverages – a little more than half on alcoholic beverages and the rest for non-alcoholic beverages (Table 1). The neighboring countries, Germany and France, spend an average of 11.20% and 13.84% on food, respectively (Destatis, 2013). Only the United States, Canada, Singapore, and the United Kingdom reveal similarly small expenditure shares for food as Switzerland (Destatis, 2013). The observation that the income levels of Western European countries such as Switzerland is negatively correlated with the food share, and that poorer families spend a greater proportion of their total expenditure on food than high income families, is often referred as Engel‘s law. This law has been confirmed in several studies (e.g. in Anker, 2011 and Seale & Regmi, 2006).

General introduction

Table 1: Household expenditure on food and beverages in Switzerland, 2011 (in Swiss francs)

Gross income 9 604.12 Share of expenditure on Expenditure total expenditure Total expenditure 9 085.79 100.00% Expenditure on food and beverages 718.78 7.91% Expenditure on food 591.12 6.51% Expenditure on bread and cereal products 101.48 1.12% Expenditure on meat 145.23 1.60% Expenditure on fish 20.58 0.23% Expenditure on milk, cheese and eggs 99.85 1.10% Expenditure on fat and oils 15.02 0.17% Expenditure on fruits 53.36 0.59% Expenditure on vegetables 72.42 0.80%

Expenditure on sugar, honey, confectionary 39.33 0.43% etc. Expenditure on sauces, salt etc. 43.84 0.48% Expenditure on beverages 127.66 1.41% Expenditure on coffee, tea and cocoa beverages 23.77 0.26%

Expenditure on mineral water, non-alcoholic soft 33.65 0.37% drinks and fruit and vegetable juices

Expenditure on alcoholic beverages 70.24 0.77%

Source: Swiss household expenditure survey

Within the expenditure for food and beverages, almost half is spent on meat, bread and cereal products and milk, cheese and eggs (Figure 1), while fruit and vegetables account only for 17%.

Chapter I

Figure 1: Household expenditure on food and beverages in Switzerland, 2011 (in Swiss francs)

Mineral water, Acoholic non-alcoholic soft beverages Bread and cereal drinks, fruit and 10% products vegetable juices 14% 5% Coffee, tea, cocoa beverages 3% Sauces, salt etc. 6% Meat Sugar, honey, 20% confectionary etc. 6%

Vegetables 10% Fish 3% Fruits 7% Fat and oils Milk, cheese, eggs 2% 14% Source: Swiss household expenditure survey

1.2 Prices for food and beverages in Switzerland

Swiss consumer prices for food and beverages increased until 2000 (Figure 2). Since then, prices have remained almost constant, with slight variations (SFSO, 2014). The price peak in world markets for agricultural products in 2007 and 2008 also impacted food prices in Switzerland, though this impact was much lower on the consumer level than on the producer level.

General introduction

Figure 2: Consumer Price Index for food and beverages in Switzerland

Source: SFSO (2014)

While prices for food and beverages have been at least stable or decreasing for some years, the price differences between Switzerland and its neighboring countries remains high (on average about 20%, BAK, 2010). This is mainly due to tariff and non- tariff trade barriers for agricultural products. The price differences are especially high for those product groups which are highly protected, for example meat (+38% in comparison to neighboring countries) and fat and oils (+37%) (BAK, 2010). These differences have major economic consequences and lead to economic losses at the consumer level. Therefore, Switzerland has decided to open its agricultural markets bilaterally, beginning with the EU. Though this process has stagnated in recent years, the question of how consumers react to downward price shifts, especially in regard to food, is still of great interest. Moreover, little research has been performed in this field with respect to Swiss consumers.

Chapter I

2 Research questions

Econometric demand analyses of food and beverage consumption in Switzerland are rare. However, during the last couple of years, there has been increasing interest in this research area, particularly related to price and income elasticities. These factors are important to the field of market analysis and are key parameters within market models. Market models reflect the demand of food and beverage products, while supply models reflect supply for raw agricultural products.

This dissertation is part of a research project funded by the Swiss Federal Office for Agriculture. Its goal is to estimate the final demand elasticities for a variety of product groups which are used to integrate Switzerland into the CAPRI model (Common Agricultural Policy Regionalised Impact model). CAPRI is an EU-wide partial equilibrium model consisting of a supply and demand module and allowing for estimates of the impact of different kinds of policy options, such as a free trade agreement between the EU and other economic areas (e.g. Mercosur or the United States). In this dissertation, new methods are developed and combined with state-of- the-art models to estimate the demand elasticities of policy-relevant agricultural product groups. Before estimating elasticities a literature analysis on different models was conducted to show which demand system is appropriate for estimating the price and income elasticities of food and beverages in Switzerland given the fact that some product groups at the most disaggregated level have a high share of zero consumption and the household expenditure survey does not provide prices. This part of the dissertation is mainly described in section 1.3 and Chapter 2.

With the aid of the methodological preliminary study the following research questions will be answered:

Research question 1: To what extent do consumers respond to price and income changes with respect to milk, meat, bread, and cereal products? 6

General introduction

Research question 2: What are the own-price, cross-price, and income elasticities for beverages, with a special focus on alcoholic beverages? Is there a difference between light, moderate, and heavy drinking households?

The first research question with focus on food is answered in Chapters 3 and 5, while the second research question with focus on beverages is answered in Chapter 4.

Changes in meat consumption in response to changes in income and meat prices have received particular attention in the literature. Gallet (2009, 2010), in separate studies, identified 419 and 393 studies that estimate income and price elasticities, respectively, with respect to meat consumption. Price elasticities are crucial as they allow policymakers to tailor market interventions that influence food consumption patterns. Those policies have taken many forms. For example, in recent years, the establishment of a ―fat tax‖ has been discussed several times in Switzerland. Certain food price policies that address the magnitude and composition of meat consumption may be superfluous if changing income levels shift consumption toward the desired patterns (Gallet, 2010). Therefore, it is as important to understand income elasticities as price elasticities. Another form of food price policy would be trade agreements with respect to agricultural products, with European countries, the United States, or Mercosur. Such trade agreements would lead to price reductions for certain food and beverages. In addition to meat, in Switzerland, milk (and milk products), bread, and cereal products are some of the most important product groups, from the perspectives of production and consumption. As the Swiss government is heading for a free trade for milk products as a first step – in addition to the existing agreement with respect to cheese – milk demand elasticities are of particular interest. Nevertheless, estimations of final demand elasticities in Switzerland for milk products as well as meat are scarce. For example, Schluep Campo (2004) analyzed Swiss meat demand at the wholesale level using time-series data for the period of January 1996 to December 2002, by applying a semi-flexible Almost Ideal Demand System. The latest final demand 7

Chapter I

estimation for meat and milk products and (only) aggregated cereal products was undertaken by Jaquet et al. (2000), using household expenditure data from 1998. In addition to the literature on Switzerland, Thiele (2008) conducted a similar final demand analysis for meat in Germany, by applying a linearized Almost Ideal Demand System; additionally, Bouamra-Mechemache et al. (2008) estimated price and income elasticities for milk products in France and Italy, to model the impact on demand of reforms to the Common Agricultural Policy of the EU (CAP) reform.

The second research question focuses on the demand for beverages in particular, for alcoholic beverages. As explained above, having information on elasticities allows one to estimate the consequences of policy measures on consumer demand. The most frequently discussed policy measures with respect to beverages relate to a (higher) tax on alcoholic beverages, or a tax on sugar. Both measures are expected to have positive health effects by reducing demand. In our research, we ask the question of whether heavy drinking households are less price-sensitive than low or moderate drinking households. If there is indeed such a relationship, it would have a major effect on the effectiveness of a new alcohol tax. In the case of lower price elasticities for heavy drinking households in comparison to low or moderate drinking households, a higher tax on alcohol – which is currently under discussion in Switzerland – would especially lead to decreased demand among low and moderate drinking households, but not as marked as that among heavy drinking households. Therefore, the relatively lower health effects on heavy drinking households would be probably offset by welfare losses among low or moderate drinking households. Although a considerable body of research has been amassed on alcohol demand, most studies rely on time-series data (Gallet 2007), or do not investigate the possible substitution effects between alcoholic and non-alcoholic beverages. Two studies of Switzerland have been undertaken with respect to price sensitivity and alcoholic beverages. Heeb et al. (2003) analyzed the effect of a price reduction on spirits in 1999 that was due to tariff reductions that stemmed from a World Trade Organization agreement. They found that low drinkers are more elastic than heavy drinkers. Kuo et al. (2003) used the same natural 8

General introduction

experiment as Heeb et al. (2003) and found that young people are more price- sensitive than people aged 60 or older.

3 Theoretical Background

This chapter provides background information on the classical microeconomic demand theory, which is the basis for understanding the chosen model and specifications as well as the interpretation of final demand elasticities. We mainly follow Seel‘s (1991) explanations1.

3.1 Duality

Duality is one of the most important concepts in demand analysis. It means that an optimization problem can be rearranged, in the sense that the object function turns into the constraint, and the constraint turns into the object function. In the derivation of demand functions, this "switch" characterizes the difference between the Marshallian and Hicksian approaches. The Marshallian demand is a function of prices and income (or total expenditure). Consumers are assumed to maximize utility (objective function) under a budget constraint. To illustrate, we use a Cobb–Douglas utility function: A price increase will cause consumers to decrease the quantity they demand and choose another bundle of goods that lies on a lower indifference curve:

( ) (Marshallian demand function) (1)

 max! (objective function: Cobb-Douglas utility function) (2)

(restriction: budget constraint) (3)

Normally:

1 The following chapter was partly taken from a report, which was submitted to the Swiss Federal Office for Agriculture in June 2012 (Aepli, 2012). The report has been unpublished until now (May, 2014). 9

Chapter I

where is quantity, is price, is income, and and are the elasticities of goods 1 and 2, respectively.

Conversely, Hicks assumes that households minimize total expenditure (objective function) while holding the utility level constant (constraint), implying that price changes must be compensated by income changes. The Hicksian demand, therefore, is a function of prices and utility level:

( ) (Hicksian demand function) (4)

 min! (objective function: total expenditure) (5)

(restriction: Cobb-Douglas utility function) (6)

Normally:

Briefly, Marshallian demand functions are based on the maximization of utility under a budget constraint, whereas Hicksian demand functions minimize total expenditure while holding the utility level constant. As the utility level is normally not observable, Hicksian demand functions are difficult to validate empirically. Therefore, an empirical analysis of demand will estimate Marshallian demand functions using observable variables. However, by applying the Slutsky equation, Hicksian demand elasticities (compensated elasticities) could be calculated, distinguishing the "pure" substitution effect from the income effect of a price change (see the next section).

3.2 Slutsky equation

A change in the price of one good will not only lead to a change in the exchange ratios of all other goods but also will affect the real income of households, and thus their purchasing power. The resulting effects are called substitution and income effect, and they are related by the Slutsky equation:

(7)

General introduction

where is the uncompensated (Marshallian) demand change caused by a change in price, (holding nominal constant), is the compensated (Hicksian) demand change due to a change in price, (hypothetically holding utility constant by compensating changes in income), and the third term captures the demand effect resulting from the fact that utility level actually changes without a compensating change in income.

Using some mathematical transformations and moving to infinitesimal price changes leads to a more common form of the Slutsky equation for goods and :

(8)

where S is the substitution effect and E is the income effect.

For , the substitution effect is the ―own-price substitution effect,‖ which describes how significantly a change in the own price affects demand, if changes in real income would be compensated (in the first or own-price Slutsky equation). For i ≠ j, the substitution effect is called the ―cross price substitution effect,‖ and it describes consumer reaction with respect to a price change of another good (the second or cross-price Slutsky equation).

The substitution effects are summarized by the substitution matrix (Slutsky matrix). The general form of an n-by-n matrix is:

( ) (9)

where the diagonal entries are the own-price substitution effects, and the remaining parameters are the cross-price substitution effects, which indicate whether goods are

Chapter I

substitutes (positive parameters) or complements (negative parameters). The Slutsky matrix should be symmetrical and negative semidefinite.

3.3 Elasticities of demand

Price and income elasticities measure consumer response with respect to price and income changes. Mathematically expressed, price elasticities measure the percentage change in quantities triggered by a one-percent change in price (income and other determinants held constant), whereas income elasticities measure the percentage change in quantities triggered by a one-percent change in income (prices and other determinants held constant).

For Marshallian and Hicksian demand functions, price elasticities are defined as follows:

(11)

(12)

where is the price of good , and is the quantity of good . The subscripts and indicate Marshallian and Hicksian demand functions; and are the corresponding elasticities.

If , is called an own-price elasticity. A negative own-price elasticity indicates decreases in quantity when prices increase. If , the good is considered to be elastic. A good is considered to be inelastic when the own-price elasticity is between - 1 and 0. Positive own-price elasticities are possible, though they are not very common (e.g. a Giffen good). If the quantity is not affected by the price, the good is said to be

General introduction

completely inelastic ( ). Otherwise, if , the good is completely elastic (graphically described by a vertical line).

If , is called a cross-price elasticity, and it measures the effect of a price change of good on good . A positive coefficient indicates that the goods are substitutes; otherwise, they are complements.

In addition to price elasticity, further information about consumers‘ reactions is captured by income elasticity, which describes the percentage change of quantity caused by a one-percent change in the income of households. Income elasticity could be derived directly from a Marshallian demand function:

(13)

where indicates that quantity increases with increasing income (normal goods, prices, and other determinants remaining constant). If the elasticity is greater than 1, demand increases are disproportionate (luxury good), and for 0, demand decreases with increasing income (indicating an inferior good).

3.4 Restrictions

Imposing restrictions on the parameters of demand functions improves consistency with the demand theory. A main advantage of demand systems is the possibility of imposing and testing these restrictions. There are four main restrictions:

 Adding up: This restriction guarantees that the demand function satisfies the budget constraint. The sum of all individual expenditures is equal to the total expenditure (or income) of the Marshallian and Hicksian demand functions:

∑ ∑ ( ) ∑ ( ) (14)

Deriving and multiplying with :

Chapter I

∑ ∑ (15)

where is utility and is the budget share. The other variables are as defined above. The products of the budget shares and the income elasticity must total one.

 Homogeneity: The second restriction indicates that the Marshallian demand function is homogenous (degree 0) with prices and income as well as a degree 0 Hicksian demand function of prices. In other words, for a Marshallian demand function, a proportional change in prices and income will not have an effect on demand (no money illusion). Beginning with the budget restriction and applying Euler's Theorem, we obtain the following restrictions for a Marshallian demand function:

∑ (16)

and for a Hicksian demand function,

∑ (17)

The price elasticities and income elasticity must total zero for a Marshallian demand function. For Hicksian demand functions, only compensated price elasticities must total zero.

 Symmetry: The cross-price derivatives of the Hicksian demand are symmetric (Deaton & Muelbauer, 1980b):

( ) ( ) (18)

A marginal change in the price of one good, , has the same substitution effect on as an equivalent change in the price of good on , implying that the Slutsky Matrix is symmetrical.

General introduction

 Negativity: The Slutsky Matrix must be negative semidefinite, and the diagonal entries (own-price substitution effects) must be non-positive.

(19)

For further explanations see, for example Ahlheim and Rose (1992).

3.5 Multi-stage budgeting and separability

Due to the huge variety of products in the food sector, it is important to group the products into meaningful aggregates. If this is not done properly, the elasticity estimates could be biased. At the same time, a useful grouping reduces the number of cross-price elasticities, thus preventing a reduction in the degree of freedom. Therefore, it is common to apply the concept of multi-stage budgeting. This type of budgeting assumes that a household is able to first allocate its income to aggregated product groups (e.g. food, entertainment). Afterwards, the budget provided for every product group is individually optimized and allocated to specific products within the group independently of allocation of other product groups.

Multi-stage budgeting works only if a household has the ability to form price indices to determine the expenditure for each product group, a step called ―price aggregation‖ (Brehe, 2007). For each of these product groups, we assume sub-utility functions

( ), which could be optimized individually. Together, they create the overall benefit ( ). Therefore, separability of the utility function is an important precondition.

Multi-stage budgeting and the separability of the utility function culminate in the utility tree. Normally, only weak separability is required. Further explanations are found in Gorman (1959) and Brehe (2007).

Due to multi-stage budgeting and the assumption of separability, it is important that goods that have a relationship with each other (e.g. substitutes) are organized into the same category (Deaton & Muellbauer, 1980b), and that basic goods and luxury goods should be in separate groups. 15

Chapter I

4 Data material

Our study‘s estimates are based on data from a household expenditure survey collected by the Swiss Federal Statistical Office (SFSO). It is a nationally representative survey consisting of repeated cross-sectional microdata with a periodicity of one month (for details see SFSO, 2011). Every month, about 3,000 households are chosen at random from the Swiss register of private telephone lines. The sample is stratified by the seven major regions of Switzerland (SFSO, 2011). Data were available from 2000 to 2009. Due to changes in the system over the years, some food and beverage product categories had to be aggregated to obtain consistent product groups for the whole data set. In addition to income, expenditure variables, and quantities for almost all food and beverage categories, the SFSO also collects data on different household characteristics, such as household size, age, educational level, and the number of children. Similar to many other household surveys, there is no information about prices. Therefore, we had to generate market prices using information about expenditure, quantities, and household characteristics.

We organized the data set into three aggregation levels. Stage 1 contained the most aggregated product groups, and Stage 2 contained the aggregated product groups for food as well as the disaggregated product groups for beverages (Figure 3).

General introduction

Figure 3: Structure of the product groups in the dataset

Total expenditure

Stage 1 Other products Food Beverages and services

Bread and cereal Coffee products

Meat and Fruit and meat Fruits Tea vegetable products juices

Cocoa Spirits, Fish Vegetables Stage 2 beverages sweet wines etc.

Milk, Sauces, salt Mineral cheese, Wine etc. water eggs

Sugar, honey, Non- Fat and oils confectionary alcoholic Beer etc. soft drinks

Stage 3

Source: based on Aepli (2012)

The most disaggregated food product groups are at Stage 3 (Table 2), and allow us to estimate a wide variety of price and income responses for different product groups.

Chapter I

Table 2: Product groups at stage 2 and 3

Stage 2 Stage 3 Bread and cereal products Rice, bread, pasta, sandwiches, wheat flour, other flours Beef (fresh or frozen), veal (fresh or frozen), pork (fresh or frozen), mutton and goat meat (fresh or frozen), poultry Meat and meat products (fresh or frozen, grilled or smoked), sausages and pasties, pork (processed meat), other meat (cooked, dried, salted, or smoked; e.g., game and rabbit, horse, preserved meat) Fish Fish (fresh or frozen), seafood (fresh or frozen), prepared fish and seafood, canned fish and seafood Milk, cheese, and eggs Whole milk, milk drinks, skim milk, cheese, cream, yogurt, other milk products, eggs (fresh or processed) Fat and oils Butter, margarine, vegetable fats and oil, olive oil Fruits Lemons and other citrus fruits, bananas, apples, pears and quinces, stone fruit, grapes, other exotic fruits, nuts and other edible nuts, oily fruits Vegetables Green salads and other leafy vegetables, stem vegetables, tomatoes, beans and peas, onions and garlic, turnips and other root vegetables, mushrooms and other dried vegetables, processed vegetables and mushrooms, potatoes, potato-containing products, other tuber vegetables Sugar, honey, and confectionary Sugar, jam and compote (incl. honey), chocolate, products confectionaries, ice cream

5 Model evaluation and theoretical framework2

In the literature on demand analysis, there is a methodological distinction between single-equation models (e.g. the double-log demand model) and demand systems (multi-equation models like the Almost Ideal Demand System or the Rotterdam model). Single-equation models are easy to use. Elasticities can be estimated with relatively little effort for a narrow selection of product groups. Unfortunately, these models do not allow for the implementation of the important restrictions mentioned above. In the double-log model, only the homogeneity restriction could be implemented. Therefore, these models are only partially consistent with the demand

2 The following chapter was taken from a report submitted to the Swiss Federal Office for Agriculture in June 2012 (Aepli, 2012). The report has been unpublished until now (May, 2014). 18

General introduction

theory. In contrast, demand systems allow for the implementation of most or all of the restrictions, causing them to be the preferred systems. Especially in the case of a broad elasticity estimate, demand systems are superior to single-equation models.

In addition to choosing the appropriate model, whether to conduct a time series analysis or a cross-sectional analysis had to be decided. The latter have been more common in recent publications; household data allow for detailed analyses of different income groups, and the sample size is usually larger for household data than for time series. Since being able to broadly estimate parameters is preferred, reducing the degree of freedom would lead to challenges in the case of time series data. In addition, for time series analyses, cross-price effects are often influenced by the collinearity of prices (Yen et al., 2003). For these reasons, cross-section data provides better elasticity estimates. Also, the expenditure survey provides detailed socio- demographic information. Therefore, we decided to conduct a cross-sectional analysis.

5.1 Model choice

The methodological section is built on a large body of research that began in the 1980s. Several models have been proved extensively during the last 40 years: the Almost Ideal Demand System (AIDS) model (Deaton & Muellbauer, 1980a; 1980b; Akbay et al., 2007), with all its variations (e.g. the Quadratic AIDS model) (Banks et al., 1997; Abdulai, 2002); the Rotterdam model (Barten, 1964; Theil, 1965; Barnett & Seck, 2008); the Translog (Christensen et al., 1975; Holt & Goodwin, 2009); and the linear and quadratic expenditure systems (Pollak & Wales, 1978; 1980; De Boer & Paap, 2009). The less popular models are not discussed here. Considering their features, single-equation models are not as favorable as demand systems. Therefore, an evaluation of the models is limited to the most popular demand systems. Also excluded are special models like the Florida model, which uses cross-country data. To decide which model was most appropriate, we first analyzed the theoretical properties of every model, and then we analyzed the empirical model evaluations in the literature. 19

Chapter I

The main theoretical properties of the models are shown in Table 3.

Table 3: Theoretical properties of the most popular demand systems

Model Theoretical properties  Nested within QES LES  Rank two demand system  Engel effects not as flexible as in QUAIDS or AIDADS  Does not allow for the existence of inferior commodities, elastic demand, or gross substitution (negative cross-price elasticity) (de Boer & Paap, 2009)  Rank three demand system QES  Engel flexibility limited by the fact that marginal expenditures are linear (Cranfield et al., 2003) BTL/GTL  Rank two demand system  Engel effects not as flexible compared to QES, QUAIDS, or AIDADS  Nested within QUAIDS  Rank two demand system AIDS  Engel effects not as flexible compared to QES, QUAIDS, or AIDADS  Locally flexible functional form  Satisfies the axioms of choice and allows aggregation across consumers LA/AIDS  Rank two demand system  Biased and inconsistent (Buse, 1994 and Liao & Chern, 2007)  Rank three demand system  Allows for general income responses, capture flexible Engel effects  Allows for luxury goods at low expenditure levels and normal goods at high expenditure levels QUAIDS  Satisfies the axioms of choice and allows aggregation over consumers. Exact aggregation is due to the fact that the underlying functions of QUAIDS are of Generalized Gorman Polar Form (Blackorby et al., 1978).  More parameters to estimate than in AIDS  Rank three demand system  Generalization of and more parameters to estimate than in LES (see, e.g., Van der Mensbrugghe, 2005). AIDADS  Allows for general income responses and has flexible Engel effects  Not able to aggregate across consumers (Cranfield et al., 2003)  Appropriate for considerable variations in expenditure and aggregated goods  Rank two demand system Rotterdam Model  Locally flexible functional form  Engel effects not as flexible compared to QES, QUAIDS, or AIDADS 20

General introduction

One of the most important criteria is the ability to approximate non-linear Engel curves, which is best satisfied by the QUAIDS and AIDADS. The other models, which do not contain a quadratic term for total expenditure, do not allow for this flexibility. Due to the insufficiency of the AIDADS regarding exact aggregation and disaggregated estimates, we prefer the QUAIDS. Additionally, it is easier to test whether the quadratic term is statistically significant and whether the quadratic specification is more appropriate than the AIDS with the QUAIDS.

The literature analysis is summarized in Figures 4 and 5. Although the suitability of the model depends on various factors (e.g. data material), the decision matrix helps support decision-making. The most common criteria for the model comparisons are goodness-of-fit, predictive performance, and the sign and magnitude of estimated elasticities. If the models are nested, the comparison is normally conducted by a likelihood ratio or Wald test, which is an approximation of the likelihood test and estimates only one model. Otherwise, suitability measures like AIC, SBC, root mean squared error (RMSE), system-wide RMSE (SRMSE), and information inaccuracy (IIA) are used (see e.g. Cranfield et al., 2003).

The decision matrix allows for comparisons between every model. Although the literature does not cover all comparisons, we can obtain information about the suitability of different models. For each comparison, we provide additional information about the data type and preferred model. As discussed in section 1.2, we used a cross-sectional microdata household survey. Therefore, models that focus on time series data are generally inappropriate and will not be included on the short list.

Chapter I

Figure 4: Decision matrix, part 1

LES QES BTL GTL AIDS QUAIDS

LES • Cranfield et al., 2003 • Katchova and Chern, 2004  QES 1  AIDS 1 • Pollak and Wales, 1978  QES 1 QES • Pollak and Wales, 1980  QES, +/- 1

BTL • Pollak and Wales, 1980 • Holt and Goodwin, 2009  GTL 1  +/- 2 • Lewbel, 1989  +/- 2 GTL

AIDS • Abdulai, 2002 (CH)  QUAIDS 1 • Cranfield et al., 2003  QUAIDS 1 QUAIDS

Legend: LES: Linear Expenditure System GTL: Generalized Translog Model QES: Quadratic Expenditure System AIDS: Almost Ideal Demand System BTL: Basic Translog Model QUAIDS: Quadratic Almost Ideal Demand System

1 Household data 2 Time series data +/-: no clear decision

From Figure 4 we get similar information as we get from the theoretical properties of the models. Throughout, we expect the QUAIDS to be the best suitable model. The main reason is its flexibility concerning non-linear Engel effects. This is also an advantage in comparison with the Rotterdam model (Figure 5). The AIDS, especially the QUAIDS, is better rated overall than the Rotterdam model. Even though the AIDADS has the same flexibility as the QUAIDS, we expect the QUAIDS to be superior for estimations of elasticities for disaggregated product groups.

General introduction

Figure 5: Decision matrix, part 2

AIDADS Rotterdam

AIDS • Barnett and Seck, 2008  +/- (simulation) • Fousekis and Revell., 2003  AIDS (in diff.) 1 • Lee et al., 1994  AIDS (in diff.) 1 LA/AIDS • Barnett and Seck, 2008  +/- (simulation) • Kastens und Brester, 1996  forecasting: Rotterdam, otherwise: +/- 2 • Alstom and Chalfant, 1993  Rotterdam 2 QUAIDS • Cranfield et al., 2003 • Fousekis and Revell, 2003  QUAIDS  QUAIDS 1 (especially for a disaggregated analysis) 1

Legend: AIDS: Almost Ideal Demand System LA/AIDS: Linear Approximated Almost Ideal Demand System QUAIDS: Quadratic Almost Ideal Demand System AIDADS: An Implicit Direct Additive Demand System

1 Household data 2 Time series data +/-: no clear decision

5.2 The QUAIDS

The starting point of the AIDS is a concave cost function (Deaton & Muellbauer, 1980a; 1980b). By deriving the cost function, we obtain the Hicksian demand function. Due to the dependence of the Hicksian demand function upon the utility level, which could not be observed directly, we needed to derive the Marshallian demand function. By inverting the cost function, we obtained the indirect utility function. Applying Roy‘s identity results in the Marshallian demand function:

( ) ( ( ) ( )) (20)

( ) ( ) ( ) (21)

Chapter I

where ( ) and ( ) are the cost function and the logarithmic cost function, respectively, and ( ) and ( ) are functions of prices. Their functional forms are:

( ) ∑ ∑ ∑ (22)

( ) ∏ (23) where n is the number of goods and , and are parameters to estimate. The whole cost function is:

( ) ∑ ∑ ∑ ∏ (24)

If we apply the steps we mentioned above, we obtain the Marshallian demand function:

∑ ( ) (25)

( ) ∑ ∑ ∑ (26)

where is the budget share for good , is the price of goods, is the income/total expenditure, and is the error term. The budget share is calculated by dividing the expenditure for good by the total expenditure.

The adding-up, homogeneity, and symmetry restrictions are expressed as:

∑ ; ∑ ; ∑ (27)

∑ (28)

(29)

The fourth restriction, negativity, which guarantees a concave cost function, could not

General introduction

be implemented with a parameter restriction, but could be tested after the model was estimated.

Determining the effects of non-linear Engel curves is difficult using the AIDS. Banks et al. (1997) added a quadratic term in the log income to the model and generated the QUAIDS, in which the AIDS is nested. It has the same degree of price flexibility as the AIDS. The demands generated by this class are shown to be of rank three (Lewbel, 1991; Gorman, 1981). The main advantage of the QUAIDS is that it produces a large regular region in comparison to the AIDS (see e.g. Cooper & Mclaren, 1996).

The model specification is as follows:

∑ ( ) * ( )+ (30) ( ) with a(p) and b(p) defined as above. The introduced parameter allows for the characterization of luxuries at low total expenditures and necessities at high expenditures. The restrictions are similar to the AIDS, but the adding-up restriction must be expanded with ∑ . The influence of socio-demographic variables, such as household size and education, is considered by the intercept ∑ , where is the jth variable of the total number of variables (demographic translating approach proposed by Pollak & Wales,1981, and used in Abdulai, 2002, and Liao & Chern, 2007).

The uncompensated elasticities could be calculated with the estimated parameters, and the compensated elasticities could be calculated by applying the Slutsky equation.

6 Structure of the thesis

The thesis is structured as follows (Figure 6). In Chapter 2, the first study is presented with focus on the approach to generate quality adjusted prices on the example of sheep and goat meat applying a tobit model. The approach to generate quality

Chapter I

adjusted prices is fundamental for the further studies. The subsequent chapter presents the second study with the estimation of meat and milk demand elasticities applying the QUAIDS. In Chapter 4 price and income elasticity for beverages for low, moderate and heavy drinking households were estimated and discussed. Chapter 5 presents the last study which is addressed to the question of endogeneity in demand models. This is a further methodological development of the approach used in Chapters 3 and 4. Elasticities for bread and cereal products are calculated applying different model specifications. The last chapter concludes.

Figure 6: Structure of the thesis

Chapter 1: Introduction and model evaluation

Chapter 2: Demand for sheep and goat meat, quality adjusted prices

Chapter 5: Demand Chapter 3: Demand Chapter 4: Demand for bread and cereal for meat and milk for (alcoholic) products, products beverages endogeneity

Chapter 6: Conclusions

General introduction

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Akbay, C., Boz, I., & Chern, W. S. (2007). Household food consumption in Turkey. European Review of Agricultural Economics, 34(2), 209-231.

Alston, J. M., & Chalfant, J. A. (1993). The silence of the Lambdas: A test of the Almost Ideal and Rotterdam models. American Journal of Agricultural Economics, 75(2), 304-313.

Anker, R. (2011). Engel‘s law around the world 150 years later. Working paper series, Political economy research institute, University of Massachusetts.

Aepli, M. (2012). Estimation of elasticites for food in Switzerland. Methodological report for the Swiss Federal Office for Agriculture. Unpublished study, ETH Zurich, Zurich.

Banks, J., Blundell R., & Lewbel A. (1997). Quadratic Engel curves and consumer demand. Review of Economics and Statistics, 79(4), 527–539.

Barnett, W. A., & Seck O. (2008). Rotterdam model versus Almost Ideal Demand System: Will the best specification please stand up? Journal of Applied Econometrics, 23(6), 795-824.

Barten A. P. (1964). Consumer demand functions under conditions of almost additive preferences. Econometrica, 32(1/2), 1–38.

Blackorby, Ch., Boyce, R., & Russell, R. R. (1978). Estimation of demand systems generated by the Gorman Polar Form: A generalization of the S-Branch Utility Tree. Econometrica, 46(2), 345-363.

Bouamra-Mechemache, Z., Réquillart, V., Soregaroli, C., & Trévisiol, A. (2008) Demand for dairy products in the EU. Food Policy, 33(6), 644-656.

Brehe, M. (2007). Ein Nachfragesystem für dynamische Mikrosimulationsmodelle. Dissertation, University of Potsdam.

Buse, A. (1994). Evaluating the Linearized Almost Ideal Demand System. American Journal of Agricultural Economics, 76(4), 781-793. 27

Chapter I

Christensen, L. R., Jorgenson, D. W., & Lau, L. J. (1975). Transcendental logarithmic utility functions. American Economic Review, 65(3), 367–383.

Deaton, A., & Muellbauer, J. (1980a). An Almost Ideal Demand System. American Economic Review, 70(3), 312–326.

Deaton, A., & Muellbauer, J. (1980b). Economics and Consumer Behavior. Cambridge University Press, Cambridge, UK.

De Boer, P., & Paap, R. (2009). Testing non-nested demand relations: Linear Expenditure System versus Indirect Addilog. Statistica Neerlandica, 6(3), 368-384.

Destatis (2013). German Federal Statistical Office, Wiesbaden. https://www.destatis.de/DE/ZahlenFakten/LaenderRegionen/Internationales/Thema/Ta bellen/Basistabelle_KonsumN.html. Accessed 5 Feb, 2014.

Cooper, R. J., & Mclaren, K. R. (1996). A system of demand equations satisfying effectively global regularity conditions. The Review of Economics and Statistics, 78(2), 359-364.

Cranfield, J. A. L., Eales, J. S., Hertel, T. W., & Preckel, P. V. (2003). Model selection when estimating and predicting consumer demands using international, cross section data. Empirical Economics, 28(2), 353-364.

Fousekis, P., & Revell, B. (2003). Quadratic differential demand systems and the retail demand for pork in Great Britain. Journal of Agricultural Economics, 54(4), 417-430.

Gallet, C.A. (2010). The income elasticity of meat: A meta-analysis. The Australian Journal of Agricultural and Resource Economics, 54(4), 477-490.

Gallet, C.A. (2009). Meat meets meta: A quantitative review of the price elasticity of meat. American Journal of Agricultural Economics, 92(1), 258-272.

Gallet, C.A. (2007). The Demand for Alcohol: A Meta-analysis of elasticities. Australian Journal of Agricultural and Resource Economics, 51(2), 121-135.

Gorman, W. M, (1981). Some Engel curves. in The Theory and Measurement of Consumer Behaviour, A. Deaton (ed.), Cambridge University Press, Cambridge, UK.

Gorman, W. M. (1959). Separability utility and aggregation. Econometria, 27(3), 469- 481.

General introduction

Holt, M. T., & Goodwin, B. K. 2009. The Almost Ideal and Translog Demand Systems, MPRA Paper No. 15092, Munich.

Jaquet. P., Abdulai, A., & Rieder, P. (2000). Empirische Analyse des Nahrungsmittelverbrauchs in der Schweiz: Ein dreistufiges LA/AIDS Modell. Scientific report, ETH Zurich, Zurich.

Kastens, T. L., & Brester, G. W. (1996). Model selection and forecasting ability of theory-constrained food demand systems. European Review of Agricultural Economics, 30(4), 539-558.

Katchova, A. L., & Chern, W. S. (2004). Comparison of Quadratic Expenditure System and Almost Ideal Demand System based on empirical data. The International Journal of Applied Economics, 1(1), 55-64.

Lee, J-Y., Brown, M. G., & Seale, J. L. (1994). Model choice in consumer analysis: Taiwan, 1970-89. American Journal of Agricultural Economics, 76(3), 504-512.

Lewbel, A. (1991). The rank of demand systems: Theory and nonparametric estimation. Economcetrica, 59(3), 711-730.

Lewbel, A. (1989). Nesting the AIDS and Translog Demand System. International Economic Review, 30(2), 349-356.

Liao, H., & Chern, W. S. (2007). A dynamic analysis of food demand patterns in China. Selected Paper at the AAEA Annual Meeting in Portland.

Pollak, R. A., & Wales, T. J. (1981). Demographic variables in the demand analysis. Econometria, 49(6), 1533-1551.

Pollak, R. A., & Wales, T. J. (1980). Comparison of the Quadratic Expenditure System and Translog Demand Systems with alternative specifications of demographic effects. Econometrica, 48(3), 595-612.

Pollak, R. A., & Wales, T. J. (1978). Estimation of complete demand systems from household budget data: The Linear and Quadratic Expenditure Systems. American Economic Review, 68(3), 348-359.

Schluep Campo, I. (2004). Market Options in the WTO Doha Round impacts on the Swiss meat market. Disseration, ETH Zurich.

Chapter I

Seale, J. L., & Regmi, A. (2006). Modeling international consumption patterns. Review of Income and Wealth. Series 52. No. 4.

Seel, B. (1991). Ökonomie des privaten Haushalts. Verlag Eugen Ulmer, Stuttgart.

SFSO. (2014). Consumer Price Index. Swiss Federal Statistical Office, Neuchâtel.

SFSO (2011). Haushaltsbudgeterhebung – Streckbrief. Swiss Federal Statistical Office, Neuchâtel.

Theil H. (1965). The Information Approach to demand analysis. Econometrica, 33(1), 67–87.

Thiele, S. (2008). Elastizitäten der Nachfrage privater Haushalte nach Nahrungsmitteln – Schätzung eines AIDS auf Basis der Einkommens- und Verbrauchsstichprobe 2003. Agrarwirtschaft, 57(5), 258-268.

Van der Meensbrugghe, D. (2005). Linkage technical reference document, version 6.0, The World Bank, Washington DC.

Yen, S.T., Lin, B-H., & Smallwood, D.M. (2003). Quasi- and simulated-likelihood approaches to censored demand systems: Food consumption by food stamp recipients in the United States. American Journal of Agricultural Economics, 85(2), 458-478.

General introduction

Chapter II

Chapter II Determinants of sheep and goat meat consumption in Switzerland

Matteo Aepli1 and Robert Finger2

1ETH Zurich 2Wageningen University

Manuscript published as: Aepli, M. & Finger, R. (2013) Determinants of sheep and goat meat consumption in Switzerland. Agricultural and Food Economics, doi:10.1186/2193-7532-1-11

Sheep and goat meat consumption

Abstract

In this study, we estimated the influence of different meat prices, socio-demographic and geographic variables on sheep and goat meat demand using the Swiss household expenditure survey from 2000 to 2005, a micro data set on 20940 households resident in Switzerland. This study is motivated by the fact that sheep and goats play a major economic role especially for small farms in Swiss agriculture and contribute to conservation of landscape and biodiversity especially in the mountain regions. Considering the high share of households with zero consumption for sheep and goat meat, we estimated a Tobit-model. Missing prices are replaced by monthly, regional market prices where possible, or by quality adjusted prices based on unit values. Our results show that pork is a substitute for sheep and goat meat, whereas beef is a complementary good, and that socio-demographic variables like education or the presence of children and geographic variables are important in determining demand of sheep and goat meat.

Chapter II

1 Introduction

In Switzerland, there are currently almost 418,000 sheep and 82,000 goats. This production activity has a high importance especially for small-scale farms and part- time farming in Swiss agriculture (Aepli & Jörin, 2011). About 20% of all Swiss farmers keep sheep or goats (SFSO, 2012). The feeding system is mainly grassland based. Almost half of the sheeps are grazing on summer pasture with increasing tendency during the last years (Kaulfers, 2009). The production of these small ruminants contributes substantially to conservation of landscapes and diversity associated with Swiss and European agriculture (Dýrmundsson, 2006). For instance, grazing of summer pastures prevents the natural regrowth of shrubs and forests (Herzog et al., 2009). Preventing this forest regrowth is of high societal and political interest because this is associated with negative impacts on biodiversity and the valuation of mountain landscapes by people (e.g. Gellrich et al., 2007; Hunziker et al., 2008). Recent investigation show that this regrowth occurs particularly on steeper slopes and at higher elevations where cultivation costs are high and yield potential is low (Pellissier et al., 2012). These are, however, exactly the plots that are either only suitable for small ruminant grazing or where this land use implies the smallest costs of cultivating these areas (e.g. Dux et al., 2009). Thus, even though the consumption levels of sheep and goat meat are currently small (1.3 kg per capita for sheep and goat meat, Proviande, 2012), small ruminants play a decisive role in Swiss agriculture. This role is determined by consumers‘ decisions to buy sheep and goat meat.3 Despite this relevance, there is no study that has addressed this question yet.4 Switzerland as a high developed country is a particularly interesting case due to high production costs of farms for livestock and for crops on the one side (e.g. Lehmann et al., 2005; Lips et

3 Compared to other types of meat, sheep and goat meat have both a pronounced flavor. In comparison to sheep meat, goat meat tends to be darker red with a coarser texture (Webb et al., 2005; Gaili & Aili, 1985). 4 Along these lines, there is in general little research available identifying determinants of sheep and goat consumption (see e.g. Juma et al., 2010 or Bernabéu et al., 2012 for exceptions).

Sheep and goat meat consumption

al., 2007; Raaflaub et al., 2005) and high purchase power of the consumers on the other side.

Against this background, the goal of this article is to provide determinants of sheep and goat meat consumption in Switzerland. Sheep and goat farming are under great pressure in Switzerland as well as in whole Europe because of a decreasing demand, high imports from Oceania and a price drop in some countries in the recent past (Boutonnet, 1999; Proviande, 2013; Aepli & Jörin, 2012).

An important reason for this consumption decline is the substitution with other types of meat like poultry and pork as well as fish (Dýrmundsson, 2006; Jaquet et al., 2000). Such substitution was also observed in Switzerland (Aepli & Jörin, 2012), where the consumer price for sheep meat increased by 30% since 2000, whereas the consumer price for poultry and fish has remained constant (SFSO, 2013).

During the next years international competition will further increase and at the same time sheep and goat meat consumption may continue to decline (Dýrmundsson, 2006; EU Commission, 2012). Therefore the understanding of the determinants of demand is crucial for the European sheep and goat meat production. The analysis of the Swiss demand will give some useful information for the whole European market. Following recent studies (Abdulai, 2002; Gallet, 2009; Jaquet et al., 2000), we focused our research on the following determinants affecting sheep and goat meat consumption: different meat prices (sheep and goat, beef, pork, chicken), fish price, total expenditure and expenditure for food, as well as household‘s characteristics like the presence of children and education. In addition, we are also able to account for within- country cultural differences determined by language regions in Switzerland. Estimating the relevance of these determinants will contribute to a better understanding of the driving forces of demand and will provide an empirical base for marketing activities.

We use data from the Swiss household expenditure survey covering monthly data for the period 2000-2005 with a total sample size of 20‘940. Even though the use of

Chapter II

detailed micro-data (i.e. highly disaggregated data with respect to time, product groups and households) is superior to other approaches (e.g. Akbay et al., 2007; Lazaridis, 2003; Majumder et al., 2012; Thiele, 2008), we face two major methodological challenges if analyzing this household expenditure survey data for sheep and goat meat. First, many households do not purchase this kind of meat every month leading to a large share of zero observations. Second, no direct price data is provided in our data set. To overcome these problems we combine recently proposed approaches to a) avoid section biases associated with zero consumption using Tobit regressions (e.g. Ma et al., 2006; Yimer, 2011), and b) we derive quality adjusted price levels at the household level using the approach proposed by Majumder et al. (2012) or regional market prices where available.

7 Material and Methods

7.1 Methodology

Zero-consumption provides important behavioral information and has significant economic and econometric implications. The behavior of zero-consumption facing the budget restrictions and the prices is a typical corner solution of the household‘s utility maximization problem (Perali & Chavas, 2000). This constitutes an econometric problem if food items are characterized by frequent zero-consumption, as it is the case for sheep and goat meat in Switzerland. As Deaton (1990) mentioned, a deletion of the zero consumption sample points could introduce bias and only allows to estimate conditional effects. To avoid selection bias we estimate a Tobit model (Tobin, 1958; Amemiya, 1984; 1973; Solon, 2010) for censored regression which is widely used in single-equation models (e.g. Ma et al., 2006; Yimer, 2011) and is of the following form:

( ) and (1)

Where is a matrix consisting of explanatory variables for the th of n observations plus a vector of ones for the intercept term and is a vector of coefficients to be

Sheep and goat meat consumption

estimated with dimensions ( ) . ‘s are independently, identically and normally distributed random variables with zero mean and variance , and represents the latent variable. The relationship between and is defined as:

{ (2)

The Likelihood function for the observed sample of ‘s is:

( ) ( ) [ ( )] ∏ ( ) ( ) (3)

with for limited observations and otherwise. is the standard normal density function and the cumulative counterpart. The parameters and can be estimated consistently using maximum likelihood. However, the maximum likelihood (ML) estimator is inconsistent and its use problematic if the assumptions of normality and homoscedasticity of the errors in the Tobit Model are violated (Amemiya, 1984; Arabmazar & Schmidt, 1981; Caudill & Mixon, 2009; Li et al., 2007). To account for this high sensitivity of the ML estimator in the Tobit model with respect to the underlying assumptions, we corrected the standard errors with the Huber/White estimator (Huber, 1967; White, 1982). Because the estimated coefficients of the Tobit- Model refer to the latent variable, marginal effects are additionally calculated which represent the effects of the explanatory variables on the expected value of the dependent variable evaluated at the mean values of the explanatory variables.

For beef, pork, veal and sheep and goat meat the Swiss Federal Office for Agriculture calculate monthly average consumer prices by regions which we take as market prices. For the other items (chicken and fish) missing information of price data were replaced using the approach presented in Majumder et al. (2012) who extended the approach of Cox and Wohlgenant (1986). They proposed to adjust unit values for the impact of quality and use them as market prices (see also e.g. Park et al., 1996; Thiele, 2010). Unit values are obtained by dividing recorded expenditures for every household and food item by the corresponding quantity for every household. In 37

Chapter II

contrast to the approach of Cox and Wohlgenant (1986), Majumder et al. (2012) and Hoang (2009) used regionally aggregated quality adjusted unit values as market prices based on the quality adjusted unit values for every household which is consistent with the hypothesis that households are facing the same price in the same market. We enhanced the approach by additionally aggregating over month and years.

The quality effects are mainly induced by income or total expenditure and household characteristics and can be determined as the difference between the unit value paid by the household and the regional average unit value. Therefore unit values are related through the following equation:

( )

∑ (4)

where is the unit value paid by the household for item in its region , year , and month m. For income, we include both income ( ) and the square of income ( ) to allow for a non-linear relationship between income and the unit value. is the household food expenditure, is the household expenditure for food consumed outside (meals and beverages) and denotes the th of household characteristics. This includes the household size by adult equivalent, which is generated after the OECD-modified equivalence scale (Hagenaars et al., 1994). Furthermore, binary dummy variables for having children and for having a university degree are included. , and are dummies for region, year and month. We follow Majumder et al. (2012) and estimate equation (4) by using medians instead of means which is a more robust statistic with respect to outliers. Furthermore, we applied a robust M- estimator instead of OLS which limits the influence of outliers (see e.g. Finger, 2013 for details).

After estimating (4) for each item with a sub-sample of the households which consumed of the item, the region-, monthly-, and yearly-wise quality adjusted prices are generated by adding

Sheep and goat meat consumption

the median unit value for this item to the median of the estimated residuals from equation (4).

( ) ( ) ( ) (5)

Thus, the calculated market prices assigned to each household account for the households region as well as month and year.

7.2 Data and descriptive statistics

The data used in this study are taken from the Swiss household expenditure survey for the period 2000-2005 that was conducted by the Swiss Federal Statistic Office. The Swiss household expenditure survey consists of repeated cross-sectional data with a periodicity of one month. Approximately 3,000 households participate every year. They are chosen at random from the register of private telephone lines. The sample is stratified by the seven major regions of Switzerland. The distribution of individuals is matched by a calibration method to the known distribution of the population and the sample is therefore representative for the Swiss population (see SFSO, 2011 for details). In addition to expenditures, quantities bought, and income, the dataset also includes several household characteristic variables (e.g. household size, age and sex of the head of household).

The explanatory variables included in our model are prices for sheep and goat meat, for pork, beef, chicken and fish, total household expenditure and household expenditure for food5, household characteristics like the presence of children (<=18 years) and education (having a university degree or not), and dummies for the regions of Switzerland.6 The total sample availbale to our study consists of 20‘940 households where 81.75% of the households did not consume sheep and goat meat during the

5 Due to relatively low correlation between total expenditure and expenditure for food we included both variables in the model. 6 We first estimated the model with the factor variable month to capture seasonality. Since we calculate monthly prices for the meat categories there has been some multicollinearity between the variable month and the prices. Seasonality is partly captured by the price variables. We therefore decided to estimate the model without the variable month. 39

Chapter II

recording period. Table 1 presents the 1% trimmed mean values for the expenditure variables and the region dummies including the quantity purchased, which is used to calculate the unit values in the first stage of quality adjusted prices.7

Table 1: Characteristics of the sample and descriptive statistics of sheep and goat meat consumption and household expenditure

Total German-speaking French-speaking part Italian-speaking

Switzerland part of Switzerland of Switzerland part of Switzerland

Number of households 20940 14637 4497 1806

Mean in the Mean in the French- Mean in the Italian-

Mean in total German-speaking speaking part of speaking part of

Switzerland part of Switzerland Switzerland (1% Switzerland (1%

(1% trimmed) (1% trimmed) trimmed) trimmed)

Sheep and goat meat consumed per

household and month (in Swiss Francs) 3.74 3.05 6.48 2.81

Sheep and goat meat consumed per

household and month (in kg) 0.12 0.09 0.23 0.11

Household monthly total expenditure in CHF 7811.18 7892.46 7869.05 7035.58

Household monthly expenditure for food in

CHF 601.29 588.38 641.66 606.03

Household size in adult equivalents 1.63 1.63 1.63 1.66

Number of children (<=18 years) 0.64 0.63 0.67 0.62

Age of the household reference person 4.90 4.90 48.50 50.29

Highest education of the household reference

person* 2.64 2.65 2.65 2.49

*The education level is recorded with a scale from 1 to 3, whereas 1 is compulsory school, 2 is professional education and 3 is higher professional education like a university degree. See SFSO (2011), for details on the data collection and the representativeness of the sample.

7 This method removes 1% of the largest and the smallest values before calculating the mean. We decided to use the trimmed mean instead of the normal mean, because we detect some outliers in the sample and this method is more robust with respect to possible outliers. 40

Sheep and goat meat consumption

Table 2 presents the results of pairwise t-tests to determine whether differences across regions with respect to expenditures (total and food) and sheep and goat consumption (consumption and expenditures) are significant.

Table 2: Pairwise comparison of the regional means for consumption of sheep and goat meat, total household expenditure and household expenditure for food using t- test

Sheep and goat meat French-speaking part of Italian-speaking part of consumed per household Switzerland Switzerland and month (in Swiss Francs) German-speaking part of -15.55*** 0.95 Switzerland French-speaking part of --- -11.60*** Switzerland

Sheep and goat meat consumed per household and month (in kg) German-speaking part of -17.30*** -2.02* Switzerland French-speaking part of --- -9.86*** Switzerland

Household monthly total expenditure in CHF German-speaking part of 0.40 9.94*** Switzerland French-speaking part of --- -8.43*** Switzerland

Household monthly expenditure for food in CHF German-speaking part of -9.70*** -2.31* Switzerland French-speaking part of --- -4.09*** Switzerland T-values are presented in the table. * and *** denote significance at 5% and 0.1% level, respectively.

The household data show that monthly expenditure in the French-speaking part of Switzerland for sheep and goat meat is with 6.48 Swiss francs significantly higher than in the German-speaking part of Switzerland, showing an average consumption of 3.05 Swiss francs per month (Tables 1 and 2). Households in the Italian speaking part of Switzerland spend significantly more on food in general but do not spend more on sheep and goat meat than households in the German speaking part of Switzerland.

Chapter II

8 Results and Discussion

Estimations of the Tobit-model are presented in Table 3. It shows that, as expected, rising prices for sheep and goat meat lead to a decline in its demand. Furthermore, we find sheep and goat meat to be substitutes to other meat. More specifically, an additional unit on price of pork would increase the household demand for sheep and goat meat by 0.38 units. In contrast, the negative sign for beef prices indicates that households facing a one unit lower beef price spend 0.20 units more on sheep and goat meat than households facing the higher beef price. In contrast, no significant influence of fish and chicken price levels on sheep and goat meat consumption has been found.

Table 3: Parameter estimates of the Tobit-model for the influence of the explanatory variables on the household expenditure for sheep and goat meat

Robust standard

Independent variables Coefficients Marginal effect error

Sheep and goat meat price per kg in CHF -1.362* -0.214* 0.598

Pork price per kg in CHF 2.423*** 0.380*** 0.523

Beef price per kg in CHF -1.275** -0.200** 0.410

Chicken price per kg in CHF -0.153 -0.024 0.266

Fish price per kg in CHF -0.244 -0.038 0.190

Household monthly total expenditure in CHF 0.001*** 0.000*** 0.000

Household monthly expenditure for food in CHF 0.055*** 0.009*** 0.005

Presence of children (yes=1, otherwise=0) -15.116*** -2.372*** 1.939

University degree (university degree=1, otherwise=0) 9.902*** 1.554*** 1.618

Household in the French-speaking region of Switzerland (yes=1,

otherwise=0) 20.187*** 3.168*** 1.968

Household in the Italian-speaking region of Switzerland (yes=1,

otherwise=0) -5.314* -0.834* 2.505

*, ** and *** denote significance at 5%, 1% and 0.1% level, respectively.

Sheep and goat meat consumption

The finding that pork meat is a good substitute for sheep and goat meat is consistent with the conclusions in Jaquet et al. (2000) who estimated a linearized Almost Ideal Demand System to calculate price and income elasticities based on household expenditure data for Switzerland of 1998. In contrast to pork, the model estimates suggest that beef is a complementary good. These findings are in agreement with Tiffin et al. (2011), who estimated price elasticities for UK. We included also the price for fish and chicken in the model due to previous research of Jaquet et al. (2000) who reported a relatively high substitution elasticity for Switzerland between sheep and goat meat and fish and chicken meat, respectively. In contrast to the previous findings we were not able to show that the prices of these two meat categories have a significant effect on the household expenditure for sheep and goat meat. Schluep Campo (2004) analyzed Swiss meat demand at the wholesale level using time series data for the time period January 1996 to December 2002 with a semiflexible Almost Ideal Demand System for sheep, beef, pork and poultry. With respect to sheep meat, she reported positive compensated price elasticities for poultry and pork, and small but negative compensated price elasticities for beef meat. This thus supports our findings that pork is a substitute for sheep meat while beef tends to be a complementary good.

The influence of the expenditure (total, food and meat) is highly significant. We find households with higher total expenditure or higher expenditure for food, to spend more on sheep and goat meat. This implies positive income elasticity for sheep and goat meat.

Furthermore, socio-demographic variables are found to be significant in affecting consumer demand for sheep and goat meat. This is consistent with other research findings (Dhraief et al., 2012; Gao et al., 1994; Lazaridis, 2003). For the presence of children, the sign is negative, indicating that households with children consume a lower amount of sheep and goat meat.8 The marginal rate indicates a decrease in

8 We started with a model also including household size. Due to some correlation of the variables especially between household size and the dummy for children, we tested the model in several 43

Chapter II

sheep and goat meat expenditure by -2.37 CHF. Children have less strong preferences for sheep and goat meat. This is consistent with the results of a study for sheep meat consumption in Spain (Sepúlveda et al., 2011) – which probably also holds for goat meat – stating that young people consume sheep meat less frequently than older people, which may be caused by the influence of stereotypes, unfamiliarity and the strong flavor of sheep and goat meat (Karakuş, 2006).

In contrast, education has been found to have a positive influence on sheep and goat meat consumption. If the head of the household has a university degree, the household spends a 1.6 CHF more on sheep and goat meat. The revealed influence of education on demand is in expectation with the findings of previous studies on meat consumption. Educational attainment is a main determinant in food consumption choices and is positively correlated with healthy dietary intake (Moreira & Pardão, 2004). Sheep and goat meat has several health benefits in comparison to other red meat, especially if the animals are reared under extensive conditions on natural pastures like in Switzerland (zur Hausen, 2012; Polidori et al., 2011), which also potentially coincides with advantages with respect to environmental impacts and animal welfare (Stott et al., 2005). As education is positively correlated with the awareness of these aspects associated with food consumption (Daniel et al., 2011; Monaco Bissonnette & Contento, 2001), households with a better educated household head tend to have a higher demand for sheep and goat meat.

Regional differences in sheep and goat meat consumption pattern in Switzerland are significant, especially for the French-speaking part of Switzerland in comparison to the German-part of Switzerland. Accounting for all the influences of the other variables in our regression, the marginal rates are 3.17 and -0.83 for the French-speaking and Italian-speaking part, respectively. This confirms earlier research showing that sheep and goat meat shows high level of spatial heterogeneity (e.g. in Aepli, 2013), which is expected to be caused by different consumer attitudes towards sheep and goat meat specifications. In every specification the coefficients have the same sign, almost the same magnitude and significance. The qualitative interpretation remains the same. 44

Sheep and goat meat consumption

(Sañudo et al., 2007) based on different cultural and traditional backgrounds (Karakuş, 2006).

9 Conclusions

Our results suggest that prices for different meat categories, expenditure as well as different household characteristics and regional aspects are predominant factors determining the demand for sheep and goat meat in Switzerland. The coefficients in our Tobit model are all, except of the price for chicken and fish, statistically significant. Our results confirm earlier research that sheep and goat meat is a substitute to chicken and pork, but is complementary to beef. Our results allow some insights how to cope with future challenges for sheep and goat production in Switzerland. For instance, the positive sign for the French-speaking part of Switzerland implies that marketing activities should be especially focused on this area, where sheep and goat meat consumption is traditionally higher and well accepted by the consumers (LID, 2000). One of the major challenges for Swiss sheep and goat producer will be price reductions resulting from trade liberalization between Switzerland and the European Union on the one hand and trade liberalization as a result of WTO negotiation on the other hand. Switzerland still has a highly restrictive border protection for agricultural products. This is particularly the case for meat, where the border protection consists of tariff-rate quotas. In the case of sheep and goat meat the quota is always exhausted. Above-quota imports are currently scarce, because the over-quota tariff is so high that it has normally a prohibitive effect on trade (Jörin & Lengwiler, 2004). Thus, a relaxation of the highly restrictive tariff-rate quota along with liberalization steps, would probably lead imports to rise quickly (Schluep Campo 2004; Aepli & Jörin, 2011). But, earlier research has also shown that Swiss consumers are willing to pay higher prices for high-quality and domestically produced products (Bolliger, 2012). Thus, the here indicated effects may offer a high-quality strategy as a potential path for Swiss sheep and goat producers. Our results suggest, for instance, that an improved communication of potential health and environmental benefits may help to cope with 45

Chapter II

future challenges. But, a more detailed understanding of the determinants of demand is a core issue to intensify and tailor marketing activities and prevent Swiss as well as other European producers to suffer from a sharp price decline after a reduction of trade barriers. In addition, higher prices will encourage the production of small ruminant which could have direct positive effects on conservation of landscapes and biodiversity especially for summer pastures, which should be addressed in further research.

Sheep and goat meat consumption

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Aepli, M. (2013). Schätzen von Engelkurven für verschiedene Fleischkategorien. Diploma thesis, ETH Zurich, Zurich.

Aepli, M., & Jörin, R. (2011). Der Schweizer Lammfleischmarkt: Marktanalyse und Wettbewerb, ETH Zurich, Zurich.http://www.afee.ethz.ch/people/Staff/maepli/Publications/Der_Schweizer_Lam mfleischmarkt_Marktanalyse_und_Wettbewerb. Accessed 10 Dec, 2012.

Akbay, C. Boz, I., & Chern, W.S. (2007). Household food consumption in Turkey. European Review of Agricultural Economics, 34(2), 209-231.

Amemiya, T. (1984). Tobit models: A survey. Journal of Econometrics, 24(1-2), 3-61.

Amemiya, T. (1973). Regression analysis when the dependent variable is truncated normal. Econometria, 41(6), 997-1016.

Arabmazar, A., & Schmidt, P. (1981). Further evidence on the robustness of the Tobit estimator to heteroscedasticity. Journal of Econometrics, 17(2), 253-258.

Bernabéu, R. Tendero, A., & Olmeda, M. (2012). Consumer preferences versus commercial differentiation: a Spanish case study. British Food Journal, 114(11), 1626- 1639.

Bolliger, C. (2012). Ökonomische Analyse von Herkunftsangaben bei Agrarerzeugnissen - Präferenz und Zahlungsbereitschaft für die "Herkunft Schweiz". Dissertation, ETH Zurich.

Boutonnet, J.P. (1999). Perspectives of the sheep meat world market on future production systems and trends. Small Ruminant Research, 34(3), 189-195.

Caudill, S.B., & Mixon, F.G. (2009). More on testing the normality assumption in the Tobit Model. Journal of Applied Statistics, 36(12), 1345-1352.

Cox, T.L., & Wohlgenant, M.K. (1986). Prices and Quality Effects in Cross-Sectional Demand Analysis. American Journal of Agricultural Economics, 68(4), 908-919.

Chapter II

Daniel, C.R., Cross, A.J., Koebnick, C., & Rashmi, S. (2011). Trends in meat consumption in the United States. Public Health Nutrition, 14(4), 575-583.

Deaton, A. (1990). Price elasticities from survey data: Extensions and Indonesian results. Journal of Econometrics, 44(3), 281-309.

Dhraief, M.Z., Oueslati, M., & Dhehibi, B. (2012). Meat and fish demand in Tunisia: economic and socio-demographic factors effects. Paper presented at the International Association of Agricultural Economists Conference 2012, Foz do Iguacu, 18-24 August 2012.

Dux, D., Matz, K., Gazzaring, C., & Lips, M. (2009). Was kostet offenes Grünland im Berggebiet? Agrarforschung, 16(1), 10-15.

Dýrmundsson, O.R. (2006). Sustainability of sheep and goat production in North European countries – From the Arctic to the Alps. Small Ruminant Research, 62(3), 151-157.

EU Commission (2012). Prospects for Agricultural Markets and Income in the EU 2012-2022, December 2012. European Commission, Brussels.

Finger, R. (2013). Investigating the performance of different estimation techniques for crop yield data analysis in crop insurance applications. Agricultural Economics, 44(1), 217-230.

Gaili, E.S., & Aili, A.E. (1985). Meat from Sudan desert sheep and goats: Part 2 – composition of the muscular and fatty tissue. Meat Science, 13(4), 229-236.

Gallet, C.A. (2009). Meat meets meta: A quantitative review of the price elasticity of meat. American Journal of Agricultural Economics, 92(1), 258-272.

Gao, X.M., & Spreen, T. (1994). A Microeconometric Analysis of the U.S. Meat Demand. Canadian Journal of Agricultural Economics, 42(3), 397-412.

Gellrich, M., Baur, P., Koch, B., & Zimmermann, N.E. (2007). Agricultural land abandonment and natural forest re-growth in the Swiss mountains: a spatially explicit economic analysis. Agriculture, Ecosystems and Environment, 118(1/4), 93-108.

Hagenaars, A.J.M., De Vos, K., & Zaidi, M.A. (1994). Poverty Statistics in the Late 1980s: Research Based on Micro-data. Kontoret for De Europæiske Fællesskabers Officielle Publikationer, Luxembourg.

Sheep and goat meat consumption

Herzog, F., Böni, R., Lauber, S., Schneider, M., & Seidl, I. (2009). AlpFUTUR – an inter- and transdisciplinary research program on the future of summer pastures in Switzerland. Paper presented at the 15th meeting of the FAO-CIHEAM Mountain Pastures Network, Agroscope Changins-Wadenswil Research Station ACW, Les Diablerets, 7-9 October 2009.

Hoang, L.V. (2009). Estimation of food demand from household survey data in Vietnam. Working Paper. Development and Policies Research Center (DEPOCEN), no. 2009/12.

Huber, P.J. (1967). The behavior of maximum likelihood estimated under nonstandard conditions. Paper presented at the Fifth Berkeley Symposium on Mathematical Statistics and Probability, University of California Press, Berkeley, 1967.

Hunziker, M., Felber, P., Gehring, K., Buchecker, M., Bauer, N., & Kienast, F. (2008). Evaluation of Landscape Change by Different Social Groups. Mountain Research and Development, 28(2), 140-147.

Jaquet, P., Abdulai, A., & Rieder, P. (2000). Empirische Analyse des Nahrungsmittelverbrauchs in der Schweiz: Ein dreistufiges LA/AIDS Modell. ETH Zurich, Zurich.

Jörin, R., & Lengwiler, Y. (2004). Learning form financial markets: Auctioning tariff-rate quotas in agricultural trade. Swiss Journal of Economics and Statistics, 140(4), 521- 541.

Juma, G.P., Ngigi, M., Baltenweck, I., & Drucker, A.G. (2010). Consumer demand for sheep and goat meat in Kenya. Small Ruminant Research, 90(1/3), 135-138.

Karakuş, F. (2006). Consumer preferences on sheep and goat meat in the world. Paper presented at the 57th Annual Meeting of the European Association for Animal Production, Antalya, Turkey, 2006.

Kaulfers, C. (2009). Weide- und Bewegungsverhalten von Schaf und Ziege auf der Alp und dessen Einfluss auf den Knochen- und Energiestoffwechsel. Dissertation, University of Zurich, Zurich.

Lazaridis, P. (2003). Household meat demand in Greece: A demand systems approach using microdata. Agribusiness, 19(1), 43-59.

Chapter II

Lehmann, B., Gerwig, C., & Huber, R. (2005). Wettbewerbsfähigkeit der Schweizerischen Schweine- und Geflügelproduktion. ETH Zurich, Zurich. http://www.aviforum.ch/downloads/Studie_%20ETH.pdf. Accessed 18 Dec, 2012.

Li, L., Simonoff, J.S., & Tsai, C.L. (2007). Tobit model estimation and sliced inverse regression. Statistical Modelling, 7(2), 107-123.

Lips, M., Ammann, H., Signer, A., & Steingruber, E. (2007). Kostenvergleich im Ackerbau zwischen der Schweiz und Baden-Württemberg, ART-Berichte, Forschungsanstalt Agroscope Reckenholz-Tänikon, Ettenhausen.

LID. (2000). Lammfleisch: Die gefragten Stücke kommen hauptsächlich aus Übersee. In: Mediendienst Nr. 2460. Landwirtschaftlicher Informationsdienst. http://www.lid.ch/de/medien/mediendienst/artikel/infoarticle/8989/. Accessed 17 Nov, 2012.

Ma, H., Huang, J., Fuller, F., & Rozelle, S. (2006). Getting rich and eating out: consumption of food away from home in urban China. Canadian Journal of Agricultural Economics, 54(1), 101-119.

Monaco Bissonnette, M., & Contento, I.R. (2001). Adolescents‘ perspectives and food choice behaviors in terms of the environmental impacts of food production practices: application of a psychosocial Model. Journal of Nutrition Education, 33(2), 72-82.

Majumder, A., Ray, R., & Sinha, K. (2012). Calculating rural-urban food price differentials from unit values in household expenditure surveys: A comparison with existing methods and a new procedure. American Journal of Agricultural Economics, 94(5), 1218-1235.

Moreira, P.A., & Padrão, P.D. (2004). Educational and economic determinants of food intake in Portuguese adults: a cross sectional survey. BMC Public Health, doi: 10.1186/1471-2458-4-58.

Park, L.J., Holcomb, R.B., Raper, K.C., & Capps, O. (1996). A demand system analysis of food commodities by U.S. households segmented by income. American Journal of Agricultural Economics, 78(2), 290-300.

Pellissier, L., Anzini, M., Maiorano, L., Dubuis, A., Pottier, J., Vittoz, P., & Guisan, A. (2012). Spatial predictions of land-use transitions and associated threats to

Sheep and goat meat consumption

biodiversity: the case of forest regrowth in mountain grasslands. Applied Vegetation Science, 16(2), 227-236.

Perali, F., & Chavas, J-P. (2000). Estimation of censored demand equations form large cross-section data. American Journal of Agricultural Economics, 82(4), 1022- 1037.

Polidori, P., Ortenzi, A., Vincenzetti, S., Beghelli, D. (2011). Dietary properties of lamb meat and human health. Mediterranean Journal of Nutrition and Metabolism, 4(1), 53- 56.

Proviande (2013). Der Fleischmarkt im Überblick 2012. Proviande, Bern.

Proviande (2012). Der Fleischmarkt im Überblick 2011. Proviande, Bern .

Raaflaub, M., Durgiai, B., & Genoni, M. (2005). Wettbewerbsfähigkeit der tierischen Veredelung in der Schweiz. School of Agricultural, Forest and Food Science, Zollikofen. http://www.bell.ch/Portaldata/1/Resources/6_Medien/Dossier_Wettbewerbsfaehigkeit/ dossied_wettbewerbsfaehigkeit_studie_shl.pdf. Accessed 18 Dec, 2012.

Sañudo, C., Alfonso, M., San Julián, R., Thorkelsson, G., Valdimarsdottir, T., Zygoyiannis, D., Stamataris, C., Piasentier, E., Mills, C., Berge, P., Dransfield, E., Nute, G.R., Enser, M., & Fisher, A.V. (2007). Regional variation in the hedonic evaluation of lamb meat from diverse production systems by consumers in six European countries. Meat Science, 75(4), 610-621.

Schluep Campo, I. (2004). Market Options in the WTO Doha Round impacts on the Swiss meat market. Disseration, ETH Zurich.

Sepúlveda, W.S., Maza, M.T., & Pardos, L. (2011). Aspects of quality related to the consumption and production of lamb meat. Consumer versus producers. Meat Science, 87(4), 366-372.

SFSO (2013). Landesindex der Konsumentenpreise, Swiss Federal Statistical Office, Neuchâtel.

SFSO (2012). Statistisches Lexikon der Schweiz, Landwirtschaftliche Betriebsstrukturerhebung. Swiss Federal Statistical Office, Neuchâtel. http://www.bfs.admin.ch/bfs/portal/en/index.html. Accessed 7 Dec 2012.

Chapter II

SFSO (2011). Haushaltsbudgeterhebung 2005, Swiss Federal Statistical Office, Neuchâtel.

Solon, G. (2010). A simple microeconomic foundation for a Tobit model of consumer demand. Economics Letters, 106(2), 131-132.

Stott, A.W., Milne, C.E., Goddard, P.J., & Waterhouse, A. (2005). Projected effect of alternative management strategies on profit and animal welfare in extensive sheep production systems in Great Britain. Livestock Production Science, 97(2), 161-171.

Thiele, S. (2010). Erhöhung der Mehrwertsteuer für Lebensmittel: Budget- und Wohlfahrtseffekte für Konsumenten. Journal of Economics and Statistics, 230(1), 115- 130.

Tiffin, R., Balcombe, K., Salois, M., & Kehlbacher, A. (2011). Estimating Food and Drink Elasticities. University of Reading, Reading. http://www.defra.gov.uk/statistics/files/defra-stats-foodfarm-food-price-elasticities- 120208.pdf. Accessed 17 Dec 2012.

Tobin, J. (1958). Estimation of relationships for limited dependent variables. Econometrica, 26(1), 24-36.

Webb, E.C., Casey, N.H., & Simela, L. (2005). Goat meat quality. Small Ruminant Research, 60 (1-2), 153-166.

White, H. (1982). Maximum likelihood estimation of misspecified models. Econometrica, 50(1). 1-25.

Yimer, S. (2011). Determinants of food consumption expenditure in Ethiopia. International Journal of Economics and Research, 2(5), 151-165. zur Hausen, H. (2012). Red meat consumption and cancer: Reasons to suspect involvement of bovine infectious factors in colorectal cancer. International Journal of Cancer, 130(11), 2475-2483.

Sheep and goat meat consumption

Chapter III

Chapter III Meat and milk demand elasticities for Switzerland: A three- stage budgeting Quadratic Almost Ideal Demand System

Matteo Aepli1 and Christian Kuhlgatz2

1ETH Zurich

2Thünen-Institute, Braunschweig, Germany

elasticities demand milk and Meat

Manuscript under review, Agribusiness 54

Meat and milk demand elasticities

Abstract

In this study, we estimated price and income elasticities for various meat and milk product categories using cross-sectional data from the Swiss household expenditure survey from 2004 to 2009 covering almost 20,000 households. The demand for meat and milk products plays a major role in public health and could have a negative environmental impact due to greenhouse gas emissions in production and trade. We applied a three-stage quadratic almost ideal demand system, taking into account the high shares of zero consumption and correcting for endogeneity of the expenditure variable. Missing prices were replaced by monthly, yearly, and regionally quality- adjusted prices. The results show that almost all product groups are necessary goods, most of the meat and milk product categories are substitutes, and own-price elasticities have been increasing over the last decade.

Chapter III

1 Introduction

Investigating determinants of consumption patterns is highly relevant for industry and policy analysis. This is particularly true of the demand for animal products given the global trend of increased meat consumption and its consequences for human health and the environment. While meat consumption can have positive effects on health due to its high content of protein, iron, and other micronutrients, high meat-based diets can result in obesity and increase the risk of heart disease, certain types of cancer, stroke, and diabetes (Walker et al., 2005). The negative environmental impact of meat and other ruminant by-products such as milk and cheese is mainly caused by livestock production due to greenhouse gas emissions, and, to a smaller extent, by the subsequent stages such as slaughtering or cheese manufacturing (e.g. waste water production or energy consumption) (Alig et al., 2012; Kim et al., 2013). Furthermore, there are contentious debates on the environmental and social impacts of feed production in poor regions of the world, particularly in soybean growing parts of Latin America. Against this backdrop, elasticities demand milk and Meat a tax on meat or fatty products is being considered in many high-income countries (e.g. the Swedish proposal for an EU-wide meat tax).

In contrast to a fat tax policy, trade liberalization agreements between Switzerland and the EU will have a negative effect on domestic prices of meat and milk products during the next five to ten years. This is because Switzerland still has a highly restrictive border protection for agricultural products and, therefore, high domestic prices for meat and some milk products in comparison to other European countries. Consumer responses to

price reductions are also relevant for suppliers in adjacent EU countries given that high price differentials for animal products tend to cause cross-border travel. For example, the introduction of a fat tax in Denmark resulted in significant cross-border travel to northern Germany due to high price differentials. This was one reason the Danish government decided to abolish the tax just one year after its implementation.

In order to assess the effect of a price reduction on demand, price elasticities are crucial. In addition, the response of meat consumption to changes in income has received 56

Meat and milk demand elasticities

particular attention in the literature (Gallet (2009; 2010) identified 419 and 393 studies that estimated the income and price elasticity of meat consumption). Information on income elasticities and price elasticities may allow policy makers to tailor market interventions to influence food consumption patterns. During the last years, a fat tax has been discussed several times in Switzerland. Certain food price policies towards magnitude and composition of meat consumption may be superfluous if changing income levels shift consumption in the desired patterns (Gallet, 2010). This issue is also relevant for other highly developed countries such as some member states of the European Union, as new trade agreements will result in lower border protection for agricultural products in the future. The high production costs of farms (see Lips et al., 2007) and high purchasing power of Swiss consumers are major reasons why Switzerland is a particularly interesting case (Aepli & Finger, 2013).

Estimations of final demand elasticities for Switzerland are scarce. Schluep Campo (2004) analyzed Swiss meat demand at the wholesale level using time series data for the period January 1996 to December 2002 with a semi-flexible almost ideal demand system. The latest final demand estimation for meat was conducted by Jaquet et al. (2000) using household expenditure data from 1998, and elasticities for milk product were estimated by Joerin (1983). Against this background, the goal of this study is to estimate price and income elasticities for meat and milk product groups for Switzerland. Since it has been found that health risks differ based on the type of meat consumed (Norat et al., 2001; 2002), our elasticity estimations for various meat types are valuable additions to public health debates centered on meat price changes.

The study adopts the quadratic almost ideal demand system (QUAIDS) using six years of repeated cross-sectional data from the Swiss household expenditure survey, a microdata set covering almost 20,000 households. It combines diverse recent estimation and data handling techniques, thereby methodologically contributing to the existing literature of similar studies in Europe (e.g. Thiele, 2008). In particular, we derived quality- adjusted prices after a recently proposed approach by Aepli and Finger (2013) who refer

Chapter III

to Majumder et al. (2012), corrected for endogeneity of total expenditure, and took into account censoring of the dependent variable using Shonkwiler and Yen‘s (1999) approach. To account for the influence of socio-demographic effects on demand, we included household characteristics in the model. Unconditional elasticities are derived by adopting a three-stage budgeting approach assuming weak separability of preferences.

The paper is structured as follows. In section 2, the economic theory behind the QUAIDS model and three-stage budgeting is explained. The subsequent section summarizes the data source and indicates how price data are derived from unit value information. Given issues of censoring and possible endogeneity, section 4 presents the econometric strategy applied in order to get unbiased elasticity estimates. The estimated demand responses are discussed in section 5, and section 6 concludes.

2 Theoretical framework Meat and milk demand elasticities demand milk and Meat To estimate a complete demand system, we apply the quadratic almost ideal demand system, developed by Banks et al. (1997). The QUAIDS is fully consistent with utility theory. When compared to the linear approximate almost ideal demand system (LA- AIDS, Deaton & Muellbauer, 1980) of rank three, it captures flexible Engel effects and therefore allows for general income responses, whereas the LA-AIDS only captures linear Engel effects and linear income responses. Furthermore, the QUAIDS satisfies the axioms of choice and allows exact aggregation across consumers because the preferences underlying the QUAIDS are of the Generalized Gorman Polar Form (Banks et al., 1997; Blackorby et al., 1978).

The Marshallian demand function in budget shares:

∑ * + , * +- (1) ( ) ( ) ( )

Meat and milk demand elasticities

Where is the budget share for product group , m is total expenditure, and p is a vector of prices. ( ) is the translog price aggregator, ( ) is the Cobb-Douglas price aggregator, which are defined as follows (after Banks et al., 1997):

( ) ∑ ∑ ∑ (2)

( ) ∏ (3)

Where n is the number of goods, and indexes i and j indicate specific goods.

This equation is the basis for our estimation for each product i, with residuals assumed to be multivariate normal distributed with zero mean and a finite variance-covariance matrix.

Consistency with demand theory requires that the adding-up, homogeneity, symmetry, and negativity conditions be fulfilled. In the QUAIDS framework, adding-up, homogeneity, and symmetry can be imposed by the following parameter restrictions:

Adding up: ∑ ; ∑ ; ∑ ; ∑ . (4)

Homogeneity: ∑ (5)

Symmetry: (6)

While negativity cannot be imposed, it can be checked whether the estimates are consistent by checking the Hicks own-price elasticities.

In practice, the condition of adding-up is satisfied by dropping one equation and estimating an n-1 equation system. The parameters of the dropped equation are computed from the restriction and the parameters of the n-1 equations.

In our study, we adopt a three-stage budgeting demand system. At each stage, we assume sub-utility functions for all product groups, which sum up overall benefit and could be optimised individually. In other words, we assume weak separability of the

Chapter III

utility function (see e.g. Moschini et al., 1994), which is a necessary condition for the estimation of demand systems in a multi-stage budgeting structure (e.g. Deaton & Muellbauer, 1980). Due to the three-stage budgeting and the assumption of weak separability, it is important that goods that are in a relationship with each other (e.g. substitutes) are put together in the same category (Deaton & Muellbauer, 1980).

Thus, we assume at stage 1 that the household first allocates its income to highly aggregated product groups, which in our case includes food, beverages, and other products and services. At Stage 2, we estimate this allocation process for the food budget, which is allocated to the product categories meat, milk, and other food products. At the third stage, we estimate the allocation of meat budget on beef, veal, pork, sheep and goat meat (all fresh or frozen), poultry (fresh or frozen, grilled or smoked), sausages, sausage products and pasties, prepared pork meat, other prepared meat (cooked, dried, salted, or smoked), and other meat (fresh or frozen or other preparation).

We further estimate the allocation of the milk budget (incl. eggs) on whole milk, milk elasticities demand milk and Meat drinks and skim milk, cheese, cream, curd, yoghurt and other milk products, and eggs. At each stage, demand responses are captured by calculating the respective elasticities.

The influence of socio-demographic variables, seasonality, and time trends is considered using the demographic translating approach proposed by Pollak and Wales

(1978). The intercept is modified to ∑ , where is the jth variable of a total number of variables (e.g. Abdulai, 2002 or Liao & Chern, 2007). The budget share

equation using Pollak and Wales‘ (1978) approach is therefore in the following form:

∑ ∑ * + , * +- ( ) ( ) ( ) (7)

The dependent and independent variables included in the demand system are described in the following section.

Meat and milk demand elasticities

3 Data description and price computation

3.1 Summary statistics

Our research utilizes data from the Swiss household expenditure survey for the period 2004-2009 that was conducted by the Swiss Federal Statistic Office. This nationally representative survey consists of repeated cross-sectional data with a periodicity of one month. In every year, almost 3,000 households participate in total. They are chosen at random from the register of private telephone lines. The sample is stratified by the seven major regions of Switzerland. The distribution of individuals is matched by a calibration method to the known distribution of the population (see SFSO, 2011 for details). In addition to information on income, expenditures, and quantities of products bought, the dataset also includes detailed information on several household characteristics (e.g. household size and age and sex of the head of household).

The household characteristics considered in the further analysis are household size, a dummy variable for the presence of young children (<5 years), the age of the household‘s reference person, and a dummy variable for the household‘s reference person having a university degree or not. Household size is calculated in terms of adult equivalents according to the OECD-modified equivalence scale (Hagenaars et al., 1994). We used the variables month and year to capture seasonality and trends respectively.

The total sample consists of 19,593 household residents in Switzerland. Table 1 presents the mean and standard deviation for total expenditure and expenditure in Swiss francs on the product group at the three stages as well as for the household characteristics that are included in the budget share equation. In addition, we provide the share of zero consumption for every expenditure variable. Given that some households do not consume meat or milk products at all, the estimations for products within the meat and milk groups are based on a reduced sample of 18,698 and 19,411 households respectively (see Table 1).

Chapter III

Table 1: Descriptive statistics of the expenditure variables and the household characteristics

Total Number of households 19,593

Percentage of zero Mean Standard deviation consumption

Total Expenditure 8763.29 5255.45 0.00

Stage 1

620.51 366.76 0.19 Expenditure for food (in Swiss Francs)

125.71 203.28 3.32 Expenditure for beverages (in Swiss Francs)

8017.07 5061.37 0.00 All other expenditure (in Swiss Francs)

Stage 2

Expenditure for meat and meat products (in Swiss Francs) 150.94 147.24 4.38

Expenditure for milk, cheese, eggs (in Swiss Francs) 106.91 71.36 0.74 elasticities demand milk and Meat

Expenditure for other food (in Swiss Francs) 363.87 221.59 0.04

Stage 3

Meat

22.23 60.61 52.28 Expenditure for beef, fresh or frozen (in Swiss Francs)

9.41 50.61 81.55 Expenditure for veal, fresh or frozen (in Swiss Francs)

21.84 47.12 50.44 Expenditure for pork, fresh or frozen (in Swiss Francs)

Expenditure for sheep and goat meat, fresh or frozen (in Swiss 5.02 22.34 87.08 Francs)

Expenditure for chicken, fresh or frozen, as well as grilled or 22.56 31.11 37.15 smoked (in Swiss Francs)

34.12 35.77 13.99 Expenditure for sausage products and pasties (in Swiss Francs)

Expenditure for pork: ham, bacon and remaining parts, cooked, 21.66 25.01 23.98 dried, salted or smoked (in Swiss Francs)

Meat and milk demand elasticities

Percentage of zero Mean Standard deviation consumption

Expenditure for other meat, cooked, dried, salted or smoked (in 6.58 17.38 66.49 Swiss Francs)

Expenditure for other meat (e.g. Game and Rabbit, Horse) (in 14.43 31.94 51.86 Swiss Francs)

Milk products

11.18 18.17 39.54 Expenditure for whole milk (in Swiss Francs)

7.14 13.79 57.00 Expenditure for milk drink and skim milk (in Swiss Francs)

48.73 43.14 5.44 Expenditure for cheese (in Swiss Francs)

8.19 11.16 32.50 Expenditure for cream (in Swiss Francs)

Expenditure for curd, yoghurt, other milk products and milk 22.99 22.75 9.67 surrogates (in Swiss Francs)

9.46 11.20 27.83 Expenditure for eggs (fresh and processed) (in Swiss Francs)

Household characteristics

1.61 0.52 -- Household size in adult equivalents

0.14 0.34 -- Young children (<=5 years, yes or no)

50.18 15.43 -- Age of the household‘s reference person

University degree with respect to the household‘s reference person -- 0.15 1.30 (yes or no)

Source: Own calculations based on the Swiss Household Expenditure Survey 2004-2009

In Switzerland, the average share of food expenditure on total expenditure is low (7.08%) in comparison to other European countries (Destatis, 2010). More than 40% of total expenditure for food is spent on meat and milk products. Expenditure on meat is higher (150.94 Swiss francs) than expenditure on milk products (106.91 Swiss francs). As can be seen from the second part in Table 1, Swiss households spend a high proportion of their meat expenditure on prepared meat such as sausage products or

Chapter III

other prepared pork products. This is followed by chicken, beef (fresh or frozen), and pork (fresh or frozen). The total expenditure on milk products is dominated by expenditure on cheese and curd, yoghurt, and other milk products.

Another point of interest is the share of zero consumption, which is low at stage 1 and stage 2, but relatively high for several product groups at stage 3, rising to 87.08% for sheep and goat meat. This implies that zero consumption should be considered at stage 3 to avoid biased estimation parameters of the budget share equation.

3.2 Retrieving price data

Like most surveys of its kind, the Swiss household expenditure survey does not contain information on prices. Therefore, most demand analyses use unit values, i.e., the division of expenditures by consumed quantity, as market prices in the model (e.g. Akbay et al., 2007). Otherwise, they employ adjusted prices for the impact of quality

based on unit values (Cox & Wohlgenant, 1986, for the application of this see e.g. elasticities demand milk and Meat Lazaridis, 2003; Park et al., 1996; Thiele, 2010). However, Majumder et al. (2012) argue, ―Unit values cannot be used as prices due to (a) measurement errors, (b) quality effects, and (c) household compositional effects on expenditure patterns‖. As mentioned by Chung et al. (2005), differences in unit values represent both price and quality variations. This especially holds for composite commodities such as beef that contain many subcategories with different quality levels. Applying a quality adjustment procedure allows us to control check for the net change in price. To generate missing price information, we therefore made use of an approach proposed by Majumder et al. (2012) and adapted by Aepli and Finger (2013) with a variable for month and year to generate regionally, monthly, and yearly quality-adjusted prices. Our choice is motivated by the fact that the approach is in contrast to Cox and Wohlgenant‘s (1986) approach consistent with the hypothesis that households are facing the same price at least in the same regional market. Furthermore, Aepli and Finger (2013) successfully tested this for the Swiss household expenditure survey using data from 2000 to 2005.

Meat and milk demand elasticities

Differences in quality of the consumed products are induced by different preferences for unobserved quality characteristics (Cox & Wohlgenant, 1986, for application see e.g. Zheng & Henneberry, 2010). Following Majumder et al. (2012) and Cox and Wohlgenant (1986), income and household characteristics are proxies for those preferences. For example, poorer households tend to choose lower quality products, and household size and composition can influence the decision of whether to take products with quantity price surcharges or price discounts (see e.g. Abdulai et al., 2009; Majumder et al., 2012). The relation between unit values and proxy variables is as follows (Aepli & Finger, 2013):

( )

∑ (8)

is the unit value paid by household for item in its region , year , and month m. The equation indicates that the difference between the unit value accepted by a particular household and the median unit value in the household‘s region can be accounted to a systematic part, which is explained by socioeconomic characteristics, income and expenditure variables of the household, and an unsystematic part captured by the error term . To allow for a non-linear relationship between income and the unit value, we include both income ( ) and the square of income ( ). is the household total food expenditure, is the household total expenditure for food consumed away from home (meals and beverages), and denotes the th of household characteristics. We include the household size, a binary dummy variables for having children, and for having a university degree. , and are dummies for region, year, and month respectively. To account for possible outliers and to limit their influence on the estimation results, we applied a robust M-estimator. Equation (8) is estimated for every item at the third stage using a sub-sample of the households that consumed the item. The regionally, monthly, and yearly quality-adjusted prices are calculated by adding the 65

Chapter III

median unit value for the item to the corresponding median of item of the estimated residuals from equation (8).

( ) ( ) ( ) (9)

The calculated market prices are assigned to each household in the whole sample account for the household‘s region as well as month and year.

For sheep and goat meat, we used monthly regional consumer market prices collected by the Swiss Federal Office for Agriculture. We were not able to generate regional quality adjusted market prices due to the very small number households in the Italian- speaking part of Switzerland that consumed sheep and goat meat. For eggs in stage 3 and all product groups in stage 1 and 2, we used Consumer Price Indices of the Swiss Federal Statistical Office (SFSO, 2013), which are equal for each observation within the

same month. Meat and milk demand elasticities demand milk and Meat

4 Econometric application and elasticity calculation

4.1 Censoring

As can be seen from the data description, zero consumption is a prominent feature for product categories at the third budget stage. With zero consumption, the dependent variables of the demand system represented by equation 7 (budget share equation of the theory section) will have dependent variables that are left-censored at zero, which

has significant econometric implications. Censoring can occur for a specific product due to a lack of preference in the household, a short survey period, or it is the expression of a typical corner solution of a household‘s utility maximization problem facing budget restrictions and prices (Perali & Chavas, 2000; Thiele, 2008). To avoid self-selection bias, Heien and Wessels (1990) proposed a two-step estimator approach based on the approach by Heckman (1979) (for application of the Heckman procedure see e.g. Lazaridis, 2003). However, Shonkwiler and Yen (1999) criticized this approach due to an

Meat and milk demand elasticities

internal inconsistency in the estimation procedure (see also Vermeulen, 2001) and proposed the consistent-two-step-estimation-procedure (SY), which is employed in this study. In the first step, a probit regression is applied to determine the probability of purchase. Instead of a univariate probit as proposed by Shonkwiler and Yen (1999) (see e.g. Lambert et al., 2006; Zheng & Henneberry, 2011), we estimated a multivariate probit, assuming that the purchase decisions for different product groups are not made independently. We can thereby account for any possible correlation among the product groups (see e.g. Yen et al., 2002 or Zheng & Henneberry, 2010). The equation system to be estimated is:

( ) , (10)

{ (11)

Where and are the observed dependent variables for the food product groups, and are the correspondent latent variables. is a binary variable and represents the decision of the household whether to consume or not. and are the and explanatory variables respectively, and and are parameters to be estimated. The explanatory variables in the multivariate probit are income, logarithmic prices, and household characteristics following Mittal (2010), Thiele (2008), and Yen et al. (2002). and are the error terms assumed to have a multivariate normal distribution, each with a mean of zero and a variance-covariance matrix V with diagonal elements of 1 and off- diagonal elements of .

Based on the multivariate probit, the standard normal cumulative distribution function ( ) (cdf) and the standard normal probability density function ( ) (pdf) are calculated for each commodity related to each household. The budget share equation system is modified as follows:

( ) ( ) ( ) , ( ) (12)

Chapter III

represents the covariance between the error term in the budget share equation and the error term of the multivariate probit model. The quadratic almost ideal demand system is estimated by the feasible generalized nonlinear least square procedure, as this approach is clearly to be preferred to the inefficient, seemingly unrelated regression technique (Drichoutis et al., 2008; Tauchmann, 2005). Tauchmann (2005) further notes that the incorporation of the standard normal cumulative distribution function and the standard normal probability density function introduces heteroscedasticity into the second step estimation (budget share equations, see eq. 7). Yen and Lin (2006), Yen (2005) and Yen et al. (2003) proposed alternatives to the SY approach. But they conclude (see also e.g. Platoni et al., 2012) that the SY approach is likely to be an attractive alternative in case of a large dataset as we are faced with. To avoid the problem of heteroscedasticity, we adjusted the variance-covariance matrix by using White‘s estimator (White, 1980, for application in demand systems see e.g. Thiele, 2008)

and calculated robust standard errors. Meat and milk demand elasticities demand milk and Meat For the stage 1 and stage 2 models, one equation is dropped to avoid singularity in the variance-covariance matrix (e.g. Abdulai, 2002; Xi et al., 2004) and the parameters of the dropped equation are then computed using the parameters of the other three equations and the adding-up restriction. For stage 1, we dropped the equation for other food and in stage 2 for meat. In this case, the adding-up condition is always fulfilled. In the stage 3 models, zero consumption is relatively high especially for some meat categories. Using the two-step estimation procedure, the right-hand side in the budget share equation generally does not add up to one anymore. Therefore, instead of dropping one equation, the second-step estimation of the stage 3 models is based on the entire set of the equations as proposed by Yen et al. (2002) and applied e.g. by Ecker and Qaim (2010) or Tefera et al. (2012). To test the adding-up condition for stage 3, we calculated the predicted budget shares and summed it for every household. We tested if the mean value of the households sum significantly differs from one by using a two-tailed t-test.

Meat and milk demand elasticities

4.2 Expenditure endogeneity

Under the assumption of weak separability, it is possible that total expenditure or expenditure on a food group is jointly determined with the budget share of a specific commodity. To avoid inconsistent estimates due to endogeneity of expenditure, we applied the augmented regression technique proposed by Blundell and Robin (1999), which was also used by Fashogbon and Oni (2013) and Tafere et al. (2010). In the first step, we regressed all the explanatory variables of the budget share equation and the logarithmic of disposable income and its square, which were used as instruments, on the logarithmic expenditure variable. In using income as an instrument, we follow Bopape (2006) in the assumption that income is exogenous in the household expenditure allocation model. Then we introduced the residuals of the first step in the budget share equation (7) to correct for endogeneity of the expenditure variable. The relevance of the instruments is verified by determining whether the estimated parameters are statistically significant in the reduced form regression (Bopape, 2006). Instead of using ordinary least squares, we applied a robust M-estimator to the reduced form regression, which limits the influence of outliers (see e.g. Finger, 2013 for details).

4.3 Elasticity estimates

With the parameters of equation (12), it is possible to estimate price and income responses via elasticities. In cases where censoring is an issue, the income elasticity

( ) was calculated using the equation , where is the differentiation of equation (12) with respect to .9 Otherwise, the income elasticity is calculated with

the simpler formula . For the price elasticity, we follow Zheng and Henneberry

(2010) and consider the full effect of a price change on demand because prices are considered in the budget share equation as well as in the multivariate probit model. The

Marshallian price elasticity considering zero consumption was calculated as

9 The differentiation results in , * +-. ( ) ( ) 69

Chapter III

( ) ( ) , where is the differentiation of equation (12) with

respect to . denotes the parameter of the price of the th good with respect to the

th good in the multivariate probit estimation and is the Kronecker delta, which is equal to one when , and otherwise zero (Banks et al., 1990; Zheng & Henneberry, 2010).10 When censoring was no issue, the Marshallian elasticity was calculated with the

formula . The Hicksian price elasticities could be calculated by applying the

Slutsky equation .

Elasticities calculated with the above equations are conditional on constant group expenditures, which is theoretically not justifiable for stages 2 and 3. We therefore make use of the three-stage budgeting demand system to calculate unconditional elasticities. The unconditional elasticities for the second and third stage are derived following

Edgerton (1997): Meat and milk demand elasticities demand milk and Meat

Second stage: ( ) ( ) (13)

Third stage: ( )( ) ( )( ) ( ) (14)

is the income elasticity (unconditional) for item , ( ) the expenditure elasticity

(conditional) for item within the th commodity group, and ( ) is the income elasticity for the th commodity group (unconditional).

( )( ) is the expenditure elasticity (conditional) for item within the th sub-commodity group within the main-commodity group , ( )( ) is the expenditure elasticity for the th sub-commodity group within the main-commodity group , and ( ) is the income elasticity for the main-commodity group (unconditional).

The unconditional, uncompensated price elasticity is calculated by:

10 The differentiation results in ( ∑ ) , * +- . ( ) ( ) 70

Meat and milk demand elasticities

Second stage: ( ) ( ) ( ) (15)

Third stage: ( )( ) ( )( ) ( )( ) ( ) ( )( ) ( )( )

( )( ) ( )( ) (16)

( ) is the conditional, compensated price elasticity for items and , ( ) is the budget share of item within the th commodity group, and is the uncompensated own- price elasticity (unconditional) of commodity group .

( )( ) is the conditional, compensated price elasticity for item with respect to the price of within the th sub-commodity group, which is within the th main-commodity group;

( )( ) is the budget share of item within the th sub-commodity group within the th main-commodity group; ( ) is the conditional, compensated own-price elasticity of sub-commodity group within the main-commodity group ; and is the unconditional, uncompensated own-price elasticity of the main-commodity group .

The unconditional, compensated price elasticity is derived analogously using the conditional, compensated price elasticity ( ) and the compensated own-price elasticity

(unconditional) of commodity group r for the second step and using ( )( ) ,

( ) and the unconditional, uncompensated own-price elasticity of main-commodity group for the third step:

Second stage: ( ) ( ) ( ) (17)

Third stage: ( )( ) ( )( ) ( )( ) ( ) ( )( ) ( )( )

( )( ) ( )( ) (18)

To assess the negativity condition, we evaluated the compensated own-price elasticities

, which should be non-positive.

Chapter III

5 Results and Discussion

For the sake of brevity, estimation results of the QUAIDS parameters are not reported here but are available in the appendix in Tables 15-18. The price and income effects will be discussed in terms of elasticities. The coefficients of determination are between 0.43 and 0.85 for stage 1 and between 0.83 and 0.96 for stage 2. For the stage 3 models, the coefficients are lower. They are between 0.09 and 0.59 for meat, and 0.25 and 0.81 for milk products (see appendix, Tables 15-18). A large proportion of the consumption behavior of the households can be explained by economic and socio-demographic variables, especially for the models at stage 1 and 2, whereas for the most disaggregated level, other determinants such as psychological factors, which could not be considered in this study, will explain the rest of the variance.

We conducted a joint significance test to consider the statistical evidence in support of

the household characteristics and the QUAIDS specification with respect to the Meat and milk demand elasticities demand milk and Meat quadratic term.11 The model specification without household characteristics is clearly rejected for all stages. The model specification in this study is therefore superior to a model without household characteristics. This is in accordance with previous findings for Switzerland (see e.g. Abdulai, 2002 or Aepli & Finger, 2013). The significance of the quadratic term in expenditure in all models except in stage 1 supports the hypothesis than the QUAIDS underlying Engel curves are of a nonlinear form and the model is superior to the linear AIDS. However, the linear AIDS model would be sufficient for the stage 1 model. This can be compared with Abdulai (2002), who found that the null

hypothesis of overall absence of the quadratic term is rejected using the same expenditure survey based on the year 1998.

As indicated before, we included all product groups in the third stage estimations so that the adding up condition is not automatically fulfilled at that stage. The results of the test

11 -test statistics for joint significance of the household characteristics: Stage 1: 6983.8, Stage 2: 719.8, Stage 3 meat: 2206.8, Stage 3 milk: 6286.7. -test statistics for joint significance of the quadratic expenditure term: Stage 1:3.2, Stage 2: 29.8, Stage 3 meat: 55.7, Stage 3 milk: 97.5. 72

Meat and milk demand elasticities

for adding-up are presented in Table 2. It can be seen that the adding-up condition is rejected for the meat model at the 1% level but not for the milk model. As Klonaris and Hallam (2003) explained, the rejection of restrictions in demand models is not uncommon (see also e.g. Klonaris, 1999 or Mergos & Donatos, 1989) and could also depend on the aggregation of the product groups. We had the same experience with our data set.

Table 2: Adding-up condition

mean of the households sum of t-value (two- standard Error p-value conclusion the predicted budget sided) shares

Stage 3: Meat 0.999 0.000 -3.855 0.000 not satisfied

Stage 3: Milk 1.000 0.000 -0.725 0.468 satisfied

Tables 3 to 6 report the unconditional price and income elasticities for stage 3 based on the estimates of coefficients of the budget share equation (see appendix). Standard errors and the significance level are computed by the delta method (Oehlert, 1992). The conditional elasticities at stage 3, as well as the elasticities at stage 1 and 2, are presented in the appendix.

Chapter III

Table 3: Unconditional, Marshallian price and income elasticities at stage 3, meat (at sample means)

Be Ve Po SG Pl SSP PHB OMp OM Income

-0.367 0.027 -0.210 0.592 -0.026 -0.039 0.009 0.117 0.192 0.353 Beef, fresh or frozen (0.074)** (0.064) (0.065)** (0.087)** (0.093) (0.102) (0.092) (0.070) (0.054)** (0.032)**

-0.255 -0.836 0.435 -2.112 0.435 0.822 1.058 0.021 0.093 1.065 Veal, fresh or frozen (0.134) (0.097)** (0.135)** (0.299)** (0.182)* (0.275)** (0.186)** (0.112) (0.112) (0.077)**

-0.048 0.178 -0.896 0.814 -0.047 1.029 0.027 -0.133 -0.130 0.471 Pork, fresh or frozen (0.068) (0.057)** (0.102)** (0.121)** (0.100) (0.120)** (0.092) (0.077) (0.063)* (0.033)**

Sheep and goat meat, -0.105 0.329 0.331 -0.791 1.101 0.188 -0.029 0.276 -0.172 0.778 fresh and frozen (0.138) (0.152)* (0.106)** (0.153)** (0.195)** (0.186) (0.177) (0.099)** (0.134) (0.071)**

Poultry, fresh or 0.130 -0.107 -0.089 0.999 -1.334 1.220 -0.976 0.012 0.092 0.022 frozen, as well as (0.054)* (0.070) (0.067) (0.121)** (0.084)** (0.127)** (0.091)** (0.079) (0.058) (0.031) grilled and smoked

Sausages, sausage -0.148 0.301 0.719 -0.723 0.952 -1.156 0.240 0.053 -0.191 0.520 Meat and milk demand elasticities demand milk and Meat products and pasties (0.061)* (0.040)** (0.071)** (0.071)** (0.086)** (0.144)** (0.092)** (0.061) (0.053)** (0.033)**

Pork: ham, bacon and remaining parts, 0.229 -0.007 -0.174 -1.154 0.312 0.316 0.159 0.343 0.472 0.708 cooked, dried, salted (0.084)** (0.053) (0.095) (0.108)** (0.117)** (0.180) (0.130) (0.085)** (0.072)** (0.045)** or smoked

Other meat, cooked, 0.019 0.254 0.007 -0.334 0.076 0.024 0.264 -0.770 0.183 0.632 dried, salted or (0.111) (0.098)** (0.125) (0.253) (0.155) (0.178) (0.151) (0.172)** (0.101) (0.065)** smoked

Other meat (e.g. game and rabbit, 0.182 0.206 -0.197 1.023 -0.045 -0.206 0.077 0.375 -0.327 0.401

horse, preserved (0.068)** (0.068)** (0.078)* (0.104)** (0.106) (0.111) (0.101) (0.086)** (0.090)** (0.034)** meat)

* and ** denote significance at 5% and 1% level, respectively.

Note: Be: Beef, fresh or frozen, Ve: Veal, fresh or frozen, Po: Pork, fresh or frozen, SG: Sheep and goat meat, fresh and frozen, Pl: Poultry, fresh or frozen, as well as grilled and smoked, SSP: Sausages, sausage products and pasties, PHB: Pork: ham, bacon and remaining parts, cooked, dried, salted or smoked, OMp: Other Meat, cooked, dried, salted or smoked, OM: Other meat (e.g. game and rabbit, horse, preserved meat)

Meat and milk demand elasticities

Table 4: Unconditional, Hicksian price and income elasticities at stage 3, meat (at sample means)

Be Ve Po SG Pl SSP PHB OMp OM Income

-0.366 0.027 -0.209 0.593 -0.025 -0.037 0.010 0.118 0.192 0.353 Beef, fresh or frozen (0.074)** (0.064) (0.065)** (0.087)** (0.093) (0.102) (0.092) (0.070) (0.054)** (0.032)**

-0.253 -0.836 0.437 -2.112 0.438 0.827 1.061 0.022 0.095 1.065 Veal, fresh or frozen (0.134) (0.097)** (0.135)** (0.299)** (0.182)* (0.275)** (0.186)** (0.112) (0.112) (0.077)**

-0.047 0.179 -0.895 0.815 -0.046 1.031 0.028 -0.133 -0.129 0.471 Pork, fresh or frozen (0.068) (0.057)** (0.102)** (0.121)** (0.100) (0.120)** (0.092) (0.077) (0.063)* (0.033)**

Sheep and goat -0.104 0.330 0.333 -0.790 1.103 0.191 -0.027 0.277 -0.171 0.778 meat, fresh and (0.138) (0.152)* (0.106)** (0.153)** (0.195)** (0.186) (0.177) (0.099)** (0.134) (0.071)** frozen

Sausages, sausage -0.147 0.302 0.720 -0.722 0.953 -1.154 0.241 0.054 -0.190 0.520 products and pasties (0.061)* (0.040)** (0.071)** (0.071)** (0.086)** (0.144)** (0.092)** (0.061) (0.053)** (0.033)**

Pork: ham, bacon and remaining parts, 0.230 -0.007 -0.173 -1.154 0.313 0.320 0.161 0.344 0.473 0.708 cooked, dried, salted (0.084)** (0.053) (0.095) (0.108)** (0.117)** (0.180) (0.130) (0.085)** (0.072)** (0.045)** or smoked

Other meat, cooked, 0.021 0.254 0.009 -0.334 0.078 0.027 0.266 -0.769 0.184 0.632 dried, salted or (0.111) (0.098)** (0.125) (0.253) (0.155) (0.178) (0.151) (0.172)** (0.101) (0.065)** smoked

Other meat (e.g. game and rabbit, 0.183 0.206 -0.196 1.023 -0.044 -0.204 0.078 0.376 -0.327 0.401 horse, preserved (0.068)** (0.068)** (0.078)* (0.104)** (0.106) (0.111) (0.101) (0.086)** (0.090)** (0.034)** meat)

* and ** denote significance at 5% and 1% level, respectively.

Chapter III

Table 5: Unconditional, Marshallian price and income elasticities at stage 3, milk products (at sample means)

Wm MS Ch Cr CY Eg Income

-1.235 0.399 1.291 0.220 0.328 -0.617 0.111 Whole milk (0.168) ** (0.119) ** (0.215) ** (0.094) * (0.158) * (0.198) ** (0.026) **

0.002 -0.289 1.399 -0.024 0.155 -0.638 0.320 Milk drink and skim milk (0.117) (0.126) * (0.312) ** (0.082) (0.187) (0.167) ** (0.034) **

Cream, soft, medium- 0.880 0.069 0.266 -0.399 0.285 0.116 0.317 hard, hard, extra hard and (0.145) (0.081) (0.059) ** (0.277) (0.057) ** (0.085) (0.021) ** processed cheese **

0.580 -0.103 1.270 -0.826 -0.033 1.298 0.309 Cream (0.129) ** (0.096) (0.289) ** (0.106) ** (0.171) (0.195) ** (0.025) **

Curd, yoghurt, other milk -0.850 0.824 -0.524 1.036 0.199 0.253 0.496 products and milk (0.244) (0.144) ** (0.111) ** (0.443) * (0.103) (0.174) (0.034) **

surrogates ** Meat and milk demand elasticities demand milk and Meat Eggs, fresh and 0.167 0.160 0.926 -0.468 -0.158 0.059 0.059 processed (0.170) (0.145) (0.242) ** (0.109) ** (0.232) (0.319) (0.033)

* and ** denote significance at 5% and 1% level, respectively.

Note: Wm: Whole milk, MS: Milk drink and skim milk, Ch: Cream, soft, medium-hard, hard, extra hard and processed cheese, Cr: Cream, CY: Curd, yoghurt, other milk products and milk surrogates, Eg: Eggs, fresh and processed

Meat and milk demand elasticities

Table 6: Unconditional, Hicksian price and income elasticities at stage 3, milk products (at sample means)

Wm MS Ch Cr CY Eg Income

-1.235 0.399 1.292 0.221 0.329 -0.617 0.111 Whole milk (0.168) ** (0.119) ** (0.215) ** (0.094) * (0.158) * (0.198) ** (0.026) **

0.003 -0.289 1.401 -0.023 0.156 -0.637 0.320 Milk drink and skim milk (0.117) (0.126) * (0.312) ** (0.082) (0.187) (0.167) ** (0.034) **

Cream, soft, medium- 0.881 0.069 0.267 -0.397 0.285 0.116 0.317 hard, hard, extra hard and (0.145) (0.081) (0.059) ** (0.277) (0.057) ** (0.085) (0.021) ** processed cheese **

0.580 -0.102 1.272 -0.825 -0.032 1.299 0.309 Cream (0.129) ** (0.096) (0.289) ** (0.106) ** (0.171) (0.195) ** (0.025) **

Curd, yoghurt, other milk -0.849 0.825 -0.524 1.039 0.200 0.253 0.496 products and milk (0.244) (0.144) ** (0.111) ** (0.443) * (0.103) (0.174) (0.034) ** surrogates **

Eggs, fresh and 0.167 0.160 0.926 -0.468 -0.158 0.059 0.059 processed (0.170) (0.145) (0.242) ** (0.109) ** (0.232) (0.319) (0.033)

* and ** denote significance at 5% and 1% level, respectively.

Income elasticities are all positive and, expect for poultry and eggs, statistically significantly different from zero for milk product categories and meat product categories, which is consistent with the findings in Jaquet et al. (2000) who estimated a demand system based on household expenditure data for Switzerland in 1998. The magnitude of the income elasticity implies that all of the product categories except veal are rated as necessary goods. This is in line with the findings of Jaquet et al. (2000), who reported that all meat and milk categories are necessities. The reason for the deviation between Jaquet et al. (2000) and our estimates with respect to veal probably lies in the fact that they estimated a linear version of the AIDS without considering nonlinearity in Engel curves or censoring the endogenous variable of the budget share equation. This could

Chapter III

lead to bias estimates, especially for product groups with a high share of zero consumption such as veal.

The majority of the point estimates of the Hicksian own-price elasticities for meat and milk product categories are significantly negative, whereas only the own-price elasticity for pork (ham, bacon etc.) and eggs are not statistically different from zero. Overall, this implies that the expenditure function is concave in prices and the negativity restriction is fulfilled. Furthermore, the magnitude of the own- and cross-price elasticities ranks the commodities as price inelastic with some exceptions, especially for the meat product categories.

Most of the milk products are substitutes for each other, which was expected and is in line with the findings in Jaquet et al. (2000). There are some exceptions, e.g. eggs, which is a complementary good to whole milk and milk drink/skim milk. Nevertheless, the

elasticities for eggs have to be interpreted carefully since the consumer price index, Meat and milk demand elasticities demand milk and Meat which has a relatively low variance, has been used as market price in the budget share equation. For whole milk and curd, yogurt and cheese, the compensated own-price elasticity is lower than in Jaquet et al. (2000). This indicates that consumers have been getting more price sensitive in recent years, which is in line with a recently published study on Switzerland that notes Swiss consumers are price sensitive, especially to fluid milk prices (Schwarzenbach et al., 2013). This is probably caused by the fact that the potential for product diversification is lower in comparison to other highly processed milk products.

As can be seen from table 4, meat product groups are generally substitutes for each other, which confirms the results of other studies for Switzerland (see e.g. Bernegger & Strasser, 1986 or Jaquet et al., 2000). As observed already for some milk products, the own-price elasticities are much lower than estimated in Jaquet et al. (2000), which means that consumers respond more elastically now than they did in 1998. This holds true especially for sausages, sausage products, pork, veal, and poultry, and it confirms the assumption that price sensitivity has increased in recent years, which can also be 78

Meat and milk demand elasticities

seen in the increasing shopping tourism of Swiss residents abroad (Anwander Phan- Huy, 2006). Looking at the compensated elasticities, an additional increase in poultry prices by 1% would result in an increase of the household demand for the same product category by 1.33%.

Aepli and Finger (2013) conducted more recent research for Switzerland. They estimated the influence of different meat prices and household characteristics as well as geographic variables on demand for sheep and goat meat, applying a tobit model based on the Swiss household expenditure survey from 2000 to 2005. In comparison to our model, Aepli and Finger‘s estimation contained only a single equation, taking into account any possible interactions between product groups in a very limited way. Our results show that rising prices for sheep and goat meat lead to a decline in its demand, that beef is more a complementary good (nevertheless, the elasticity is not significantly different from zero), and pork is a substitute, which is in line with the findings in Aepli and Finger‘s (2013) report.

6 Conclusions and policy implications

This paper reports on income and price elasticities for different food categories, especially meat and milk products, obtained by a three-stage quadratic almost ideal demand system including household characteristics based on the Swiss household expenditure survey from 2004 to 2009 including almost 20,000 Swiss households. Prices were generated by a recently proposed method in Aepli and Finger‘s (2013) work. We corrected for endogeneity of the expenditure variable as well as for zero consumption using Shonkwiler and Yen‘s (1999) approach. Overall testing of the model for household characteristics and the quadratic expenditure term confirms the specification of the budget share equations. The evidence presented in this study supports the idea that Engel curves for food product groups are often of a nonlinear form and, in particular, that meat and milk consumption cannot be only explained by price and income changes. Household characteristics such as household size or the age of the

Chapter III

household‘s reference person significantly influence consumption. According to the results, almost all meat and milk product categories have positive and significant income elasticities, with milk products more responsive to income fluctuations. Furthermore, own-price elasticities for meat and milk product groups are substantially higher than in previous studies for Switzerland, confirming presumptions that Swiss consumers have become more price sensitive in recent years.

Switzerland has one of the highest GDP per capita in the world (Worldbank, 2014) and is a particularly interesting case to investigate consumer‘s price and income responses against the background that Swiss consumers only spend a very small amount of their total expenditure on food. For policy makers, this study provides some useful insight on the potential consequences of setting price policies directly through a tax or indirectly through free trade agreements. Switzerland still has very high border protection as well as a high level of domestic support for agriculture, especially for meat and some milk

products. Tariff-rate quotas have a highly restrictive impact on trade and are a major elasticities demand milk and Meat reason for significantly higher consumer prices in Switzerland compared to other European countries (BAK, 2010). A free trade agreement for agricultural products will decrease the average price level for food by about 10% (Bösch et al., 2011) and possibly even by as much as 40% (SFOA, 2008) for highly protected product groups like meat and some milk products. Following our results, the price decrease will have a positive effect on demand for meat and milk. For instance, a 1% decrease in the price of sausages will increase demand by 1.15%. While the findings for veal are in line with previous findings in literature, the relatively high price elasticities for sausages, pork, or poultry in the context of a high-income country are new. With red and processed meat, in particular sausages, higher consumption is a risk factor (e.g. for colon and gastric cancer [see e.g. Paar et al., 2013 or Zhu et al., 2013]), and a price reduction could have a direct influence on public health due to the high price elasticity for these product groups. In addition to public health concerns, a higher demand for meat from ruminants and milk products due to trade liberalization will lead to higher greenhouse gas emissions in Switzerland or abroad depending on whether the products are domestically 80

Meat and milk demand elasticities

produced or imported. For instance, 1kg of fluid milk generates an average of 2kg of

CO2-aequivalents, and 1kg of Mozzarella generates 7kg CO2-aequivalents (Kim et al., 2013; Thoma et al., 2013). A recently published study for Switzerland shows that domestically produced beef generates more than 25kg CO2-aequivalents per kg (Alig et al., 2012). A further concern is that the production of animal products needs significant amounts of feed. In some cases, this feed could be used for direct human consumption. Price reduction will boost consumption of whole milk as well as beef and other food products and therefore have a direct influence on the environment. These estimations call for serious considerations of further policy measures in the case of trade liberalization to avoid negative environmental effects due to price reductions.

From the methodological perspective, further research is needed, in particular to theoretically implement the adding-up condition under the consisting two-step estimation approach, which is not fulfilled for the meat model. This would increase the precision of the estimations. Furthermore, research should be conducted to estimate alternative specifications of the QUAIDS and to check the robustness of the results. Extended models might allow for flexible parameter estimations with respect to different household types e.g. different income groups or different education levels to find out if high-income households behave differently in comparison to middle- or low-income households. As price policies directly influence the household welfare, specific elasticities for different household types would give an overview of the extent of the liability of individual household types.

Chapter III

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Abdulai, A., Kuhlgatz, C., & Schmitz, S. (2009). Empirical Investigation of Price Setting and Quantity Surcharges in the German Food Sector. Agribusiness, 25(3), 331-350.

Aepli, M., & Finger, R. (2013). Determinants of sheep and goat meat consumption in Switzerland. Agricultural and Food Economics. doi:10.1186/2193-7532-1-11

Akbay, C., Boz, I., & Chern, W.S. (2007). Household food consumption in Turkey. European Review of Agricultural Economics, 34(2), 209-231.

Alig, M., Grandl, F., Mieleitner, J., Nemecek, T., & Gaillard, G. (2012). Life cycle assessment of beef, pork and poultry. Swiss Federal Department of Economic Affairs, Agroscope Reckenholz-Tänikon, Tänikon.

Anwander Phan-Huy, S. (2006). Das nahe Ausland als Konkurrent im Schweizer Meat and milk demand elasticities demand milk and Meat Detailhandel. Die Volkswirtschaft, 6, 24-27.

Armagan, G., & Akbay, C. (2008). An econometric analysis of urban households‘ animal products consumption in Turkey. Applied Economics, 40(15), 2029-2036.

BAK (2010). Kosten, Preise, Performance. Der Schweizer Detailhandel im internationalen Vergleich. Scientific report, BAKBASEL, Basel.

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel Curves and Consumer Demand. The Review of Economics and Statistics, 79(4), 527-539.

Bernegger, U., & Strasser, M. (1986). Analyse der Fleischnachfrage in der Schweiz.

Scientific report, ETH Zurich, Zurich.

Blackorby, Ch., Boyce, R., & Russell, R.R. (1978). Estimation of demand systems generated by the Gorman Polar Form: A generalization of the S-Branch Utility Tree. Econometrica, 46(2), 345-363.

Blundell, R., & Robin, J.M. (1999). Estimation in large and disaggregated demand systems: An estimator for conditionally linear systems. Journal of Applied Economics, 14(2), 209-232.

Meat and milk demand elasticities

Bopape, L. (2006). The influence of demand model selection on household welfare estimates: An application to South African food expenditures. Unpublished doctoral dissertation, Michigan State University, Michigan.

Bösch, I., Weber, M., Aepli, M., & Werner, M. (2011). Folgen unterschiedlicher Öffnungsszenarien für die Schweizer Nahrungsmittelindustrie. Scientific report, ETH Zurich, Zurich.

Brehe, M. (2007). Ein Nachfragesystem für dynamische Mikrosimulationsmodelle. Unpublished doctoral dissertation, University of Potsdam, Potsdam.

Chung, C., Dong, D., Schmit, T.M., Kaiser, H.M., & Gould, B.W. (2005). Estimation of price elasticities from cross-sectional data. Agribusiness, 21(4), 565-584.

Cox, T.L., & Wohlgenant, M.K. (1986). Prices and Quality Effects in Cross-Sectional Demand Analysis. American Journal of Agricultural Economics, 68(4), 908-919.

Deaton, A. (1990). Price elasticities from survey data: Extensions and Indonesian results. Journal of Econometrics, 44(3), 281-309.

Deaton, A., & Muellbauer, J. (1980). An almost ideal demand system, American Economic Review, 70(3), 312-326.

Destatis. (2010). German Federal Statistical Office, Wiesbaden. https://www.destatis.de/DE/ZahlenFakten/LaenderRegionen/Internationales/Thema/Tab ellen/Basistabelle_KonsumN.html. Accessed 16 Oct, 2013.

Drichoutis, A.C., Klonaris, S., Lazaridis, P., & Nayga, R.M. (2008). Household food consumption in Turkey: a comment, European Review of Agricultural Economics, 35(1), 93-98.

Ecker, O., & Qaim, M. (2010). Analyzing nutritional impacts of policies: An empirical study for Malawi. World Development, 39(3), 412-428.

Edgerton, D.L. (1997). Weak Separability and the Estimation of Elasticities in Multistage Demand Systems. American Journal of Agricultural Economics, 79(1), 62-79.

Fashogbon, A. E., & Oni, O.A. (2013). Heterogeneity in rural household food demand and its determinants in Ondo State, Nigeria: An application of quadratic almost ideal demand system. Journal of Agricultural Science, 5(2), 169-177.

Chapter III

Finger, R. (2013). Investigating the performance of different estimation techniques for crop yield data analysis in crop insurance applications. Agricultural Economics, 44(2), 217-230.

Gallet, C.A. (2010). The income elasticity of meat: A meta-analysis. The Australian Journal of Agricultural and Resource Economics, 54(4), 477-490.

Gallet, C.A. (2009). Meat meets meta: A quantitative review of the price elasticity of meat. American Journal of Agricultural Economics, 92(1), 258-272.

Gorman, W. M. (1959). Separability utility and aggregation. Econometrica, 27(3), 469- 481.

Green, R., & Alston, J. (1990). Elasticities in AIDS Models. American Journal of Agricultural Economics, 72(2), 442-445.

Heckman, J. (1979). Sample selection bias as a specification error. Econometrica, 47(1), elasticities demand milk and Meat 153-161.

Heien, D., & Wessels, C.R. (1990). Demand Systems Estimation with Microdata: A censored Regression Approach. Journal of Business & Economic Statistics, 8(3), 365- 371.

Huang, K. S., & Lin, B-H. (2000). Estimation of food demand and nutrient elasticities form household survey data. Food and Rural Economics Division, Economic Research Service, U.S. Department of Agriculture. Technical Bulleting No. 1887.

Jaquet. P., Abdulai, A., & Rieder, P. (2000). Empirische Analyse des

Nahrungsmittelverbrauchs in der Schweiz: Ein dreistufiges LA/AIDS Modell. Scientific report, ETH Zurich, Zurich.

Joerin, R. (1983). Ökonomische Analyse der Entscheidungen über die Steuerung des Milchmarktes. Unpublished doctoral dissertation, ETH Zurich, Zurich.

Kim, D., Thoma, G., Nutter, D., Milani, F., Ulrich, R., & Norris, G. (2013). Life cycle assessment of cheese and whey production in the USA. International journal of life cycle assessment, 18(5), 1019-1035.

Meat and milk demand elasticities

Klonaris, S., & Hallam, D. (2003). Conditional and unconditional food demand elasticities in a dynamic multistage demand system. Applied Economics, 35(5), 503-514.

Klonaris, S. (1999). Applied demand analysis for food in Greece: Exploration of an alternative AIDS Model. Unpublished doctoral dissertation, University of Reading, Reading.

Lambert, R., Larue, B., Yélou, C., & Criner, G. (2006). Fish and meat demand in Canada: Regional differences and weak separability. Agribusiness, 22(2), 175-199.

Lazaridis, P. (2003). Household meat demand in Greece: A demand systems approach using microdata. Agribusiness, 19(1), 43-59.

Liao, H., & Chern, W.S. (2007). A dynamic analysis of food demand patterns in China. Paper presented at the AAEA Annual Meeting, Portland, USA.

Lips. M., Ammann, H., Signer, A., & Steingruber, E. (2007). Kostenvergleich im Ackerbau zwischen der Schweiz und Baden-Württemberg. Scientific report, Forschungsanstalt Agroscope Reckenholz-Tänikon, Ettenhausen.

Mergos, G.J., & Donatos, G.S. (1989). Consumer behaviour in Greece: An application of the almost ideal demand system. Applied economics, 21(7), 983-993.

Mittal, S. (2010). Application of the QUAIDS Model to the Food sector in India. Journal of Quantitative Economics, 8(1), 42-54.

Moschini, G., Moro, D., & Green, R.D. (1994). Maintaining and testing separability in demand systems. American Journal of Agricultural Economics, 76(1), 61-73.

Norat, T., Lukanova, A., Ferrari, P., & Riboli, E. (2002). Meat consumption and colorectal cancer risk: dose-response meta-analysis of epidemiological studies. International Journal of Cancer, 98(2), 241–256.

Norat, T., Bingham, S., Ferrari, P., Slimani, N., Jenab, M., Mazuir, M., Overvad, K., Olsen, A., Tjønneland, A., Clavel, F., Boutron-Ruault, M.C., Kesse, E., Boeing, H., Bergmann, M.M., Nieters, A., Linseisen, J., Trichopoulou, A., Trichopoulos, D., Tountas, Y., Berrino, F., Palli, D., Panico, S., Tumino, R., Vineis, P., Bueno-de-Mesquita, H.B., 85

Chapter III

Peeters, P.H., Engeset, D., Lund, E., Skeie, G., Ardanaz, E., González, C., Navarro, C., Quirós, J.R., Sanchez, M.J., Berglund, G., Mattisson, I., Hallmans, G., Palmqvist, R., Day, N.E., Khaw, K.T., Key, T.J., San Joaquin, M., Hémon, B., Saracci, R., Verbeke, W., & Ward, R.W. (2001). A fresh meat almost ideal demand system incorporating negative TV press and advertising impact. Agricultural Economics, 27(2/3), 339-353.

Oehlert, G.W. (1992). A note on the delta method. The American Statistician, 46(1), 27- 29.

Paar, C.L., Hjartaker, A., Lund, E., & Veierod, V.B. (2013). Meat intake, cooking methods and risk of proximal colon, distal colon and rectal cancer: The Norwegian Women and Cancer (NOWAC) cohort study. International Journal of Cancer, 133(5), 1153-1163.

Park, J.L., Holcomb, R.B., Raper, K.C., & Capps. O. (1996). A Demand Systems Analysis of Food Commodities by U.S. Households Segmented by Income. American Journal of Agricultural Economics, 78(2), 290-300.

Perali, F., & Chavas, J.P. (2000). Estimation of censored demand equations from large cross-section data. American Journal of Agricultural Economics, 82(4), 1022-1037. elasticities demand milk and Meat

Platoni, S., Schokai, P., & Moro, D. (2012). Panel data estimation techniques and farm- level data methods. American Journal of Agricultural Economics, 94(5), 1202-1217.

Pollak, R. A., & Wales, T.J. (1978). Estimation of complete demand systems from household budget data: The Linear and Quadratic Expenditure Systems. American Economic Review, 68(3), 348-359.

Schluep Campo, I. (2004). Market Options in the WTO Doha Round impacts on the Swiss 420 meat market. Dissertation, ETH Zurich, Zurich.

Schröck, R. (2013). Qualitäts- und Endogenitätsaspekte in Nachfragesystemen: Eine vergleichende Schätzung von Preis- und Ausgabenelastizitäten der Nachfrage nach ökologischem und konventionellem Gemüse in Deutschland. German Journal of Agricultural Economics, 62(1), 18-38.

Schwarzenbach, R., Blättler, T., Durgiai, B., & Haller, T. (2013). Marktöffnung ―Weisse Linie‖. Scientific report, Bern University of applied Sciences, Bern.

SFSO (2013). Consumer Price Index. Swiss Federal Statistical Office, Neuchâtel.

Meat and milk demand elasticities

SFSO (2011). Haushaltsbudgeterhebung 2005. Swiss Federal Statistical Office, Neuchâtel.

SFOA (2008). Verhandlungen Schweiz-EU für ein Abkommen im Bereich der öffentlichen Gesundheit (GesA). Swiss Federal Office for Agriculture, Bern.

Shonkwiler, J.S., & Yen, S.T. (1999). Two-step estimation of a censored system of equations. American Journal of Agricultural Economics, 81(4), 972-982.

Tafere, K., Taffesse, A.S., & Tamru, S. (2010). Food demand elasticities in Ethiopia: Estimates using household income consumption expenditure (HICE) survey data. Discussion paper, International Food Policy Research Institute, Washington.

Tauchmann, H. (2005). Efficiency of two-step estimators for censored systems of equations: Shonkwiler and Yen reconsidered. Applied Economics, 37(4), 367-374.

Tefera, N., Demeke, M., & Rashid, S. (2012). Welfare impacts of rising food prices in rural Ethiopia: A quadratic almost ideal demand system approach. Paper presented at the International Association of Agricultural Economists (IAAE), Foz de Iguacu, Brazil.

Thiele, S. (2010). Erhöhung der Mehrwertsteuer für Lebensmittel: Budget- und Wohlfahrtseffekte für Konsumenten. Jahrbücher für Nationalökonomie und Statistik, 230(1), 115-130.

Thoma, G., Popp, J., Nutter, D., Shonnard, D., Ulrich, R., Matlock, M., Kim, D.S., Neiderman, Z., Kemper, N., East, C., & Adom, F. (2013). Greenhouse gas emissions from milk production and consumption in the United States: A cradle-to-grave life cycle assessment circa 2008. International Dairy Journal, 31(1), 3-14.

Vermeulen, F. (2001). A note of Heckman-type corrections in models for zero expenditures. Applied Economics, 33(9), 1089-1092.

Walker, P., Rhubart-Berg, P., McKenzie, S., Kelling, K., & Lawrence, R.S. (2005). Public health implications of meat production and consumption. Public Health Nutrition, 8(4), 348-356.

White, H. (1980). A heteroscedasticity consistent covariance matrix and a direct test for heteroscedasticity. Econometrica, 46(4), 817-838. 87

Chapter III

Wood, B.D.K., Nelson, C.H., & Nogueira, L. (2012). Poverty effects of food price escalation: The importance of substitution effects in Mexican households. Food Policy, 37(1), 77-85.

Worldbank (2014). Data: GDP per capita. The World Bank, Washington DC.

Xi, J., Mittelhammer, R., & Heckelei, T. (2004). A quaids model of Japanese meat demand. Paper presented at the AAEA Annual Meeting, Denver, USA.

Yen, S.T. (2005). A multivariate sample-selection model: Estimating cigarette and alcohol demands with zero observations. American Journal of Agricultural Economics, 87(2), 453-466.

Yen, S.T., Kan, K., & Su S.J. (2002). Household demand for fats and oils: two-step estimation of a censored demand system. Applied economics, 34(14), 1799-1806.

Yen, S.T., & Lin, B.H. (2006). A sample selection approach to censored demand systems. American Journal of Agricultural Economics, 88(3), 742-749.

Yen, S.T., Lin, B.H., & Smallwood, D.M. (2003). Quasi- and simulated-likelihood approaches to censored demand systems: Food consumption by food stamp recipients elasticities demand milk and Meat in the United States. American Journal of Agricultural Economics, 85(2), 458-478.

Zheng, Z., & Henneberry, S.R. (2011). Household food demand by income category: Evidence from household survey data in an urban Chinese province. Agribusiness 27(1), 99-113.

Zheng, Z., & Henneberry, S.R. (2010). An Analysis of Food Grain Consumption in Urban Jiangsu Province of China, Journal of Agricultural and Applied Economics, 42(2), 337- 355.

Zhu, H., Yang, X., Zhang, C., Zhu, C., Tao, G., Zhao, L., Tang, S., Shu, Z., Cai, J., Dai,

S., Qin, Q., Xu, L., Cheng, H., & Sun, X. (2013). Red and Processed Meat Intake Is Associated with Higher Gastric Cancer Risk: A Meta-Analysis of Epidemiological Observational Studies. PLoS ONE, 8(8), 1-10.

Meat and milk demand elasticities

Appendix

Table 7: Unconditional, Marshallian price and income elasticities at stage 1 and 2 (at sample means)

Stage 1 Fo Be APS Income

-0.153 0.219 -0.470 0.405 Food (0.272) (0.189) (0.268) (0.019)**

1.446 -1.764 -0.567 0.884 Beverages (0.863) (0.942) (0.812) (0.044)**

-0.098 -0.007 -0.950 1.054 Tobacco (0.025)** (0.010) (0.029)** (0.002)**

All other products and -0.153 0.219 -0.470 0.405 services (0.272) (0.189) (0.268) (0.019)**

Stage 2 Me Mi OF Income

Meat and Meat 0.484 -0.639 -0.654 0.516 Products (0.440) (0.495) (0.328)* (0.028)**

-1.388 1.059 -0.142 0.301 Milk, Cheese, Eggs (0.479)** (0.595) (0.269) (0.019)**

0.179 -0.125 -0.673 0.395 Other food (0.171) (0.225) (0.145)** (0.020)**

* and ** denote significance at 5% and 1% level, respectively.

Note: Fo: Food, Be: Beverages, APS: All other products and services, Me: Meat and Meat products, Mi: Milk, Cheese, Eggs, OF: Other food

Chapter III

Table 8: Unconditional, Hicksian price and income elasticities at stage 1 and 2 (at sample means)

Stage 1 Fo Be APS Income

-0.121 0.225 -0.103 0.405 Food (0.272) (0.189) (0.267) (0.019))**

1.516 -1.751 0.234 0.884 Beverages (0.863) (0.942) (0.809) (0.044))**

-0.014 0.009 0.005 1.054 Tobacco (0.025) (0.010) (0.029) (0.002))**

All other products and -0.121 0.225 -0.103 0.405 services (0.272) (0.189) (0.267) (0.019))**

Stage 2 Me Mi OF Income Meat and milk demand elasticities demand milk and Meat

Meat and Meat 0.493 -0.631 0.071 0.493 Products (0.440) (0.495) (0.324) (0.440)

Milk, Cheese, -1.383 1.063 0.281 -1.383 Eggs (0.480))** (0.594) (0.268) (0.480))**

0.186 -0.120 -0.117 0.186 Other food (0.171) (0.225) (0.144) (0.171)

* and ** denote significance at 5% and 1% level, respectively.

Meat and milk demand elasticities

Table 9: Conditional, Marshallian price and income elasticities at stage 2 (at sample means)

Stage 2 Me Mi OF Income

Meat and Meat 0.242 -0.832 -0.683 1.272 Products (0.433) (0.491) (0.327)* (0.032)**

-1.529 0.946 -0.159 0.743 Milk, Cheese, Eggs (0.477)** (0.594) (0.269) (0.030)**

-0.006 -0.274 -0.695 0.975 Other food (0.160) (0.220) (0.145)** (0.013)**

* and ** denote significance at 5% and 1% level, respectively.

Table 10: Conditional, Hicksian price and income elasticities at stage 2 (at sample means)

Stage 2 Me Mi OF Income

Meat and 0.528 -0.603 0.076 1.272 Meat (0.433) (0.491) (0.324) (0.032)** Products

Milk, -1.363 1.079 0.284 0.743 Cheese, (0.477)** (0.593) (0.268) (0.030)** Eggs

0.212 -0.098 -0.114 0.975 Other food (0.160) (0.220) (0.144) (0.013)**

* and ** denote significance at 5% and 1% level, respectively.

Chapter III

Table 11: Conditional, Marshallian price and income elasticities at stage 3, meat (at sample means)

Be Ve Po SG Pl SSP PHB OMp OM Income

-0.480 -0.015 -0.327 0.565 -0.181 -0.300 -0.154 0.065 0.105 0.685 Beef, fresh or frozen (0.067)** (0.063) (0.054)** (0.087)** (0.080)* (0.068)** (0.073)* (0.068) (0.049)* (0.050)**

-0.596 -0.963 0.083 -2.194 -0.033 0.035 0.568 -0.137 -0.170 2.066 Veal, fresh or frozen (0.094)** (0.089)** (0.087) (0.300)* (0.119) (0.147) (0.107)** (0.104) (0.082)* (0.098)**

-0.199 0.122 -1.051 0.779 -0.254 0.681 -0.189 -0.203 -0.246 0.913 Pork, fresh or frozen (0.052)** (0.054)* (0.090)** (0.120)** (0.078)** (0.062)** (0.062)** (0.074)** (0.053)** (0.041)**

Sheep and goat meat, -0.354 0.237 0.074 -0.850 0.759 -0.387 -0.386 0.161 -0.364 1.508 fresh and frozen (0.123)** (0.149) (0.071) (0.153)** (0.160)** (0.081)** (0.129)** (0.094) (0.126)** (0.110)**

Poultry, fresh or 0.122 -0.110 -0.096 0.997 -1.344 1.204 -0.986 0.009 0.086 0.043 frozen, as well as (0.057)* (0.069) (0.067) (0.122)** (0.087)** (0.128)** (0.085)** (0.079) (0.060) (0.060) grilled and smoked

Sausages, sausage -0.315 0.239 0.548 -0.762 0.723 -1.541 0.000 -0.024 -0.320 1.009 Meat and milk demand elasticities demand milk and Meat products and pasties (0.038)** (0.035)** (0.049)** (0.071)** (0.051)** (0.087)** (0.057) (0.057) (0.038)** (0.035)**

Pork: ham, bacon and remaining parts, 0.002 -0.092 -0.408 -1.208 0.000 -0.207 -0.166 0.238 0.297 1.374 cooked, dried, salted (0.052) (0.047) (0.064)** (0.107)** (0.071) (0.090)* (0.085)* (0.079)** (0.051)** (0.045)** or smoked

Other meat, cooked, -0.183 0.178 -0.201 -0.382 -0.201 -0.443 -0.027 -0.863 0.027 1.227 dried, salted or (0.097)* (0.095) (0.107) (0.254) (0.128) (0.122)** (0.115) (0.169)** (0.091) (0.107)** smoked

Other meat (e.g. game and rabbit, 0.054 0.158 -0.329 0.993 -0.222 -0.503 -0.108 0.316 -0.426 0.778

horse, preserved (0.059) (0.066)* (0.067)** (0.105)** (0.091)* (0.071)** (0.079) (0.084)** (0.086)** (0.051)**

meat)

* and ** denote significance at 5% and 1% level, respectively.

Meat and milk demand elasticities

Table 12: Conditional, Hicksian price and income elasticities at stage 3, meat (at sample means)

Be Ve Po SG Pl SSP PHB OMp OM Income

-0.404 0.013 -0.248 0.584 -0.076 -0.124 -0.044 0.100 0.163 0.685 Beef, fresh or frozen (0.066)** (0.063) (0.055)** (0.087)** (0.081) (0.066) (0.078) (0.068) (0.048)** (0.050)**

-0.366 -0.878 0.320 -2.139 0.283 0.565 0.898 -0.031 0.007 2.066 Veal, fresh or frozen (0.088)** (0.090)** (0.086)** (0.298)** (0.119)* (0.145)** (0.115)** (0.102) (0.081) (0.098)**

-0.098 0.160 -0.947 0.803 -0.114 0.916 -0.044 -0.156 -0.168 0.913 Pork, fresh or frozen (0.052) (0.055)** (0.091)** (0.120)** (0.079) (0.061)** (0.065) (0.074)* (0.053)** (0.041)**

Sheep and goat -0.186 0.299 0.247 -0.810 0.989 0.000 -0.145 0.238 -0.235 1.508 meat, fresh and (0.116) (0.150)* (0.073)** (0.152)** (0.166)** (0.073) (0.141) (0.093)** (0.121) (0.110)** frozen

Poultry, fresh or 0.127 -0.108 -0.091 0.998 -1.338 1.215 -0.979 0.011 0.090 0.043 frozen, as well as (0.054)* (0.070) (0.066) (0.121)** (0.084)** (0.126)** (0.091)** (0.079) (0.058) (0.060) grilled and smoked

Sausages, sausage -0.203 0.281 0.663 -0.736 0.877 -1.281 0.162 0.028 -0.233 1.009 products and pasties (0.036)** (0.035)** (0.049)** (0.070)** (0.052)** (0.088)** (0.059)** (0.057) (0.037)** (0.035)**

Pork: ham, bacon and remaining parts, 0.155 -0.035 -0.251 -1.172 0.210 0.146 0.053 0.309 0.415 1.374 cooked, dried, salted (0.050)** (0.047) (0.064)** (0.106)** (0.072)** (0.091) (0.086) (0.079)** (0.050)** (0.045)** or smoked

Other meat, cooked, -0.047 0.229 -0.061 -0.350 -0.014 -0.128 0.169 -0.800 0.132 1.227 dried, salted or (0.093) (0.096)* (0.109) (0.252) (0.131) (0.111) (0.124) (0.170)** (0.090) (0.107)** smoked

Other meat (e.g. game and rabbit, 0.140 0.190 -0.240 1.013 -0.103 -0.303 0.016 0.356 -0.360 0.778 horse, preserved (0.057)* (0.066)** (0.067)** (0.104)** (0.092) (0.068)** (0.084) (0.084)** (0.086)** (0.051)** meat)

* and ** denote significance at 5% and 1% level, respectively.

Chapter III

Table 13: Conditional, Marshallian price and income elasticities at stage 3, milk products (at sample means)

Wm MS Ch Cr CY Eg Income

-1.315 0.347 0.962 0.162 0.160 -0.690 0.369 Whole milk (0.169)** (0.121)** (0.214)** (0.091) (0.136) (0.193)** (0.083)**

Milk drink and skim -0.228 -0.439 0.452 -0.193 -0.332 -0.846 1.065 milk (0.096)* (0.123)** (0.175)** (0.061)** (0.111)** (0.154)** (0.091)**

Cream, soft, medium- hard, hard, extra hard -0.159 0.119 -1.336 0.117 0.399 -0.091 1.053 and processed (0.046)** (0.042)** (0.057)** (0.029)** (0.041)** (0.061) (0.018)** cheese

0.357 -0.246 0.358 -0.989 -0.501 1.097 1.026 Cream (0.112)** (0.089)** (0.125)** (0.095)** (0.104)** (0.184)** (0.052)**

Curd, yoghurt, other 0.467 -0.756 -0.429 -0.063 -1.603 -0.071 1.648 milk products and milk (0.102)** (0.090)** (0.134)** (0.070) (0.108)** (0.148) (0.044)**

surrogates elasticities demand milk and Meat

Eggs, fresh and 0.124 0.132 0.751 -0.499 -0.248 0.020 0.197 processed (0.173) (0.148) (0.265)** (0.106)** (0.210) (0.319) (0.107)

* and ** denote significance at 5% and 1% level, respectively.

Meat and milk demand elasticities

Table 14: Conditional, Hicksian price and income elasticities at stage 3, milk products (at sample means)

Wm MS Ch Cr CY Eg Income

-1.277 0.372 1.122 0.190 0.242 -0.654 0.369 Whole milk (0.167)** (0.118)** (0.189)** (0.093)* (0.149) (0.197)** (0.083)**

Milk drink and skim -0.116 -0.366 0.912 -0.111 -0.096 -0.745 1.065 milk (0.096) (0.118)** (0.145)** (0.065) (0.121) (0.156)** (0.091)**

Cream, soft, medium- hard, hard, extra hard -0.048 0.190 -0.881 0.199 0.633 0.010 1.053 and processed (0.046) (0.041)** (0.057)** (0.029)** (0.041)** (0.060) (0.018)** cheese

0.465 -0.177 0.801 -0.910 -0.274 1.195 1.026 Cream (0.112)** (0.087)* (0.117)** (0.095)** (0.104)** (0.186)** (0.052)**

Curd, yoghurt, other 0.641 -0.643 0.282 0.064 -1.237 0.086 1.648 milk products and (0.101)** (0.089)** (0.130)* (0.070) (0.110)** (0.147) (0.044)** milk surrogates

Eggs, fresh and 0.145 0.146 0.836 -0.484 -0.205 0.039 0.197 processed (0.170) (0.144) (0.231)** (0.109)** (0.229) (0.318) (0.107)

* and ** denote significance at 5% and 1% level, respectively.

Chapter III

Table 15: Parameter estimates of the budget share equation, Stage 1

Coeff. for equation beverages Coeff. for equation other products

Coeff. for equation food (robust products (robust standard errors in and services (robust standard errors

Independent variables standard errors in parentheses) parentheses) in parentheses)

-0.200 (0.355) -0.098 (0.186) 1.298 (0.403)** Constant

0.076 (0.031)* 0.022 (0.013) -0.098 (0.034)** Logprice food

0.022 (0.013) -0.011 (0.014) -0.011 (0.012) Logprice beverages

-0.098 (0.034)** -0.011 (0.012) 0.109 (0.040)** Logprice other products and services

-0.047 (0.001)** -0.004 (0.001)** 0.051 (0.001)** Log (expenditure/a(p))

0.000 (0.001) -0.001 (0.000) 0.001 (0.001) (Log (expenditure/a(p)))^2

0.043 (0.001)** 0.002 (0.000)** -0.045 (0.001)** Household size in adult equivalents

0.005 (0.001)** -0.001 (0.000)* -0.004 (0.001)** Young children (<=5 years, yes or no)

0.001 (0.000)** 0.000 (0.000)** -0.001 (0.000)** Age of the household‘s reference person

University degree of the household‘s reference person -0.001 (0.001) -0.001 (0.000)** 0.002 (0.001)* (yes or no)

0.001 (0.000)** 0.000 (0.000)** elasticities demand milk and Meat -0.001 (0.000)** Month

0.000 (0.000) 0.000 (0.000) -0.000 (0.000) Year

0.011 (0.001)** 0.004 (0.001)** -0.015 (0.001)*** Residuals

0.85 0.43 2 -- R

110,000 -test statstic for the whole demand system (Wald test)

* and ** denote significance at 5% and 1% level, respectively.

Meat and milk demand elasticities

Table 16: Parameter estimates of the budget share equation, Stage 2

Coeff. for equation milk products

(robust standard errors in Coeff. for equation other food (robust Coeff. for equation meat (robust

Independent variables parentheses) standard errors in parentheses) standard errors in parentheses)

Constant -3.370 (1.272)** 1.883 (1.939) 2.487 (1.756)

Logprice milk 0.514 (0.164)** -0.117 (0.120) -0.398 (0.139**

Logprice other food -0.117 (0.120) 0.154 (0.070)* -0.037 (0.106)

Logprice meat -0.398 (0.139)** -0.037 (0.106) 0.435 (0.182)*

Log (expenditure/a(p)) -0.099 (0.015)** 0.010 (0.017) 0.089 (0.01)**

(Log (expenditure/a(p)))^2 -0.008 (0.002)** 0.004 (0.002) 0.004 (0.001)**

Household size in adult equivalents 0.029 (0.004)** -0.022 (0.006)** -0.007 (0.006)

Young children (<=5 years, yes or no) 0.012 (0.002)** 0.010 (0.003)** -0.022 (0.003)**

Age of the household‘s reference person 0.000 (0.000)* -0.001 (0.000)** 0.000 (0.000)**

University degree of the household‘s reference person 0.002 (0.002) 0.041 (0.003)** -0.043 (0.002)** (yes or no)

Month -0.000 (0.000)** 0.001 (0.000)** -0.000 (0.000)

Year 0.002 (0.001)** -0.001 (0.001) -0.001 (0.001)

Residuals 0.008 (0.005) -0.006 (0.008) -0.003 (0.008)

R2 0.83 0.96 --

-test statstic for the whole demand system (Wald test) 880,000**

* and ** denote significance at 5% and 1% level, respectively.

Chapter III Chapter

Table 17: Parameter estimates of the budget share equation, Stage 3: Meat

Coeff. for equation

Coeff. for equation pork: ham, bacon Coeff. for equation

Coeff. for equation Coeff. for equation chicken, fresh or Coeff. for equation and remaining parts, other meat, cooked, Coeff. for

beef, fresh or frozen Coeff. for equation Coeff. for equation sheep and goat meat, frozen, as well as sausage products and cooked, dried, salted dried, salted or equation other

(robust standard veal, fresh or frozen pork, fresh or frozen fresh and frozen grilled and smoked pasties (robust or smoked (robust smoked (robust meat (robust

errors in (robust standard (robust standard errors (robust standard (robust standard standard errors in standard errors in standard errors in standard errors

Independent variables parentheses) errors in parentheses) in parentheses) errors in parentheses) errors in parentheses) parentheses) parentheses) parentheses) in parentheses)

Constant 0.363 (0.071)** -0.220 (0.124) 0.039 (0.074) 0.814 (0.130)** 0.076 (0.145) 0.387 (0.126)** -1.007 (0.091)** 0.034 (0.087) 0.359 (0.067)**

Logprice beef, fresh or 0.099 (0.018)** 0.019 (0.018) -0.083 (0.012)** 0.089 (0.016)** -0.065 (0.020)** -0.091 (0.010)** 0.036 (0.012)** 0.001 (0.015) -0.005 (0.010) frozen

Logprice veal, fresh or 0.019 (0.018) -0.044 (0.046) 0.055 (0.014)** -0.070 (0.021)** 0.009 (0.044) 0.066 (0.011)** -0.064 (0.018)** 0.012 (0.017) 0.016 (0.015) frozen

Logprice pork, fresh or -0.083 (0.012)** 0.055 (0.014)** -0.024 (0.022) 0.025 (0.017) -0.023 (0.019) 0.181 (0.015)** -0.030 (0.014)* -0.033 (0.018) -0.069 (0.013)** frozen

Logprice sheep and goat 0.089 (0.016)** -0.070 (0.021)** 0.025 (0.017) -0.032 (0.029) 0.035 (0.026) -0.080 (0.019)** -0.134 (0.020)** 0.083 (0.021)** 0.082 (0.016)** meat, fresh and frozen

Logprice chicken, fresh

or frozen, as well as -0.065 (0.020)** 0.009 (0.044) -0.023 (0.019) 0.035 (0.026) -0.091 (0.052) 0.205 (0.016)** 0.018 (0.023) -0.017 (0.023) -0.071 (0.019)**

grilled and smoked

Logprice sausage -0.091 (0.010)** 0.066 (0.011)** 0.181 (0.015)** -0.080 (0.019)** 0.205 (0.016)** -0.172 (0.027)** 0.029 (0.017) -0.042 (0.017)* -0.096 (0.011)** products and pasties

Logprice pork: ham,

bacon and remaining 0.036 (0.012)** -0.064 (0.018)** -0.030 (0.014)* -0.134 (0.020)** 0.018 (0.023) 0.029 (0.017) 0.109 (0.020)** -0.007 (0.017) 0.043 (0.011)** parts, cooked, dried,

salted or smoked

Logprice other meat,

cooked, dried, salted or 0.001 (0.015) 0.012 (0.017) -0.033 (0.018) 0.083 (0.021)** -0.017 (0.023) -0.042 (0.017)* -0.007 (0.017) -0.002 (0.028) 0.004 (0.015)

smoked

Coeff. for equation

Coeff. for equation pork: ham, bacon Coeff. for equation

Coeff. for equation Coeff. for equation chicken, fresh or Coeff. for equation and remaining parts, other meat, cooked, Coeff. for

beef, fresh or frozen Coeff. for equation Coeff. for equation sheep and goat meat, frozen, as well as sausage products and cooked, dried, salted dried, salted or equation other

(robust standard veal, fresh or frozen pork, fresh or frozen fresh and frozen grilled and smoked pasties (robust or smoked (robust smoked (robust meat (robust

errors in (robust standard (robust standard errors (robust standard (robust standard standard errors in standard errors in standard errors in standard errors

Independent variables parentheses) errors in parentheses) in parentheses) errors in parentheses) errors in parentheses) parentheses) parentheses) parentheses) in parentheses)

Logprice other meat -0.005 (0.010) 0.016 (0.015) -0.069 (0.013)** 0.082 (0.016)** -0.071 (0.019)** -0.096 (0.011)** 0.043 (0.011)** 0.004 (0.015) 0.095 (0.017)**

Log (expenditure/a(p)) -0.046 (0.017)** 0.341 (0.032)** -0.025 (0.014) 0.098 (0.030)** -0.306 (0.017)** 0.023 (0.013) 0.079 (0.014)** 0.026 (0.021) -0.043 (0.014)**

(Log 0.015 (0.005)** 0.030 (0.008)** -0.002 (0.004) -0.010 (0.006) -0.031 (0.005)** 0.010 (0.004)* -0.001 (0.004) -0.005 (0.005) -0.001 (0.004) (expenditure/a(p)))^2

Household size in adult 0.047 (0.010)** -0.196 (0.021)** -0.022 (0.010) -0.116 (0.021)** 0.168 (0.009)** 0.072 (0.009)** -0.016 (0.009) -0.108 (0.013)** 0.049 (0.009)** equivalents

Young children (<=5 -0.017 (0.007)* 0.058 (0.011)** -0.005 (0.007) 0.023 (0.015) -0.028 (0.006)** 0.034 (0.005)** -0.012 (0.005)** -0.023 (0.007)** -0.014 (0.006)* years, yes or no)

Age of the household‘s 0.001 (0.000)** 0.000 (0.000) 0.000 (0.000) -0.001 (0.000) 0.000 (0.000) 0.001 (0.000)** 0.000 (0.000) -0.002 (0.000)** 0.000 (0.000)* reference person

Highest education of the

household‘s reference 0.012 (0.007) 0.059 (0.015)** 0.020 (0.006)** -0.067 (0.005)** -0.004 (0.005) 0.037 (0.007)** -0.021 (0.006)** 0.005 (0.010) -0.012 (0.006) person

Month 0.001 (0.001) -0.001 (0.001) -0.001 (0.001) 0.000 (0.001) -0.003 (0.001)** 0.003 (0.000)** 0.001 (0.001) 0.000 (0.001) 0.002 (0.001)** Meat and milk demand elasticities demand milk and Meat Year 0.000 (0.000)** 0.001 (0.000)** 0.000 (0.000) 0.000 (0.000) 0.000 (0.000)** 0.000 (0.000)** 0.001 (0.000)** 0.000 (0.000)* 0.000 (0.000)**

Residuals 0.157 (0.011)** -0.197 (0.026)** 0.095 (0.010)** -0.091 (0.024)** 0.225 (0.009)** -0.064 (0.009)** -0.118 (0.010)** -0.080 (0.017)** 0.062 (0.010)**

pdf 0.160 (0.019)** 0.169 (0.021)** 0.004 (0.014) 0.166 (0.022)** 0.192 (0.024)** 0.652 (0.013)** 0.525 (0.013)** 0.126 (0.015)** 0.249 (0.015)**

R2 0.33 0.15 0.36 0.09 0.40 0.59 0.46 0.17 0.26

-test statstic for the

whole demand system 710,000

(Wald test)

* and ** denote significance at 5% and 1% level, respectively.

100 Chapter III Chapter

Table 18: Parameter estimates of the budget share equation, Stage 3: Milk products

Coeff. for equation

curd, yoghurt, other milk

Coeff. for equation milk drink products and milk

and skim milk (robust Coeff. for equation cheese Coeff. for equation cream surrogates (robust Coeff. for equation

Coeff. for equation whole milk (robust standard errors in (robust standard errors in (robust standard errors in standard errors in eggs (robust standard

Independent variables standard errors in parentheses) parentheses) parentheses) parentheses) parentheses) errors in parentheses)

Constant 0.920 (0.461)* 1.085 (0.223)** 2.362 (0.243)** -0.274 (0.152) -2.345 (0.609)** -0.299 (0.504)

Logprice whole milk -0.163 (0.043)** -0.017 (0.021) -0.055 (0.023)* 0.054 (0.014)** 0.250 (0.052)** -0.069 (0.042)

Logprice milk drink and -0.017 (0.021) 0.109 (0.013)** 0.076 (0.016)** -0.024 (0.008)** -0.071 (0.029)* -0.073 (0.022)** skim milk

Logprice cheese -0.055 (0.023)* 0.076 (0.016)** -0.100 (0.032)** 0.045 (0.014)** 0.136 (0.024)** -0.102 (0.028)**

Logprice cream 0.054 (0.014)** -0.024 (0.008)** 0.045 (0.014)** 0.026 (0.010)* -0.070 (0.017)** -0.030 (0.016)

Logprice curd, yoghurt,

other milk products and 0.250 (0.052)** -0.071 (0.029)* 0.136 (0.024)** -0.070 (0.017)** -0.404 (0.073)** 0.158 (0.051)**

milk surrogates

-0.069 (0.042) -0.073 (0.022)** -0.102 (0.028)** -0.030 (0.016) 0.158 (0.051)** 0.117 (0.048)* Logprice eggs

Log (expenditure/a(p)) -0.085 (0.020)** 0.017 (0.021) 0.021 (0.014) -0.001 (0.011) 0.171 (0.018)** -0.130 (0.017)**

(Log 0.003 (0.002) 0.001 (0.002) 0.000 (0.001) 0.000 (0.001) 0.002 (0.002) -0.003 (0.001)* (expenditure/a(p)))^2

Household size in adult 0.120 (0.006)** 0.001 (0.010) -0.044 (0.005)** -0.006 (0.005) -0.091 (0.007)** 0.068 (0.005)** equivalents

Young children (<=5 0.055 (0.004)** 0.011 (0.005)* -0.033 (0.004)** -0.015 (0.002)** 0.008 (0.004)* -0.002 (0.003) years, yes or no)

Age of the household‘s 0.001 (0.000)** 0.001 (0.000)** 0.000 (0.000) 0.001 (0.000)** -0.003 (0.000)** 0.001 (0.000)** reference person

Highest education of the

household‘s reference -0.007 (0.003)* -0.014 (0.004)** 0.013 (0.003)** -0.003 (0.002) -0.005 (0.004) 0.013 (0.003)**

person

Coeff. for equation curd, yoghurt, other milk

Coeff. for equation milk drink products and milk

and skim milk (robust Coeff. for equation cheese Coeff. for equation cream surrogates (robust Coeff. for equation

Coeff. for equation whole milk (robust standard errors in (robust standard errors in (robust standard errors in standard errors in eggs (robust standard

Independent variables standard errors in parentheses) parentheses) parentheses) parentheses) parentheses) errors in parentheses)

Month 0.000 (0.000) -0.001 (0.000) 0.002 (0.000)** 0.001 (0.000)** -0.003 (0.000)** 0.000 (0.000)

Year -0.001 (0.000)** 0.000 (0.000)** -0.001 (0.000)** 0.000 (0.000)** 0.001 (0.000)** 0.000 (0.000)

Residuals 0.154 (0.007)** 0.105 (0.009)** 0.316 (0.016)** 0.013 (0.005)* 0.545 (0.008)** 0.087 (0.010)**

pdf 0.090 (0.007)** -0.022 (0.013) 0.034 (0.007)** 0.000 (0.006) -0.183 (0.010)** 0.084 (0.006)**

R2 0.37 0.25 0.81 0.40 0.62 0.45

-test statstic for the

whole demand system 1,900,000**

(Wald test)

* and ** denote significance at 5% and 1% level, respectively.

Meat and milk demand elasticities demand milk and Meat

101

Chapter IV

Chapter IV Chapter Chapter IV

Consumer demand for alcoholic beverages in Switzerland:

A two-stage Quadratic Almost Ideal Demand System for low, moderate and heavy drinking households

Matteo Aepli

ETH Zurich

Manuscript under review, Agricultural and Food Economics

102

Consumer demand for alcoholic beverages

Abstract

In this study, we estimate final demand for beverages with a particular focus on alcoholic beverages and calculate elasticities using microdata from the Swiss household expenditure survey from 2000 to 2009, which contains data from more than 34,000 households. We estimate price and income responses for three household segments, light, moderate, and heavy drinking households, to assess whether higher alcohol consumption could be described by different price and income elasticities in comparison to lower alcohol consumption. We obtain unconditional estimates by applying a two-stage budgeting quadratic almost ideal demand system. To generate missing price data, we used the recently proposed quality adjusted price approach by Aepli and Finger (2013) based on the approach of Majumder et al. (2012). Due to a high share of zero consumption for some beverages categories, we correct the model with Shonkwiler and Yen‘s (1999) two-step estimation procedure. Estimation results show that heavy drinking households are much less price elastic with respect to wine and beer in comparison to moderate or light drinking households, while the price response for spirits is almost constant over the three segments. Before implementing a new tax for alcoholic beverages in Switzerland, the social, health, and economic effects of a rather small decrease in alcohol consumption among heavy drinking households must be weighed against possible negative consequences of a sharp decline in light or moderate drinking households.

103

Chapter IV

Chapter IV Chapter 1 Introduction

Knowledge about the determinants of consumption of alcoholic beverages and the price and income elasticity of different consumer segments is highly policy-relevant. Alcohol consumption is a public health priority. Following for example Wakabayashi (2013) and Corrao et al. (2000), a regular but moderate consumption of alcohol in general can have a positive effect on health by increasing the level of HDL cholesterol and reducing the risk of heart and vascular diseases. Other studies do not report positive health effects of low or moderate alcohol consumption (Estruch et al., 2014; WHO, 2007). The discussion is complex and still ongoing. However, there is consensus that excessive alcohol consumption can be detrimental to health in that it can lead to liver cancer, liver disease, raised blood pressure, stroke, or mental decline (Rehn et al., 2010; Schwartz et al., 2013; Sabia et al., 2014). In addition, excessive alcohol consumption can also have negative psychological, behavioral consequences. It can lead to higher rates of crime or violence (Jacobs & Steyn, 2013; Fergusson et al., 2013) and have negative economic effects e.g. leading to substantial higher state costs (Sacks et al., 2013).

Policy measures have been shown to reduce alcohol consumption and to be a cost effective health care intervention (Xuan et al., 2013; Doran et al., 2013). Nevertheless, alcohol taxation and its effectiveness is still a controversial discussion point in Switzerland and other European countries (to get an overview of the different alcohol policy regulations in Europe see e.g. FOPH, 2014a). While total alcohol consumption in Switzerland has been slightly decreasing since 1990, with a current per capita consumption of about 8.5 liter of pure alcohol per year (SAB, 2013), excessive forms of alcohol drinking, such as binge drinking, have been increasing, especially among young people (FOPH, 2013; Annaheim & Gmel, 2004). Therefore, Switzerland is currently discussing a tightening of the alcohol legislation. While wine is excluded from the alcohol tax, beer and spirits are heavily taxed. The current legislation has the

104

Consumer demand for alcoholic beverages

purpose of a positive health effect on one hand and a fiscal benefit with annual tax revenue of about 440 million Swiss francs (FDF, 2009).

In order to assess the effect of a new tax on consumption, especially in the case of heavy drinkers, the estimation of price elasticities is crucial. However, price policies may be superfluous if changing income level affect consumption (Aepli & Finger, 2013; Gallet, 2010). Therefore, an estimation of income elasticities as a supplement to price elasticities is necessary. An additional tax burden to reduce the light or moderate drinkers‘ consumption level may not be desirable for example with respect to social aspects but also concerning health effects depending on which assumptions were made with respect to health effects of low or moderate alcohol consumption. In the case of heavy drinkers it is clear: A reduction in consumption would reduce public health costs and abate social problems. To address the question of how price and income affect demand for alcoholic beverages in Switzerland, we estimate final demand price and income elasticities separately for light, moderate, and heavy drinking households based on a repeated cross-sectional household expenditure survey. We combine several recently used methods in one demand system, expecting that price and income responses are not constant among the three segments. Switzerland is a particularly interesting case due to its high purchasing power. Findings from this case can also be transferred to a considerable extent to other European countries.

Elasticity estimates based on household data are scarce. Most studies rely on time- series data (Gallet, 2007) which only allows to estimate elasticities for different household segments in a very limited way. Furthermore, most of them did not investigate in possible substitution effects between alcoholic and non-alcoholic beverages. Our study contributes to fill this gap and allows to a more detailed understanding of alcohol demand with respect to the demand response of different household segments.

105

Chapter IV

Chapter IV Chapter We apply a two-stage quadratic almost ideal demand system (QUAIDS) (Banks et al., 1999 [for application see e.g. Abdulai, 2002; Jithitikulchai, 2011]). It uses cross- sectional data from the Swiss household expenditure survey from 2000 to 2009 that contains data from more than 34,000 households. Final demand elasticities are estimated for food, beverages and other products and services at the first stage and for wine, beer, spirits, and several non-alcoholic beverages at the second stage. Due to a high zero consumption for some product groups, we modified the QUAIDS using Shonkwiler and Yen‘s (1999) approach (for application see e.g. Thiele, 2008; Tafere et al., 2010) and corrected for possible heteroscedasticity by applying a parametric bootstrap. We received missing price data using a recently proposed method by Majumder et al. (2012), further developed by Aepli and Finger (2013) and introducing a variable for seasonality and trend applied by Aepli and Kuhlgatz (2014) for the same expenditure survey. Furthermore, we corrected for endogeneity of the expenditure variable in the model by using the augmented regression technique.

For Switzerland, Heeb et al. (2003) analyzed the effect of a price reduction on spirits in 1999. The reduction was due to a tariff reduction under the WTO agreement and Heeb et al. focused on heavy drinkers using a longitudinal survey. Overall, spirits are rated as inelastic, whereas low volume drinkers are more elastic than high volume drinkers are. This would mean that taxes would reduce light or moderate drinkers‘ level of consumption more than the heavy drinkers‘ levels (Manning et al., 1995). Kuo et al. (2003) used the same period as Heeb et al. (2003) with the price reduction due to the WTO agreement to estimate demand reaction to the price decrease. They found that young people in particular responded with a higher demand, whereas people who are aged 60 or older did not respond at all.

The paper is structured as follows. In section 2, we describe the theoretical framework and the two-stage budgeting system, and in section 3, we present the descriptive statistics and indicate how price data are generated. Section 4 summarizes the

106

Consumer demand for alcoholic beverages

estimation procedure and the elasticity calculation. The income and price responses are discussed in section 5, and section 6 concludes.

2 Theoretical framework and two-stage budgeting

2.1 Theoretical framework of the QUAIDS

Equation systems have been widely used in demand analysis. The most frequently applied models are as follows: the almost ideal demand system (AIDS) or the linearized AIDS (Deaton & Muellbauer, 1980a; 1980b; Akbay et al., 2007; Mhurchu et al., 2013); the quadratic version of the AIDS (QUAIDS) (Banks et al., 1997; Abdulai, 2002; Dey et al., 2011 or Stasi et al., 2010 and Cembalo et al., 2014 in particular for wine); the Rotterdam model (Barten, 1964; Theil, 1965; Barnett & Seck, 2008; Barnett & Kanyama, 2013); the Translog (Christensen et al., 1975; Holt & Goodwin, 2009); and the Linear and Quadratic expenditure system (Pollak & Wales, 1978; De Boer & Paap, 2009). One of the most important criteria is the approximation of non-linear Engel curves, which is best satisfied by the QUAIDS and allows general income responses that are not captured by the AIDS or many other models. The QUAIDS model is a rank three demand system and it satisfies the axioms of choice. It allows exact aggregation over consumers due to underlying preferences that are of the generalized Gorman polar form (Banks et al., 1997; Blackborby et al., 1978). The recently proposed EASI demand system (Lewbel & Pendakur, 2009, for application see e.g. Stasi et al., 2011) would have been an alternative due to its flexibility with respect to the approximation of Engel curves. But Cranfield et al. (2013) mentioned that the QUAIDS is especially suitable in the case of a disaggregated analysis as in our case. Furthermore, the QUAIDS has already been successfully applied to the Swiss household expenditure survey (see e.g. Abdulai, 2002).

Following Banks et al. (1997) the indirect utility function of the QUAIDS is given by:

( ) (* + ( )) (1) ( ) 107

Chapter IV

Chapter IV Chapter Where is total expenditure and are prices. ( ) is the translog price aggregator ( ) and ( ) is the Cobb-Douglas price aggregator. The term is the indirect

( ) utility function of a PIGLOG demand system and ( ) is a differentiable, homogeneous function of degree zero of prices. ( ) and ( ) are defined as follows:

( ) ∑ ∑ ∑ (2)

( ) ∏ (3)

Where and are specific goods and is the number of goods.

The Marshallian demand function in budget shares is obtained by applying Roy‘s identity to the indirect utility function:

∑ * + , * +- (4) ( ) ( ) ( )

Where is the budget share for product category , and , , , and are parameters to be estimated. The residuals are assumed to be multivariate normal distributed with zero mean and a finite variance-covariance matrix.

To meet utility maximization theory, the restriction of adding-up (5), of homogeneity of the Marshallian cost function in prices and total expenditure (6) and of symmetry of the Slutsky matrix (Young‘s theorem) (7) are implemented:

Adding-up: ∑ ; ∑ ; ∑ ; ∑ . (5)

Homogeneity: ∑ (6)

Symmetry: (7)

Theoretical restrictions allow us to reduce the number of estimated parameters as well as improve the efficiency of the estimated model (Barnett & Seck, 2008). To avoid singularity in the variance-covariance matrix and to satisfy the condition of adding-up,

108

Consumer demand for alcoholic beverages

one equation is dropped and an n-1 equation system is estimated. The parameters of the th equation are obtained from the restriction and the parameters of the n-1 equations.

In addition to the economic determinants, we take household characteristics and a variable for seasonality and time trend into account using the demographic translation approach by Pollak and Wales (1978) (e.g. applied by Abdulai, 2002; Bopape, 2007), modifying the intercept in equation (4) to

∑ , (8)

Where is the fth variable of a total number of variables including household characteristics and a variable for month and year.

2.2 Two-stage budgeting design

To obtain unconditional elasticities for alcoholic beverages, which are more suitable for policy recommendations than conditional elasticities (Abdulai, 2002; Klonaris & Hallam, 2003), we adopt a two-stage budgeting system, assuming separability of the utility function (for further explanations see e.g. Moschini et al., 1994 or Brehe, 2007).12 First, we assume that the household allocates its budget on the three aggregated product groups at stage 1: food, beverages, and other products and services. At stage 2, the total expenditure for beverages is allocated to the disaggregated product categories: wine, beer, spirits, coffee, tea, cocoa beverages, mineral water, non-alcoholic soft drinks, and fruit and vegetables juices. We estimate income and price response at both stages, while the elasticity at stage 2 is conditional

12 After Klonaris and Hallam (2003) conditional elasticities does only contain direct effects on demand in comparison to unconditional elasticities which contain direct and indirect effects. The latter are therefore more relevant in welfare analysis and for policy purposes (Klonaris and Hallam 2003). The reason lies in the multistage budgeting. A change in the price for one beverage category at the second stage within beverages (first stage) has a direct effect on the demand of all beverage categories (second stage) but has also an effect on the price index for beverages (first stage) and therefore will influence the allocation at the first stage (beverages and other commodity groups). This could have a further indirect effect on the demand for the beverage categories at the second stage. 109

Chapter IV

Chapter IV Chapter with respect to total expenditure for beverages. Unconditional elasticities for stage 2

could be obtained using the elasticity estimates at stage 1 (see section 4.3).

3 Data description and price computation

3.1 Summary statistics and definition of the household segments

For our study, we used data from the Swiss household expenditure survey from 2000 to 2009 collected by the Swiss Federal Statistic Office, a national representative survey that consists of cross-sectional data with a periodicity of one month. Almost 3,000 households participate in the survey every year. We can base our estimates on more than 34,000 households over a period of 10 years. Households are chosen randomly from the register or private telephone lines (see SFSO, 2011 for details). The sample is stratified with respect to the seven major regions in Switzerland and the distribution of households is calibrated to the distribution of households in the Swiss population (SFSO, 2011). For every household, data on expenditure and quantity bought (in liter per month) for all products and services, household income, and detailed information on household characteristics were gathered. To increase the explanatory power of the model and to test whether household characteristics have an influence on alcohol demand, we introduced the following criteria into the model: household size in terms of adult equivalents, a dummy variable for the presence of young children (<5 years), the age of the household‘s reference person, and a dummy variable for whether the household‘s reference person has a university degree. Household size is calculated following the OECD-modified equivalence scale (Hagenaars et al., 1994). Following Angulo et al. (2001) we tend to expect a decreasing expenditure share as the number of equivalence increase and the age of the household‘s reference person decrease. With respect to education and the presence of young children there are some contradictory or no clear findings in the literature (see e.g. Van Oers et al., 1999; Reavley et al., 2011; Rice et al., 1998).

110

Consumer demand for alcoholic beverages

The sample is divided into three segments: light drinking households, moderate drinking households, and heavy drinking households. We provide elasticities for the whole sample as well as for the three household segments. The classification is constructed based on the recommendations for Switzerland in Annaheim and Gmel (2004). These are the official recommendations for Switzerland from the Swiss institute for alcohol and drug problems. Table 1 summarizes the class boundaries in pure alcohol per person and day.

Table 1: Class boundaries in pure alcohol per person and day

Pure alcohol (in g per day)

light drinkers <20

moderate drinkers 20-39

heavy drinkers >40

According to FOPH (2014b), we assume the following levels of pure alcohol for each beverage category: 86.9 g pure alcohol per litre for wine, 37.9 g pure alcohol per litre for beer, and 316.0 g pure alcohol per litre for spirits. We calculated the amount of pure alcohol bought per month for every household based on the data of the household expenditure survey. We divided the total amount of pure alcohol by the adult equivalent for every household13 and classified the households according to the class boundaries (Table 1) based on the Swiss recommendations in Annaheim and Gmel (2004).

Tables 2 to 4 present summary statistics for stage 1, stage 2, and the household characteristics considered in the model. Furthermore, we provide the share of zero consumption for every product group. Given that some households did not buy

13 Adult equivalent is a better proxy for the number of alcohol consuming people in a household than the number of adults because of the relatively high alcohol consumption among young people in Switzerland (15 years or younger) (Annaheim and Gmel 2004). 111

Chapter IV

Chapter IV Chapter beverages during the collection period, the sample for stage 2 reduces to 33,364

households.

Table 2: Summary statistics for the stage 1 model, total number of households and expenditure per month

Total number of households 34,176

Percentage of zero Mean Standard deviation consumption

Total Expenditure (in Swiss Francs) 8,432.70 5,132.24 0.00

Expenditure on food (in Swiss Francs) 616.63 354.78 0.14

Expenditure on beverages (in Swiss Francs) 121.63 186.97 2.38

All other expenditure (in Swiss Francs) 1395.42 4933.60 0.00

112

Table 3: Summary statistics for the stage 2 model, sample sizes (reduced samples)14 and expenditure per month

Percentage Standard Standard Standard Mean (only Standard of zero Mean (only deviation deviation (only Mean (only Mean (all deviation moderate deviation (only consumption light drinking (only light moderate heavy drinking households) (all drinking heavy drinking (all households) drinking drinking households) households) households) households) households) households) households)

33,364 25,699 4,152 3,513 Number of households 124.59 188.25 2.38 79.28 69.85 192.70 128.76 375.54 444.81 Expenditure on beverages (in Swiss Francs)

67.57 175.72 32.09 23.93 39.15 131.25 109.74 311.51 435.48 Expenditure on alcoholic beverages (in Swiss Francs) 57.02 53.95 1.61 55.35 53.30 61.45 53.57 64.03 58.19 Expenditure on non-alcoholic beverages (in Swiss Francs) Wine (in Swiss Francs) 48.31 162.57 49.10 15.57 34.37 91.17 114.12 237.16 425.85

Beer (in Swiss Francs) 8.23 20.55 69.52 4.36 11.18 17.23 26.65 25.87 41.73 beverages alcoholic for demand Consumer

Spirits, sweet wines, etc. (in Swiss Francs) 11.03 32.44 71.33 4.00 13.47 22.85 34.81 48.48 72.84

Coffee (in Swiss Francs) 14.46 31.21 44.66 13.49 30.03 17.23 31.81 18.32 37.81

Tea (in Swiss Francs) 3.87 8.83 59.92 3.91 9.09 3.88 7.76 3.57 8.05

Cocoa beverages (in Swiss Francs) 1.92 6.36 85.05 2.03 6.61 1.64 5.45 1.48 5.45

Mineral water (in Swiss Francs) 10.89 19.33 39.88 10.33 19.17 12.46 19.28 13.10 20.26

Non-alcoholic soft drinks (in Swiss Francs) 15.87 25.01 32.62 15.67 24.64 16.17 25.77 17.02 26.68

Fruit juices and vegetable juices (in Swiss 10.01 19.81 38.50 9.92 20.65 10.06 15.62 10.54 17.89 Francs)

14 The reduction in sample size in comparison to Table 1 arises due to the deletion of households without consumption of beverages during the data 113 collection period.

Chapter IV

Chapter IV IV Chapter Chapter

Table 4: Summary statistics for the household characteristics

Household characteristics Mean Standard deviation

1.63 0.52 Household size in adult equivalents

0.14 0.35 Young children (<5 years, yes or no)

49.69 15.22 Age of the household‘s reference person

0.13 1.27 University degree with respect to the household‘s reference person (yes or no)

As can be seen in Table 2, Swiss households spend only 1.4% of their budget on beverages, with 0.8% on alcoholic beverages and 0.7% on non-alcoholic beverages (Table 3). This is much lower than in the European Union with average values of about 1.5% and 1.2% of the household budget being spent on alcoholic and non-alcoholic beverages, respectively (Eurostat, 2013). Within alcoholic beverages, the highest share of the budget is spent on wine, followed by spirits and beer. The difference between the three household segments can be clearly noted in Table 3. Moderate drinking households spend on average four to five times more money on wine, beer, or spirits than light drinking households, whereas heavy drinking household spend about twice as much or more in comparison to moderate drinking households. On average, the light drinking households consume 49.3 g of pure alcohol per month, moderate drinking households consume 864.5 g, and heavy drinking households consume 1890.0 g.

A particular point of interest is the share of zero consumption. Zero consumption means that the left side of equation (4) (budget share) is left censored at zero in the case of demand systems. Besides a short collection period, which may play a particular role in the context of beverages, there are other determinants for zero consumption as e.g. income restrictions which force the household into a corner solution or missing preferences for certain products (e.g. see Perali & Chavas, 2000; Thiele, 2008). On the aggregated level (stage 1), zero consumption is low, but at the 114

Consumer demand for alcoholic beverages

stage 2 level, a large proportion of households did not consume during the collection period. This has serious implications on the model specification. Censoring of the expenditure data should therefore be considered, in particular at stage 2, to avoid biased parameter estimates.

3.2 Quality-adjusted prices

Only a few household expenditure surveys collect price data. In the literature, there are mainly two methods to retrieve price data from expenditure and quantity variables: using unit values as proxy for market prices, as done by Abdulai (2002) or Akbay et al. (2007) by dividing the expenditure by the quantity consumed, or quality adjusted unit values. The latter method was originally introduced by Cox and Wohlgenant (1986) and is frequently applied in the literature (e.g. in Lazaridis, 2003; Thiele, 2010; Zheng & Henneberry, 2010). In comparison to quality-adjusted prices, unit values do also contain quality effects. The reason lies in the heterogeneity of products within a product category (e.g. wine). The consumer can select between different quality levels. This leads to different unit values among households. Differences in unit values could therefore be attributed to price and quality variations (Chung et al., 2005). An increase in the unit value could be induced by a price increase or by a shift in the household‘s demand for more expensive products within this particular product category. Thus, taking unit values as proxies for market prices could lead to biased parameter estimates. This is particularly the case for composite commodities (Chung et al., 2005), such as product groups for beverages used in this study, which contain products of varying qualities. As McKelvey (2011) notes, unit values will be an ―error- ridden indicator of prices‖.

Cox and Wohlgenant (1986) proposed correcting the unit value for every household and product group by regressing proxies for quality variations (e.g. education or household size) on the unit value. This leads to quality-adjusted prices that vary between household. Majumder et al. (2012) criticized this approach stating that households should face the same price at least in the same regional market and 115

Chapter IV

Chapter IV IV Chapter Chapter proposed a new method to calculate quality adjusted prices per region. Aepli and

Finger (2013) expanded it by a variable for month and year and adapted it for the

Swiss household expenditure survey.

We give a short overview of the method (for further details we refer the reader to Majumder et al., 2012 and Aepli & Finger, 2013). As mentioned in Majumder et al. (2012), household characteristics and income are good proxies for quality preferences. For instance, households with less income tend to choose less expensive goods within a product group. A separate regression is run for each product category . Following Aepli and Finger (2013), the unit value and the proxy variables are related as follows:

( )

∑ (9)

Where is the unit value paid by household for item in its region , year , and month m. The deviation of the household‘s unit value from the median is explained by household characteristics, income, and expenditure variables, which capture the quality effect. The part of the deviation, which cannot be explained by quality, is

captured by the error term . Furthermore, x denotes income and the square of income, is the household total expenditure for beverages, and is the household total expenditure for food and beverages consumed away from home. denotes the th of household characteristics, which are household size, a binary dummy variables for having children, and a dummy variable for having a university degree. ,

and are dummies for region, year, and month, respectively. To manage possible outliers more successfully, we decided to follow the suggestions in Aepli and Finger (2013) and to estimate the equations using a robust M-estimator using only those households that consumed the good.

116

Consumer demand for alcoholic beverages

To obtain regional, monthly, and yearly quality adjusted market prices we add the median unit value to the corresponding median of item of the estimated residuals from equation (9).

( ) ( ) ( ) (10)

The prices are then assigned to all households in the sample according to region, month, and year. Due to a low number of households consuming tea and cocoa beverages in the Italian speaking part of Switzerland in some months, we took the corresponding consumer price index for those product categories as well as for all product categories at stage 1.

4 Estimation procedure and elasticity calculation

4.1 Censoring

As shown in Table 3, the share of zero consumption is relatively high for all beverage product categories at stage 2. This has some major consequences on the estimation procedure and has to be taken into account to avoid self-selection and resulting biased parameter estimates. As Deaton (1990) states, deleting the zero-consumption points only allows one to estimate conditional effects. The most popular model to cope with censored data is the Tobit approach (Tobin, 1958; Amemiya, 1984), which is widely used for single-equation demand models. Heien and Wessels (1990) proposed a new approach for equation systems based on Heckman (1979) (for its application see e.g. Dey, 2011; Lazaridis, 2003). This approach was criticized by Shonkwiler and Yen (1999) and later by Vermeulen (2001), who showed that this procedure leads to inconsistent estimates. As an alternative, Shonkwiler and Yen (1999) proposed another frequently used approach (e.g. in Thiele, 2008; Su & Zen, 2000; Zheng & Henneberry, 2011; for wine in particular see e.g. Stasi et al., 2011) that is applied in this study.

117

Chapter IV

Chapter IV IV Chapter Chapter In the first step, a multivariate probit is estimated (this is in contrast to Shonkwiler and

Yen (1999) who estimated a univariate probit). We suppose that one household‘s purchase of a particular beverage product category is not independent of the purchase decision for other beverage product categories due to substitution effects. The model to be estimated is:

( ) , (11)

{ (12)

and are the observed dependent variables for the food product groups. and are the correspondent latent variables. is a binary variable representing the decision of the household to consume or not. and are explanatory variables such as income, logarithmic prices, and household characteristics. All these variables are supposed to be important determinants with respect to the decision of the household to consume or not (in this context we also refer to Thiele, 2008 or Zheng &

Henneberry, 2011). The error terms and are assumed to have a multivariate normal distribution, each with a mean of zero and a variance-covariance matrix V with diagonal elements of 1 and off-diagonal elements of .

Based on the estimated parameters, we calculate the standard normal cumulative distribution function ( ) (cdf) and the standard normal probability density function ( ) (pdf) for every household and each product category. cdf and pdf are used to correct the budget share equation (4) for a censoring of the budget share variables as follows:

( ) ( ) ( ) , ( ) (13)

Following Zheng and Henneberry (2010), the parameter for the pdf represents the covariance between the error term in the budget share equation (4) and the error term of the multivariate probit model. Equation (13) is only applied to stage 2. At stage 1,

118

Consumer demand for alcoholic beverages

there is no need for the two-step estimation procedure due to the low level of zero consumption. We dropped the equation for expenditure on other products and services as described in section 2.1. For the two-step estimation procedure, the right hand side generally does not add up to one anymore. Therefore, we estimate the full equation system as Yen et al. (2002) or Ecker and Qaim (2010) do.15

Despite the advantages of Shonkwiler and Yen‘s (1999) procedure, Tauchmann (2005) draws attention to the problem of heteroscedasticity. By expanding the budget share equation cdf and pdf, heteroscedasticity is implicitly introduced into the model (for further explanations we refer readers to Tauchmann, 2005). To avoid this problem and to estimate a heteroscedasticity robust covariance matrix, we apply the parametric bootstrap.

4.2 Expenditure endogeneity

Assuming weak separability and applying a two-stage budgeting model could lead to problems with respect to endogeneity of the expenditure variable in the budget share equation because the expenditure variable could correlated with the equation errors (Attfield, 1985). To account for endogeneity, Blundell and Robin (1999) proposed the augmented regression technique. In a first step, we regress all the price variables, the household characteristics, and the variable for month and year of the budget share equation on the expenditure variable for every product category. On the right hand side, we further include income and its square as instruments as proposed by Bopape (2006). In contrast to other studies, where the equations are estimated using OLS (e.g. Fashogbon & Oni, 2013), we use a robust M-estimator to account for possible outliers. Residuals are then included in every budget share equation as additional right hand side variables. Testing the parameter of the residuals for significance allows us to check whether endogeneity is present or not.

15 Shonkwiler and Yen‘s (1999) two-step estimation procedure of does not allow one to implement the adding-up restriction into the model. 119

Chapter IV

Chapter IV IV Chapter Chapter 4.3 Elasticity estimates

Income elasticities for stage 2, where censoring is an issue, are derived using the formula proposed by

( ) (14)

which reduces to (Banks et al., 2010) for stage 1 where is the differentiation of equation (4) with respect to 16 For the Marshallian price elasticity , we follow Zheng and Henneberry (2010) and calculate the full effect of a price change on demand based on the effect of the budget share equation and the effect of the multivariate probit model for stage 2

( ) ( ) (15)

17 where is the differentiation of equation (4) with respect to is the estimated parameter for price with respect to product category in the multivariate probit model. is defined as above. denotes the Kronecker delta which is equal to one when , otherwise it is zero. For stage 1, the formula for the Marshallian price

elasticity reduces to .

The Slutsky equation is applied to get the Hicksian price elasticities:

(16)

For stage 2, we first estimate conditional elasticities with respect to total expenditure on beverages. Unconditional elasticities are derived following Edgerton (1997). The formula for the income elasticity (unconditional) is:

16 , * +- ( ) ( )

17 ( ∑ ) , * +- ( ) ( ) 120

Consumer demand for alcoholic beverages

( ) ( ) (17)

Where ( ) is the expenditure elasticity (conditional) for item within the th product group and ( ) the income elasticity for the th product group (unconditional). The unconditional Marshallian price elasticity is calculated by:

( ) ( ) ( ) (18)

( ) is the conditional Marshallian price elasticity for item and , ( ) the budget share of item within the th product category and the uncompensated Hicksian own-price elasticity of product category .

The unconditional, Hicksian price elasticity is derived analogously using ( ) and the Hicksian own-price elasticity (unconditional) of product category r :

( ) ( ) ( ) (19)

5 Results and Discussion

Overall, the variables in the model explain between 0.47% and 0.86% of total variance at stage 1 and between 0.06% and 0.80% of total variance at stage two. For the sake of brevity, we report the parameter estimates of stage 1 and 2, coefficients of determination, and -test statistics for the whole models in the appendix. While prices and socio-demographic variables only explain a part of the variance in consumption, the rest is determined by other factors such as lifestyle or psychographic or behavioral aspects, which are not considered further in this study (for determinants of Swiss wine demand see e.g. Brunner & Siegrist, 2011). While the household size is mostly negatively correlated with the budget share, the age of the households‘ reference person shows a positive correlation which is in line with the findings of previous studies. For the presence of young children and education we couldn‘t show a clear 121

Chapter IV

Chapter IV IV Chapter Chapter relationship with respect to the budget share. The sign of the parameters vary

depending on the household segment.

To verify the model specification with respect to the implementation of household characteristics and the quadratic income term, we carried out Wald-tests for every model. The test statistics reject the null-hypothesis of no household characteristics clearly for all models (Table 13 in the appendix), which can also be derived from the estimation results for the parameter estimates in Tables 24 to 28 (in the appendix). The hypothesis that Engel curves for beverages in Switzerland are often of a nonlinear form is supported by the test results. The QUAIDS specification is therefore superior to the AIDS specification for stage 1 and stage 2 except for the models for light drinking household and moderate drinking households (Table 14 in the appendix). These findings are consistent with other findings for food demand in Switzerland (Abdulai, 2002; Aepli & Kuhlgatz, 2014). Furthermore, we tested for endogeneity of the expenditure variable in the model. The hypothesis of no endogeneity is clearly rejected for all models and confirms the importance of a correction for endogeneity (Table 15 in the appendix), which we did by applying the augmented regression technique.

Unconditional Hicksian price and income elasticities for beverages are reported in Tables 5 to 12 for all households and the three segments. We provide standard errors and the significance level, which are computed by the delta method discussed in Oehlert (1992). We provide the conditional elasticities for stage 2 in the appendix. The elasticities for stage 1 can be obtained from the author on request.

122

Consumer demand for alcoholic beverages

Table 5: Unconditional, Marshallian price and income elasticities at stage 2, all households (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.328 1.288 -0.986 -0.177 0.046 -0.141 0.353 0.354 -0.256 0.054 (0.325) (0.182) (0.208) *** (0.142) (0.097) (0.181) (0.248) (0.203) (0.296) (0.135) Wine *** ***

-0.634 0.571 -1.260 -1.036 0.296 0.188 0.100 0.153 0.002 0.096 (0.242) (0.260) (0.238) (0.180) (0.153) (0.139) (0.093) (0.112) (0.109) (0.085) Beer *** ** *** *** *

-0.164 -0.950 0.519 0.918 0.397 -0.082 0.565 0.132 -0.070 0.003 (0.072) (0.043) (0.215) (0.101) (0.146) *** (0.122) (0.357) (0.116) (0.209) (0.069) Spirits ** *** ** ***

0.185 -0.786 -2.436 -1.508 -0.320 0.172 0.350 0.055 0.098 -0.126 (0.084) (0.081) (0.256) (0.158) (0.136) (0.079) (0.132) (0.088) (0.077) (0.110) Coffee ** *** *** *** ** ** ***

-4.638 -2.442 0.419 -0.157 0.007 -0.030 -0.143 0.300 -0.045 0.030 (0.506) (0.294) (0.160) (0.114) (0.108) (0.089) (0.128) (0.218) (0.180) (0.097) Tea *** *** ***

-0.593 -3.365 1.175 1.441 0.443 -0.138 0.028 -0.753 -0.305 -0.267 (0.219) (0.679) (0.236) (0.106) (0.242) * (0.100) (0.207) (0.484) (0.330) (0.167) Cocoa beverages ** *** *** ***

-0.416 -2.969 1.156 -0.068 -0.106 0.016 -0.064 -0.186 -0.213 0.058 (0.115) (0.273) (0.107) (0.182) (0.066) (0.166) (0.229) (0.148) (0.262) (0.110) Mineral water *** *** ***

0.099 -0.406 -0.565 -0.403 -1.169 0.226 0.518 -0.011 0.054 -0.031 Non-alcoholic soft (0.034) (0.148) (0.124) (0.079) (0.122) (0.065) (0.044) (0.094) (0.076) (0.076) drinks *** *** *** *** *** *** ***

0.318 0.164 -0.253 -1.933 -1.279 -0.745 -0.757 0.363 0.033 -0.053 Fruit juices and (0.109) (0.046) (0.074) (0.210) (0.167) (0.124) (0.086) (0.057) (0.073) (0.101) vegetables juices *** *** *** *** *** *** *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

123

Chapter IV

Chapter IV IV Chapter Chapter Table 6: Unconditional, Hicksian price and income elasticities at stage 2, all

households (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.324 0.355 1.288 -0.983 -0.176 0.046 -0.138 0.356 -0.252 0.055 (0.325) (0.203) (0.182) (0.208) *** (0.142) (0.097) (0.181) (0.248) (0.295) (0.135) Wine *** * ***

-0.634 0.571 -1.260 -1.036 0.296 0.188 0.100 0.153 0.002 0.096 (0.242) (0.260) (0.238) (0.180) (0.153) (0.139) (0.093) (0.112) (0.109) (0.085) Beer *** ** *** *** *

-0.163 -0.950 0.520 0.918 0.399 -0.080 0.567 0.133 -0.067 0.004 (0.072) (0.043) (0.215) (0.101) (0.146) *** (0.122) (0.357) (0.116) (0.209) (0.069) Spirits ** *** ** ***

0.185 -0.785 -2.435 -1.507 -0.320 0.172 0.350 0.056 0.098 -0.125 (0.084) (0.081) (0.256) (0.158) (0.136) (0.079) (0.132) (0.088) (0.077) (0.110) Coffee ** *** *** *** ** ** ***

-4.637 -2.442 0.419 -0.156 0.008 -0.030 -0.142 0.300 -0.044 0.031 (0.506) (0.294) (0.160) (0.113) (0.108) (0.089) (0.128) (0.218) (0.180) (0.097) Tea *** *** ***

-0.592 -3.361 1.176 1.441 0.446 -0.138 0.031 -0.751 -0.300 -0.266 Cocoa (0.219) (0.679) (0.236) (0.106) (0.242) * (0.100) (0.207) (0.484) (0.330) (0.167) beverages *** *** *** ***

-0.415 -2.966 1.156 -0.066 -0.106 0.018 -0.062 -0.185 -0.209 0.059 (0.115) (0.273) (0.107) (0.182) (0.066) (0.166) (0.229) (0.148) (0.262) (0.109) Mineral water *** *** ***

0.099 -0.404 -0.564 -0.402 -1.167 0.226 0.518 -0.010 0.054 -0.030 Non-alcoholic (0.034) (0.148) (0.124) (0.079) (0.122) (0.065) (0.044) (0.094) (0.076) (0.076) soft drinks *** *** *** *** *** *** ***

Fruit juices and 0.318 0.164 -0.252 -1.932 -1.278 -0.744 -0.757 0.363 0.034 -0.052 vegetables (0.109) (0.046) (0.074) (0.210) (0.167) (0.124) (0.086) (0.057) (0.073) (0.101) juices *** *** *** *** *** *** *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

124

Consumer demand for alcoholic beverages

Table 7: Unconditional, Marshallian price and income elasticities at stage 2, light drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.484 -1.275 -0.144 0.038 -0.156 -0.801 0.172 0.450 0.003 0.620 (0.177) (0.247) *** (0.141) (0.203) (0.161) (1.050) (0.718) (0.348) (0.595) (0.366) * Wine ***

-1.216 0.550 -2.940 0.679 0.000 -1.892 -1.056 0.480 1.082 0.469 (0.457) (0.226) (1.576) (0.707) (0.265) (2.355) (1.363) (0.545) (0.874) (0.534) Beer *** ** *

2.874 0.429 0.363 -0.011 -0.859 0.109 0.317 -0.760 0.417 0.666 (1.163) (0.235) (0.400) (0.151) (0.220) *** (0.152) (1.276) (0.730) (0.703) (0.408) Spirits ** *

-0.729 -3.209 -1.825 0.365 0.165 0.036 -0.431 0.192 0.708 0.569 (0.134) (1.311) (0.721) (0.348) (0.151) (0.163) (0.667) (0.240) (0.527) (0.307) * Coffee *** ** **

-0.606 2.528 -0.721 -1.000 -0.438 -0.070 -4.656 -0.586 -1.146 0.875 (0.297) (1.147) (0.343) (0.580) * (0.245) * (0.430) (2.844) (1.466) (1.403) (0.845) Tea ** ** **

-4.249 0.635 -0.332 0.014 -0.006 -1.244 -0.807 0.375 -0.574 1.489 Cocoa (1.628) (0.564) (0.627) (0.403) (0.446) (1.244) (0.915) (0.610) (0.825) (0.668) ** beverages ***

-6.929 2.068 -0.193 -0.854 -0.099 -0.710 0.722 0.490 -0.193 -1.239 (2.611) (0.403) (0.831) (0.670) (0.316) (0.445) (1.186) (1.599) (0.653) (0.819) Mineral water *** ***

-1.062 0.384 0.464 -0.282 0.100 -0.011 0.074 -0.168 -0.422 0.141 Non-alcoholic (0.256) (0.129) (0.109) (0.255) (0.230) (0.100) (0.154) (0.348) (0.339) (0.437) soft drinks *** *** ***

Fruit juices and -1.558 0.171 0.388 0.111 0.025 -0.812 -0.262 0.025 -0.206 0.220 vegetables (0.458) (0.275) (0.273) (0.163) (0.138) (1.077) (0.753) (0.251) (0.465) (0.248) juices ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

125

Chapter IV

Chapter IV IV Chapter Chapter Table 8: Unconditional, Hicksian price and income elasticities at stage 2, light drinkers

(at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.483 -1.273 -0.143 0.039 -0.155 -0.799 0.174 0.450 0.004 0.620 (0.176) (0.246) *** (0.141) (0.203) (0.161) (1.049) (0.718) (0.348) (0.595) (0.366) * Wine ***

-1.216 0.551 -2.938 0.680 0.001 -1.890 -1.056 0.480 1.082 0.469 (0.457) (0.225) (1.576) (0.707) (0.265) (2.355) (1.363) (0.545) (0.874) (0.534) Beer *** ** *

2.877 0.430 0.365 -0.010 -0.858 0.111 0.319 -0.760 0.417 0.666 (1.163) (0.235) (0.400) (0.151) (0.220) *** (0.152) (1.276) (0.730) (0.703) (0.408) Spirits ** *

-0.728 -3.207 -1.823 0.367 0.165 0.036 -0.431 0.193 0.708 0.569 (0.133) (1.311) (0.721) (0.347) (0.151) (0.163) (0.667) (0.240) (0.527) (0.307) * Coffee *** ** **

-0.604 2.529 -0.720 -0.998 -0.437 -0.070 -4.653 -0.584 -1.146 0.875 (0.296) (1.147) (0.342) (0.579) * (0.245) * (0.430) (2.843) (1.466) (1.403) (0.845) Tea ** ** **

-4.244 0.639 -0.330 0.014 -0.003 -1.240 -0.806 0.377 -0.573 1.489 Cocoa (1.627) (0.564) (0.627) (0.403) (0.446) (1.244) (0.915) (0.609) (0.825) (0.668) ** beverages ***

-6.923 2.068 -0.188 -0.852 -0.098 -0.705 0.727 0.491 -0.189 -1.238 (2.611) (0.403) (0.831) (0.670) (0.316) (0.445) (1.186) (1.599) (0.653) (0.819) Mineral water *** ***

-1.061 0.384 0.464 -0.280 0.100 -0.010 0.075 -0.166 -0.421 0.141 Non-alcoholic (0.256) (0.129) (0.109) (0.254) (0.230) (0.100) (0.154) (0.348) (0.339) (0.437) soft drinks *** *** ***

Fruit juices and -1.558 0.171 0.388 0.112 0.026 -0.812 -0.262 0.026 -0.206 0.220 vegetables (0.458) (0.275) (0.273) (0.163) (0.137) (1.077) (0.753) (0.251) (0.465) (0.248) juices ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

126

Consumer demand for alcoholic beverages

Table 9: Unconditional, Marshallian price and income elasticities at stage 2, moderate drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.665 0.194 0.236 -0.029 0.186 -0.183 -0.362 -0.562 -0.088 1.077 Wine (1.419) (1.155) (0.606) (0.554) (0.614) (0.548) (0.450) (0.563) (0.232) (0.355) ***

0.639 -0.825 0.365 -0.049 1.096 -0.893 -0.301 -0.317 -0.137 0.877 Beer (1.230) (1.186) (0.535) (0.699) (1.289) (0.554) (0.450) (0.704) (0.266) (0.867)

0.305 0.166 -0.799 -0.093 -0.877 -0.479 -0.143 -0.423 -0.144 0.636 Spirits (0.954) (0.453) (0.351) ** (0.249) (0.514) * (0.304) (0.310) (0.390) (0.145) (0.355) *

0.034 0.015 0.077 -0.828 -1.718 -0.250 0.173 -0.096 0.105 0.058 Coffee (1.414) (1.579) (0.686) (0.611) (1.648) (1.180) (0.473) (0.479) (0.361) (0.436)

0.264 0.005 0.043 -0.237 -1.360 -1.367 0.524 0.069 -0.170 0.241 Tea (2.398) (1.583) (0.960) (0.757) (1.488) (1.008) (0.685) (0.714) (0.403) (0.613)

Cocoa 0.337 -1.143 -0.097 -0.063 -2.195 1.858 -0.087 -0.129 0.389 -0.186 beverages (2.472) (2.480) (1.039) (1.441) (3.197) (3.379) (1.077) (1.496) (0.857) (1.539)

-0.293 -0.021 -0.033 -0.212 -0.716 -0.136 -0.512 -0.211 0.316 0.632 Mineral water (1.706) (0.873) (0.561) (0.523) (0.757) (0.620) (1.234) (0.704) (0.563) (0.509)

-1.141 0.145 -0.034 0.009 -0.004 -0.724 -0.053 0.023 -0.051 0.489 Non-alcoholic (0.550) (1.497) (0.870) (0.553) (0.435) (0.605) (0.577) (0.547) (0.286) (0.420) soft drinks **

Fruit juices and -2.267 -0.947 0.362 -0.058 0.007 -0.041 0.265 0.175 -0.419 0.503 vegetables (0.857) (0.356) (0.973) (0.937) (0.384) (0.452) (0.457) (0.638) (0.417) (0.329) juices *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

127

Chapter IV

Chapter IV IV Chapter Chapter Table 10: Unconditional, Hicksian price and income elasticities at stage 2, moderate

drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.663 0.194 0.236 -0.028 0.188 -0.182 -0.360 -0.556 -0.087 1.077 Wine (1.419) (1.155) (0.606) (0.554) (0.614) (0.548) (0.450) (0.563) (0.232) (0.355) ***

0.640 -0.825 0.365 -0.048 1.098 -0.892 -0.299 -0.312 -0.135 0.877 Beer (1.230) (1.186) (0.535) (0.699) (1.289) (0.554) (0.449) (0.701) (0.265) (0.867)

0.306 0.166 -0.799 -0.093 -0.876 -0.479 -0.141 -0.420 -0.143 0.636 Spirits (0.954) (0.453) (0.351) ** (0.249) (0.514) * (0.304) (0.310) (0.389) (0.145) (0.355) *

0.034 0.015 0.077 -0.828 -1.717 -0.250 0.173 -0.095 0.105 0.058 Coffee (1.414) (1.579) (0.686) (0.611) (1.648) (1.180) (0.473) (0.478) (0.361) (0.436)

0.265 0.006 0.043 -0.237 -1.360 -1.367 0.525 0.070 -0.170 0.241 Tea (2.398) (1.583) (0.960) (0.757) (1.488) (1.008) (0.685) (0.712) (0.402) (0.613)

Cocoa 0.336 -1.143 -0.097 -0.063 -2.195 1.857 -0.088 -0.130 0.389 -0.186 beverages (2.472) (2.480) (1.039) (1.441) (3.197) (3.378) (1.076) (1.491) (0.856) (1.539)

-0.292 -0.021 -0.033 -0.212 -0.715 -0.135 -0.511 -0.207 0.317 0.632 Mineral water (1.706) (0.873) (0.561) (0.523) (0.757) (0.620) (1.233) (0.703) (0.563) (0.509)

-1.139 0.146 -0.034 0.009 -0.003 -0.723 -0.052 0.024 -0.050 0.489 Non-alcoholic (0.549) (1.497) (0.870) (0.553) (0.435) (0.604) (0.577) (0.547) (0.286) (0.420) soft drinks **

Fruit juices and -2.266 -0.947 0.362 -0.057 0.007 -0.040 0.265 0.176 -0.416 0.503 vegetables (0.857) (0.356) (0.973) (0.937) (0.384) (0.452) (0.457) (0.638) (0.416) (0.329) juices *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

128

Consumer demand for alcoholic beverages

Table 11: Unconditional, Marshallian price and income elasticities at stage 2, heavy drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.456 -0.641 0.543 -0.519 0.573 0.251 0.178 1.533 -0.673 1.009 (0.803) (0.295) (0.325) (0.319) (0.557) (0.204) (0.119) (1.075) (0.516) (0.172) *** Wine * ** *

0.800 -0.244 0.289 0.498 4.057 -3.259 -1.306 -0.684 1.419 1.412 Beer (1.210) (2.026) (0.671) (0.604) (5.212) (4.122) (1.427) (0.987) (1.960) (0.768) **

-1.095 0.336 -0.684 -0.919 -0.260 -3.335 1.735 0.494 -0.284 0.357 (0.566) (0.512) (0.897) (0.315) *** (0.221) (1.771) * (1.339) (0.451) (0.372) (0.240) Spirits *

-0.997 0.066 -0.451 0.237 -3.220 1.173 0.408 -0.296 -0.898 0.488 (0.318) (0.451) (0.819) (0.275) (2.295) (1.721) (0.586) (0.402) (0.855) (0.338) Coffee ***

0.230 -1.307 0.199 -0.319 -2.582 3.393 1.149 -0.631 -0.859 0.258 Tea (0.830) (1.232) (0.522) (0.416) (3.903) (2.865) (0.799) (0.674) (1.050) (0.503)

Cocoa 0.131 -0.429 0.100 -0.285 -6.955 -1.639 0.055 -0.274 -0.864 0.426 beverages (0.873) (1.159) (0.569) (0.655) (5.165) (3.760) (1.388) (0.790) (2.016) (0.634)

0.079 -0.442 -0.055 -0.374 -2.208 1.556 -0.686 -0.591 0.007 0.398 Mineral water (0.468) (0.783) (0.343) (0.457) (2.190) (1.655) (0.775) (0.491) (1.110) (0.350)

-1.702 0.483 -0.193 0.069 -0.255 -2.633 1.908 0.308 -0.931 0.389 Non-alcoholic (0.421) (0.545) (1.027) (0.333) (0.329) (2.429) (1.947) (0.649) (0.951) (0.378) soft drinks ***

Fruit juices and 0.233 -0.969 -0.082 -0.691 -2.330 1.553 1.517 -0.160 -2.154 0.395 vegetables (0.826) (1.339) (0.410) (0.622) (3.855) (3.069) (1.340) (0.623) (1.735) (0.601) juices

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

129

Chapter IV

Chapter IV IV Chapter Chapter Table 12: Unconditional, Hicksian price and income elasticities at stage 2, heavy

drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.455 -0.639 0.545 -0.518 0.574 0.251 0.179 1.535 -0.666 1.009 (0.803) (0.295) (0.325) (0.319) (0.557) (0.204) (0.119) (1.075) (0.516) (0.172) *** Wine ** ** *

0.801 -0.244 0.289 0.499 4.059 -3.258 -1.303 -0.674 1.420 1.412 Beer (1.210) (2.026) (0.671) (0.604) (5.212) (4.122) (1.426) (0.984) (1.960) (0.768) **

-0.919 -1.094 0.337 -0.684 -0.260 -3.334 1.735 0.495 -0.281 0.357 (0.315) (0.566) (0.512) (0.897) (0.221) (1.771) * (1.339) (0.451) (0.371) (0.240) Spirits *** *

0.067 -0.451 0.237 -0.996 -3.219 1.173 0.409 -0.292 -0.897 0.488 Coffee (0.451) (0.819) (0.275) (0.318) *** (2.295) (1.721) (0.586) (0.400) (0.855) (0.338)

0.231 -1.307 0.199 -0.319 -2.582 3.393 1.150 -0.629 -0.859 0.258 Tea (0.830) (1.232) (0.522) (0.416) (3.903) (2.865) (0.799) (0.671) (1.050) (0.503)

Cocoa 0.131 -0.429 0.100 -0.284 -6.954 -1.639 0.056 -0.271 -0.863 0.426 beverages (0.873) (1.159) (0.569) (0.655) (5.165) (3.760) (1.388) (0.787) (2.016) (0.634)

0.079 -0.442 -0.055 -0.373 -2.207 1.556 -0.685 -0.588 0.007 0.398 Mineral water (0.468) (0.783) (0.343) (0.457) (2.190) (1.655) (0.775) (0.489) (1.110) (0.350)

-1.699 0.484 -0.193 0.069 -0.255 -2.633 1.908 0.309 -0.931 0.389 Non-alcoholic (0.419) (0.545) (1.027) (0.333) (0.329) (2.429) (1.947) (0.649) (0.951) (0.378) soft drinks ***

Fruit juices and 0.234 -0.969 -0.082 -0.691 -2.330 1.553 1.518 -0.157 -2.154 0.395 vegetables (0.826) (1.339) (0.410) (0.622) (3.855) (3.069) (1.340) (0.620) (1.735) (0.601) juices

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

General price and income responses for beverages and for the three household segments are discussed in terms of elasticities. Due to the cross-sectional structure of

130

Consumer demand for alcoholic beverages

the data set, elasticities should be interpreted as a short-term reaction of the households to price and income changes.

Income elasticities for alcoholic beverages are all positive in the model for all households. Wine is found to be a luxury good and spirits and beer are found to be necessity good, while spirits is very closed to the luxury goods. The findings for beer are in line with previous studies (see e.g. Nelson, 1997 or Selvanathan & Selvanthan, 2005). With respect to wine and spirits the findings in literature are partially contradictory. In most studies wine and spirits are found to be luxury goods or have an elasticity which is smaller but closed to one (Fogarty, 2008). Therefore our findings are in the range of variability of the results of previous studies.

Looking at the income elasticities of the three segments, alcoholic beverages are a necessity good for light drinking households, as well as for moderate drinking households except of wine. For heavy drinking households, beer and wine are luxury goods, while spirits are necessity goods.18 Light drinking households are clearly less income-elastic for wine and beer than moderate and heavy drinking households are, and heavy drinking households are much more income-elastic with respect to beer than moderate drinking households are.

Almost all own-price elasticities for alcoholic and non-alcoholic beverages are negative for the three household segments as well as for the model with all households. Therefore, the negativity condition is mostly fulfilled. For wine and beer, our findings show a clear decrease in the magnitude from light drinking households to heavy drinking households. Looking at the Hicksian demand elasticities, an additional increase of beer price by 1% would result in an increase of the light drinking household demand by 1.21%, while moderate drinking household increase their consumption by 0.83% and heavy drinking households even less by 0.24%. This shows that light drinking households are relatively price elastic while heavy drinking

18 A good is called a luxury good if demand increases more than proportionally as income rises ( >1). If demand increases less than proportionally the good is called a necessity good ( <1). 131

Chapter IV

Chapter IV IV Chapter Chapter households are inelastic. Beer is of special interest because it is the favorite beverage

among young people in Switzerland. A recently published study by Dey et al. (2013) shows that, in Switzerland, those who practice binge drinking or other risky drinking behavior have a high penchant for beer. The authors (Dey et al., 2013) suggest the relatively low price of beer in comparison to other alcoholic beverages explains these findings. The price argument probably applies also to middle-aged or older people in terms of risky drinking behavior and could be a reason why own-price elasticities for spirits, which are more expensive than beer, are almost constant between the three segments with a range from 0.83 and 0.92.19 This is in contrast to Kuo et al. (2003) who found that moderate drinkers show relatively higher price responses than heavy drinkers in relation to spirits.

Looking at the model for all households, non-alcoholic beverages are mostly substitutes for wine and beer. In the case of spirits, tea, cocoa beverages, and mineral water are slight complements, while coffee, non-alcoholic soft drinks, and fruit and vegetables juices are substitutes. With respect to the substitution effects between alcoholic and non-alcoholic beverages among the three households segments, we did not find a clear structure. There are substitutes as well as complementary goods depending on the household segment.

The own-price elasticities for non-alcoholic beverages for the three segments are all negative except for mineral water for light drinking households and cocoa beverages for moderate drinking households. The magnitude of the Hicksian own-price elasticities shows a tendency for higher price sensitivity from light to heavy drinking households, indicating that heavy drinking households are price-sensitive to non- alcoholic beverages in contrast to alcoholic beverages.

19 The quality adjusted price for beer (3.00 CHF/Liter) is much lower than the price for wine (10.20 CHF/Liter) or spirits (15.18 CHF/Liter). This supports the interpretation proposed by Dey et al. (2013) that the low price for beer plays an important role with respect to the drinking behavior. 132

Consumer demand for alcoholic beverages

6 Conclusions and policy implications

This paper reports on the income and price elasticities for different alcoholic and non- alcoholic beverage product categories with respect to light, moderate, and heavy drinking households in Switzerland. It is based on microdata of the Swiss household expenditure survey from 2000 to 2009 that contains data from more than 34,000 households. We applied a two-stage quadratic almost ideal demand system, correcting for the high share of zero consumption for some product categories at stage 2 and for endogeneity of the expenditure variable. Missing price data were received by adjusting unit values for quality as recently proposed by Aepli and Finger (2013) and Majumder et al. (2012). We tested the model specification with respect to household characteristics as well as the quadratic income term, which distinguishes the QUAIDS from the AIDS. Testing for the overall significance of the household characteristics and the quadratic income terms confirms the model specification. With respect to wine and beer, moderate and heavy drinking households are less price-sensitive than light drinking households are, while for spirits we did not find a difference among the three household segments.

Our findings have major policy implications with respect to the alcohol tax. In general, the higher the negative effects of alcohol consumption and the lower the elasticity, the higher the tax should be fixed and vice versa. This only holds if we assume a constant elasticity function. By dividing the sample into three segments, we have shown that this does not hold for Swiss households and heavy drinking households are less price- sensitive, in particular with respect to wine and beer, than moderate or light drinking households. To fix the optimal level of a tax on alcohol, the different responses of households to price changes should be considered. As already mentioned by Manning (1995), the optimal level for an alcohol tax is a trade-off between economic, public health, and social gains due to a reduction in consumption of heavy drinkers and the

133

Chapter IV

Chapter IV IV Chapter Chapter possible adverse social or even health effects20 with respect to light or moderate

drinkers due to an additional tax burden. Our findings clearly show that the assumption of a constant elasticity function with respect to the three defined segments is violated and that households do not respond similarly to price changes in wine and beer. A tax on those products will therefore lead to a decrease in the consumption, especially in light drinking households, and, to a smaller extent, in moderate drinking households. The effect on heavy drinking household is minimal. From a social and health perspective, this may not be a desirable development. Therefore, before implementing a new tax on alcoholic beverages in Switzerland or in other European countries, the social and health externality costs, and the economic effects of a rather small decrease in alcohol consumption among heavy drinking households must be weighed against possible adverse health or economic consequences of a sharp decline in light or moderate drinking households. To a certain extent, our findings for Switzerland can be transferred to other high-income countries and contribute to a differentiated discussion on alcohol tax.

In addition to the estimation of the negative external effects of alcohol consumption, a topic for further research will be the estimation of elasticities for different household types with respect to the age of the households‘ reference person. As Kuo et al. (2003) already mentioned, for spirits the price response depends on age. While younger or middle-aged people react more to price changes in spirits, older people are quite inelastic. Our findings show that households react differently to spirits than to wine and beer. Therefore, the findings of Kuo et al. (2003) will probably not hold for wine and beer and a further study should shed light on this issue.

Another point of interest is the substitution within a product category for example from more expensive alcoholic beverages to cheaper ones with lower quality due to a higher tax. Although we were not able to show this effect in our analysis it is possible that people would buy cheaper products which are probably more risky. Gruenewald

20 The health effects depend on the assumptions which were made with respect to low or moderate alcohol consumption. 134

Consumer demand for alcoholic beverages

et al. (2006) show that consumers respond to price increases by altering their consumption, varying their brand choices as well as substituting between quality classes. This has consequences for a tax increase. For example, a tax increase only for high quality alcohol beverages could lead to substitution towards lower quality products and operate as a positive income effect, probably associated with higher consumption of alcoholic beverages in total (Gruenewald et al., 2006). This illustrates the need for research with respect to quality substitution.

From a methodological perspective, further research should be conducted on the estimation of price and income for different household segments within a single QUAIDS by introducing cross-terms into the budget share equation, to estimate possible interaction effects between household types and price or income parameters.

135

Chapter IV

Chapter IV IV Chapter Chapter

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Aepli, M., & Finger, R. (2013). Determinants of sheep and goat meat consumption in Switzerland. Agricultural and Food Economics. doi:10.1186/2193-7532-1-11

Aepli, M., & Kuhlgatz, C. (2014). Meat and milk demand elasticities for Switzerland: A three stage budgeting Quadratic Almost Ideal Demand System. Submitted.

Akbay, C., Boz, I., & Chern W.S. (2007). Household food consumption in Turkey. European Review of Agricultural Economics, 34(2), 209-231.

Amemiya, T. (1984). Tobit models: A survey. Journal of Econometrics, 24(1/2), 3-61.

Angulo, A.M., Gil, J.M., & Gracia, A. (2001). The demand for alcoholic beverages in Spain. Agricultural Economics, 26(1), 71-83.

Annaheim, B., & Gmel, G. (2004). Alkoholkonsum in der Schweiz. Scientific study, schweizerische Fachstelle für Alkohol- und andere Drogenprobleme, Lausanne.

Attfield, C.L.F. (1985). Homogeneity and endogeneity in systems of demand equations. Journal of Econometrics, 27(2), 197-209.

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel Curves and Consumer Demand. Review of Economics and Statistics, 79(4), 527-539.

Barnett, W.A., Kanyama, I.K. (2013). Time-varying parameters in the almost ideal demand system and the Rotterdam model: will the best specification please stand up?. Journal of Applied Econometrics, 45(29), 4169-4183.

Barnett, W.A., & Seck, O. (2008). Rotterdam model versus Almost Ideal Demand System: Will the best specification please stand up? Journal of Applied Econometrics, 23(6), 795-824.

Barten, A.P. (1964). Consumer demand functions under conditions of almost additive preferences. Econometrica, 32(1/2), 1-38.

136

Consumer demand for alcoholic beverages

Blackorby, C.H., Boyce, R., & Russell, R.R. (1978). Estimation of demand systems generated by the Gorman Polar Form: A generalization of the S-Branch Utility Tree. Econometrica, 46(2), 345-363.

Blundell, R., Robin, J.M. (1999). Estimation in large and disaggregated demand systems: An estimator for conditionally linear systems. Journal of Applied Econometrics, 14(3), 209-232.

Bopape, L., & Myers, R. (2007). Analysis of household food demand in South Africa: Model selection, expenditure endogeneity, and the influence of socio-demographic effects. Paper presented at the AESA Conference, Cape Town, South Africa, July 2007.

Brehe, M. (2007). Ein Nachfragesystem für dynamische Mikrosimulationsmodelle. Dissertation, University of Potsdam.

Brunner, T.A., & Siegrist, M. (2011). A consumer-oriented segmentation study in the Swiss wine market. British Food Journal, 113(3), 353-373.

Cembalo, L., Caracciolo, F., & Pomarici, E. (2014). Drinking cheaply: the demand for basic wine in Italy. Australian Journal of Agricultural and Resource Economics, doi: 10.1111/1467-8489.12059

Christensen, L.R., Jorgenson, D.W., & Lau, L.J. (1975). Transcendental logarithmic utility functions. The American Economic Review, 65(3), 367-383.

Chung, C., Dong, D. Schmit, T.M., Kaiser, H.M. & Gould, B.W. (2005). Estimation of price elasticities from cross-sectional data. Agribusiness, 21(4), 565-584.

Corrao, G., Rubbiati, L., Bagnardi, V., Zambon, A., & Poikolainen, K. (2000). Alcohol and coronary heart disease: a meta-analysis. Addiction, 95(10), 1505-23.

Cox, T.L., & Wohlgenant, M.K. (1986). Prices and Quality Effects in Cross-Sectional Demand Analysis. The American Economic Review, 68(4), 908-919.

Cranfield, J.A.L., Eales, J.S., Hertel, T.W., & Preckel, P.V. (2003). Model selection when estimating and predicting consumer demands using international, cross section data. Empirical Economics, 28(2), 353-364. 137

Chapter IV

Chapter IV IV Chapter Chapter Deaton, A. (1990). Price elasticities from survey data: Extensions and Indonesian

results. Journal of Econometrics, 44(3), 281-309.

Deaton, A., & Muellbauer, J. (1980a). An Almost Ideal Demand System. The American Economic Review, 70(3), 312-326.

Deaton, A., & Muellbauer, J. (1980b). Economics and consumer behavior. Cambridge University Press, Cambridge.

De Boer, P., &, Paap, R. (2009). Testing non-nested demand relations: Linear expenditure system versus indirect addilog. Statistica Neerlandica, 63(3), 368-384.

Dey, M.M., Alam, M.F., & Paraguas, F.J. (2011). A multistage budgeting approach to the analysis of demand for fish: An application to inland areas of Bangladesh. Marine resource economics, 26(1), 35-58.

Dey, M., Gmel, G., Studer, J., Dermota, P., & Mohler-Kuo, M. (2013). Beverage preferences and associated drinking patterns, consequences and other substance use behaviours. European Journal of Public Health. doi: 10.1093/eurpub/ckt109

Doran, C.M., Byrnes, J.M., Cobiac. L.J., Vandenberg, B., & Vos, T. (2013). Estimated impacts of alternative Australian alcohol taxation structure on consumption, public health and government revenues. The Medical Journal of Australia, 199(9), 619-622.

Ecker, O., Qaim, M. (2010). Analyzing nutritional impacts of policies: An empirical study for Malawi. World Development, 39(3), 412-428.

Edgerton, D.L. (1997). Weak separability and the estimation of elasticities in multistage demand systems. The American journal of agricultural Economics, 79(1), 62-79.

Estruch, R., Ros, E., Salas-Salvadó, J., Covas, M-I., Corella, D., Arós, F., Gómez- Gracia, E., Ruiz-Gutiérrez, V., Fiol, M., Lapetra, J., Lamuela-Raventos., Serra-Majem, L., Pintó, X., Basora, J., Muñoz, M.A., Sorlí, J.V., Martínez, J.A., & Martínez-González, M.A. (2014). Primary prevention of cardiovascular disease with a mediterranean diet. The New England Journal of Medicine, doi: 10.1056/NEJMx140004

Eurostat. (2013). Statistiken zur Nahrungsmittelkette – Vom Erzeuger zum Verbraucher. Press release, EU Comission, Brussels.

138

Consumer demand for alcoholic beverages

Fashogbon, A.E., & Oni, O.A. (2013). Heterogeneity in rural household food demand and its determinants in Ondo State, Nigeria: An application of quadratic almost ideal demand system. Journal of Agricultural Science, 5(2), 169-177.

FDF (2009). Alkohol in der Schweiz – Übersicht. Report, Swiss Federal Department of Finance, Bern.

Ferguson, S.M., McLeod, G.G., & Horwood, J.L. (2013). Alcohol misuse and psychosocial outcomes in young adulthood: results from a longitudinal birth cohort studied to age 30. Drug and alcohol dependence, 133(2), 513-519.

FOPH (2014a). Alcohol policy regulation in Europe. Swiss Federal Office of Public Health, Bern. http://www.bag.admin.ch/themen/drogen/00039/10172/12019/index.html?lang=en. Accessed 14 Jan, 2014.

FOPH (2014b). Swiss food composition database. Swiss Federal Office of Public Health, Bern. http://naehrwertdaten.ch/request?xml=MessageData&xml=MetaData&xsl=Start&lan=e n&pageKey=Start. Accessed 16 Jan, 2014.

FOPH (2013). Faktenblatt: Entwicklung des Alkoholkonsum der Schweiz seit den 1880er Jahren. Swiss Federal Office of Public Health, Bern.

Forgarty, J. (2008). The demand for beer, wine and spirits: Insights from a meat analysis approach. AAWE Working Paper No. 31, American Association of Wine Economists.

Gallet, C.A. (2010). The income elasticity of meat: A meta-analysis. Australian Journal of Agricultural and Resource Economics, 54(4), 477-490.

Gallet, C.A. (2007). The Demand for Alcohol: A Meta-analysis of elasticities. Australian Journal of Agricultural and Resource Economics, 51(2), 121-135.

Gruenewald, P.J., Ponicki, W.R., Holder, H.D., & Romelsjö. A. (2006). Alcohol prices, beverage quality, and the demand for alcohol: Quality substitutions and price elasticities. Alcoholism: Clinical and Experimental Research, 30(1), 96-105.

Hagenaars, A.J.M., De Vos, K., & Zaidi, M.A. (1994). Poverty statistics in the late 1980s: Research based on micro-data. Kontoret for De Europæiske Fællesskabers Officielle Publikationer, Luxembourg.

139

Chapter IV

Chapter IV IV Chapter Chapter Heckman, J. (1979). Sample selection bias as a specification error. Econometrica,

47(1), 153-161.

Heeb, J.L., Gmel, G., Zurbrügg, C.H., Kuo, M., & Rehm, J. (2003). Changes in alcohol consumption following a reduction in the price of spirits: a natural experiment in Switzerland. Addiction, 98(10), 1433-1446.

Heien, D., & Wessels, C.R. (1990). Demand systems estimation with microdata: A censored regression approach. Journal of Business & Economic Statistics, 8(3),365- 371.

Holt, M.T., & Goodwin, B.K. (2009). The almost ideal and translog demand systems. MPRA Paper No. 15092, Munich.

Jacobs, L., & Steyn, N. (2013). Commentary: If you drink alcohol, drink sensibly: Is this guideline still appropriate? Ethnicity & Disease, 23(1),110-115.

Jithitikulchai, T. (2011). U.S. alcohol consumption: tax instrumental variables in Quadratic Almost Ideal Demand System. Paper presented at the AAEA & NAREA Joint Annual Meeting, Pittsburgh, Pennsylvania, July 2011.

Klonaris, S., & Hallam, D. (2003). Conditional and unconditional food demand elasticities in a dynamic multistage demand system. Applied Economics, 35(5), 503- 514.

Kuo, M., Heeb, J.L., Gmel, G., & Rehm, J. (2003). Does price matter? The effect of decreased price on spirits consumption in Switzerland. Alcoholism: Clinical and experimental research, 27(4), 720-725.

Lazaridis, P. (2003). Household meat demand in Greece: A demand systems approach using microdata. Agribusiness, 19(1), 43-59.

Lewbel, A., & Pendakur, K. (2009). Tricks with Hicks: The EASI demand system. American Economic Review, 99(3), 827-863.

McKelvey, C. (2011). Price, unit value, and quality demanded. Journal of Development Economics, 95(2), 157-169.

Manning, W.G., Blumberg, L., & Moulton, L.H. (1995). The demand for alcohol: The differential response to price. Journal of Health Economics, 14(2), 123-148.

140

Consumer demand for alcoholic beverages

Majumder, A., Ray, R., & Sinha, K. (2012). Calculating rural-urban food price differentials from unit values in household expenditure surveys: A comparison with existing methods and a new procedure. The American Journal of Agricultural Economics, 94(5), 1218-1235.

Mhurchu, C.N., Eyles, H., Schilling, C., Yang, Q., Kaye-Blakem, W., Genç, M., & Blakely, T. (2013). Food prices and consumer demand: Differences across income levels and ethnic groups. PLoS One. DOI: 10.1371/journal.pone.0075934

Moschini, G., Moro, D., & Green, R.D. (1994). Maintaining and testing separability in demand systems. The American Journal of Agricultural Economics, 76(1), 61-73.

Nelson, J.P. (1997). Economic and demographic factors in US alcohol demand: a growth-accounting analysis. Empirical Economics, 22(1), 83-102.

Oehlert, G.W. (1992). A note on the delta method. The American Statistician, 46(1), 27-29.

Perali, F., & Chavas, J.P. (2000). Estimation of censored demand equations from large cross-section data. The American Journal of Agricultural Economics, 82(4), 1022-1037.

Pollak, R.A., & Wales, T.J. (1978). Estimation of complete demand systems from household budget data: The linear and quadratic expenditure systems. The American Economic Review, 68(3), 348-359.

Reavley, N.J., Jorm, A.F., McCann, T.V., & Lubman, D.I. (2011). Alcohol consumption in tertiary education students. BMC Public Health. doi:10.1186/1471-2458-11-545

Rehn, N., Room, R., & Edwards, G. (2001). Alcohol in the European region – consumption, harm and policies, report, WHO Regional Office for Europe, Copenhagen.

Rice, N., Carr-Hill, R., Dixon, P., & Sutton, M. (1998). The influence of households on drinking behavior: A multilevel analysis. Social Science & Medicine, 46(8), 971-979.

SAB (2013). Zahlen und Fakten – Konsum. Swiss Alcohol Board, Bern. http://www.eav.admin.ch/dokumentation/00439/00564/index.html?lang=de. Accessed 14 Jan, 2014.

141

Chapter IV

Chapter IV IV Chapter Chapter Sabia, S., Elbaz, A., Britton, A., Bell, S., Dugravot, A., Shipley, M., Kivimaki, M., & Sing-Manoux, A. (2014). Alcohol consumption and cognitive decline in early old age.

Neurology. doi: 10.1212/WNL.0000000000000063

Sacks, J.J., Roeber, J., Bouchery, E.E., Gonzales, K., Chaloupka, F.J., & Brewer, R.D. (2013). State costs of excessive alcohol consumption, 2006. American journal of preventive medicine, 45(4),474-485.

Schwartz, L.M., Persson, E.C., Weinstein, S.J., Graubard, B.I., Freedman, N.D., Mannisto, S., Albanes, D., & McGlynn, K.A. (2013). Alcohol consumption, one-carbon metabolites, liver cancer and liver disease mortality. PLos One, 8(10), e78156.

Selvanathan, S., & Selvanathan, E.A. (2005). The demand for alcohol, tobacco and marijuana: International evidence, Ashgate Publishing, Aldershot, England.

SFSO (2011). Haushaltsbudgeterhebung 2005. Swiss Federal Statistical Office, Neuchâtel.

Shonkwiler, J.S., & Yen, S.T. (1999). Two-step estimation of a censored system of equations. The American Journal of Agricultural Economics, 81(4), 972-982.

Stasi, A., Seccia, A., & Nardone, G. (2011). Italian wine consumers‘ preferences and impact of taxation on wines of different quality and source. Paper presented at the 6th AWBR International Conference, France.

Stasi, A., Seccia, A., & Nardone, G. (2010). Market power and price competition in the Italian wine market. Enometrica, Review of the Vineyard Data Quantification Society (VDQS) and the European Association of Wine Economists (EuAWE) - Macerata University.

Su, S.J., & Yen, S. (2000). A censored system of cigarette and alcohol consumption. Journal of Applied Economics, 32(6), 729-737.

Tafere, K., Taffesse, A.S., & Tamru, S. (2010). Food demand elasticities in Ethiopia: Estimates using household income consumption expenditure (HICE) survey data. ESSP II Working paper 11, International Food Policy Research Institute, Washington.

Tauchmann, H. (2005). Efficiency of two-step estimators for censored systems of equations: Shonkwiler and Yen reconsidered. Journal of Applied Economics, 37(4), 367-374.

142

Consumer demand for alcoholic beverages

Theil, H. (1965). The Information Approach to demand analysis. Econometrica, 33(1), 67-87.

Thiele, S. (2010). Erhöhung der Mehrwertsteuer für Lebensmittel: Budget- und Wohlfahrtseffekte für Konsumenten. Jahrbücher für Nationalökonomie und Statistik, 230(1), 115-130.

Tobin, J. (1958). Estimation of relationships for limited dependent variables. Econometrica, 26(1), 24-36.

Van Oers, J.A.M., Bongers, I.M.B., Van de Goor, L.A.M., & Garretsen, H.F.L. (1999). Alcohol consumption, alcohol-related problems, problem drinking, and socioeconomic status. Alcohol & Alcoholism, 34(1), 78-88.

Vermeulen, F. (2001). A note of Heckman-type corrections in models for zero expenditures. Applied Economics, 33(9), 1089-1092.

Wakabayashi, I. (2013). Relationship between alcohol intake and lipid accumulation product in middle-aged men. Alcohol and Alcoholism, 48(5), 535-542.

WHO (2007). Prevention of cardiovascular disease. Report, World Health Organization, Geneva.

Xuan, Z.M., Nelson, T.F., Heeren, T., Blanchette, J., Nelson, D.E., Gruenewald, P., & Naimi, T.S. (2013). Tax Policy, adult binge drinking, and youth alcohol consumption in the United States. Alcoholism - clinical and experimental research, 37(10), 1713-1719.

Yen, S.T., Kan, K., Su, S.J. (2002). Household demand for fats and oils: two-step estimation of a censored demand system. Applied Economics, 34(14), 1799-1806.

Zheng, Z., & Henneberry, S.R. (2011). Household food demand by income category: Evidence from household survey data in an urban Chinese province. Agribusiness, 27(1), 99-113.

Zheng, Z., Henneberry, S.R. (2010). An analysis of food grain consumption in urban Jiangsu province of China. Journal of Agricultural and Applied Economics, 42(2), 337- 355.

143

Chapter IV

Chapter IV IV Chapter Chapter

Appendix

Table 13: Wald tests for household characteristics

conclusion with respect to the -test P-value nullhypothesis (5% significance statistic level)

Stage 1 23,140.44 0.000 rejected

Stage 2: All households 47,840.72 0.000 rejected

Stage 2: Light drinkers 14,633.41 0.000 rejected

Stage 2: Moderate 13,902.19 0.000 rejected drinkers

Stage 2: Heavy drinkers 1340.49 0.000 rejected

Note: Nullhypothesis: The specification with household characteristics is not superior than the specification without household characteristics.

Table 14: Wald tests for the quadratic term in the QUAIDS

conclusion with respect to the -test P-value nullhypothesis (5% significance statistic level)

Stage 1 21.55 0.171 rejected

Stage 2: All households 30.59 0.000 rejected

Stage 2: Light drinkers 7.22 0.610 not rejected

Stage 2: Moderate 10.21 0.334 not rejected drinkers

Stage 2: Heavy drinkers 25.79 0.002 rejected

Note: Nullhypothesis: The QUAIDS specification is not superior than the AIDS specification.

144

Consumer demand for alcoholic beverages

Table 15: Wald tests for endogeneity of the expenditure variable

conclusion with respect to the -test P-value nullhypothesis (5% significance statistic level)

Stage 1 432.48 0.000 rejected

Stage 2: All households 871.89 0.000 rejected

Stage 2: Light drinkers 839.37 0.000 rejected

Stage 2: Moderate 18.78 0.027 rejected drinkers

Stage 2: Heavy drinkers 39.45 0.000 rejected

Note: Nullhypothesis: There is no endogeneity problem.

145

Chapter IV

Chapter IV IV Chapter Chapter Table 16: Conditional, Marshallian price and income elasticities at stage 2, all

households (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.031 0.585 0.302 -1.033 -0.159 -0.022 -0.109 -0.079 -0.015 1.676 (0.198) (0.156) (0.126) (0.070)*** (0.112 ) (0.049 ) (0.062)* (0.068 ) (0.079 ) (0.110)*** Wine *** *** **

-0.633 -1.243 -1.040 0.291 0.185 0.095 0.155 0.593 0.015 0.125 (0.235) (0.254) (0.181) (0.146) (0.125 ) (0.076 ) (0.119 ) (0.275)** (0.133 ) (0.106) Beer *** *** *** **

-0.152 0.776 0.684 0.364 -0.998 -0.060 0.095 0.057 -0.046 1.195 (0.049) (0.256) (0.131) (0.043)*** (0.030)*** (0.027)** (0.090 ) (0.030)* (0.030 ) (0.088)*** Spirits *** *** ***

0.190 -2.356 -1.445 -0.334 0.153 0.042 0.080 -0.777 -0.078 0.456 (0.077) (0.186)** (0.122) (0.094) (0.060) (0.051 ) (0.036)** (0.042)*** (0.053 ) (0.083)*** Coffee ** * *** *** ***

-4.542 -2.367 0.283 -0.172 0.013 -0.052 -0.133 0.012 0.008 0.545 (0.519) (0.253) (0.139) (0.080)** (0.114 ) (0.049 ) (0.071)* (0.095 ) (0.093 ) (0.121)*** Tea *** *** **

-0.573 -3.034 1.116 -0.345 0.390 -0.213 0.063 -0.494 -0.106 1.875 Cocoa (0.212) (0.589) (0.192) (0.136) (0.130)*** (0.077)*** (0.112 ) (0.366 ) (0.122 ) (0.087)*** beverages *** *** *** **

-0.400 -2.703 -0.233 -0.110 -0.167 0.044 0.143 -0.054 -0.004 1.504 (0.097) (0.174) (0.088) (0.064)* (0.037)*** (0.050 ) (0.130 ) (0.058 ) (0.063 ) (0.053)*** Mineral water *** *** ***

-0.472 -0.424 -1.098 0.198 -0.030 0.061 0.072 -0.018 -0.287 0.674 Non-alcoholic (0.091) (0.061) (0.053) (0.052) (0.045 ) (0.068 ) (0.025)*** (0.037 ) (0.117)** (0.027)*** soft drinks *** *** *** ***

Fruit juices and 0.323 -1.850 -1.213 -0.760 -0.777 0.020 0.145 -0.244 -0.003 0.473 vegetables (0.103) (0.197) (0.137) (0.110) (0.087) (0.059 ) (0.037)*** (0.059)*** (0.072 ) (0.039)*** juices *** *** *** *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

146

Consumer demand for alcoholic beverages

Table 17: Conditional, Hicksian price and income elasticities at stage 2, all households (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.937 0.624 0.510 0.219 0.192 -0.670 -0.054 0.098 0.132 1.676 (0.194) (0.157) (0.136) (0.072) (0.081) (0.076)*** (0.111 ) (0.046)** (0.064)** (0.110)*** Wine *** *** *** *** **

-0.625 -1.240 -1.025 0.306 0.211 0.104 0.173 0.600 0.037 0.125 (0.234) (0.255) (0.189) (0.145) (0.132 ) (0.074 ) (0.119 ) (0.272)** (0.134 ) (0.106) Beer *** *** *** **

0.843 0.712 0.243 0.269 0.102 0.622 -0.077 -0.913 0.112 1.195 (0.254) (0.131) (0.100) (0.035) (0.031) (0.046)*** (0.047)* (0.027)*** (0.031)*** (0.088)*** Spirits *** *** ** *** ***

0.219 -2.330 -1.434 -0.277 0.209 0.141 0.112 -0.712 0.003 0.456 (0.076) (0.183) (0.123) (0.101) (0.060) (0.056)** (0.033)*** (0.045)*** (0.057 ) (0.083)*** Coffee *** *** *** *** ***

-4.511 -2.354 0.350 -0.054 0.047 -0.013 -0.054 0.109 0.075 0.545 (0.518)** (0.254) (0.151) (0.074 ) (0.113 ) (0.044 ) (0.078 ) (0.101 ) (0.093 ) (0.121)*** Tea * *** **

-0.455 -2.928 1.349 0.796 -0.080 0.333 -0.450 0.227 -0.113 1.875 Cocoa (0.211) (0.589) (0.197) (0.131)*** (0.076 ) (0.113)*** (0.366 ) (0.124)* (0.136 ) (0.087)*** beverages ** *** ***

-0.306 -2.619 0.214 0.182 0.216 -0.060 0.260 0.179 -0.046 1.504 (0.096) (0.174) (0.059) (0.064) (0.063)*** (0.036)* (0.052)*** (0.130 ) (0.091 ) (0.053)*** Mineral water *** *** *** ***

-0.456 -0.340 -0.978 0.281 0.116 0.103 0.120 0.079 -0.249 0.674 Non-alcoholic (0.091) (0.062) (0.054) (0.052) (0.046)** (0.068 ) (0.024)*** (0.037)** (0.117)** (0.027)*** soft drinks *** *** *** ***

Fruit juices and 0.352 -1.823 -1.202 -0.701 -0.718 0.122 0.178 -0.176 0.081 0.473 vegetables (0.103) (0.196) (0.137) (0.112) (0.087) (0.059)** (0.037)*** (0.059)*** (0.073 ) (0.039)*** juices *** *** *** *** ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

147

Chapter IV

Chapter IV IV Chapter Chapter Table 18: Conditional, Marshallian price and income elasticities at stage 2, light

drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.213 -0.522 -0.110 0.034 -0.129 -0.626 0.311 0.369 -0.061 0.807 (0.238) (0.151) (0.121) (0.181) (0.060)** (1.064) (0.693) (0.298) (0.662) (0.475)* Wine *** ***

-1.191 0.726 -0.003 0.571 -1.759 -2.835 -1.117 0.451 1.034 0.611 (0.471) (0.776) (0.237) (0.229)** (2.390) (1.551)* (1.276) (0.656) (0.972) (0.694) Beer **

-0.863 0.429 0.025 0.138 3.063 0.466 -0.847 0.389 0.348 0.867 (0.196) (0.440) (0.157) (0.091) (1.169)*** (1.253) (0.656) (0.305) (0.778) (0.530) Spirits ***

0.422 0.195 0.032 -0.705 -3.048 -1.698 -0.506 0.157 0.649 0.740 Coffee (0.374) (0.153) (0.146) (0.112)*** (1.326)** (0.701)** (0.620) (0.292) (0.582) (0.397)*

2.414 -0.774 -0.913 -0.390 -0.076 -0.568 -4.408 -0.391 -1.236 1.138 (1.039) (0.360) (0.626) (0.217)* (0.381) (0.189)*** (2.891) (1.413) (1.559) (1.098) Tea ** **

Cocoa 0.783 -0.251 0.004 0.059 -3.827 -0.911 -1.002 0.285 -0.727 1.938 beverages (0.452)* (0.597) (0.368) (0.260) (1.626)** (1.197) (0.936) (0.449) (0.938) (0.864)**

0.013 -0.743 -0.113 -0.620 -6.344 1.184 0.220 -0.317 -1.452 2.692 Mineral water (0.790) (0.660) (0.298) (0.341)* (2.598)** (1.132) (1.565) (0.659) (0.878)* (0.507)***

-1.090 0.336 -0.235 0.125 -0.014 0.095 -0.036 -0.318 0.080 0.603 (0.227) (0.138) Non-alcoholic (0.226) (0.223) (0.095) (0.120) (0.330) (0.331) (0.449) (0.138)*** soft drinks *** **

Fruit juices and -1.509 0.193 0.399 0.110 0.035 -0.750 -0.291 0.012 -0.228 0.286 vegetables (0.453) (0.246) (0.267) (0.153) (0.118) (1.090) (0.767) (0.212) (0.508) (0.323) juices ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

148

Consumer demand for alcoholic beverages

Table 19: Conditional, Hicksian price and income elasticities at stage 2, light drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.093 -0.354 -0.069 0.070 0.004 -0.571 0.334 0.485 0.057 0.807 (0.195) (0.139) (0.126) (0.201) (0.091) (1.032) (0.705) (0.347) (0.593) (0.475)* Wine *** **

-1.160 0.817 0.024 0.671 -1.718 -2.817 -1.030 0.578 1.123 0.611 (0.451) (0.685) (0.263) (0.165)*** (2.345) (1.568)* (1.362) (0.531) (0.873) (0.694) Beer ***

-0.825 0.569 0.558 0.070 0.280 3.121 0.491 -0.723 0.474 0.867 (0.218) (0.203) (0.365) (0.135) (0.045)*** (1.144)*** (1.268) (0.729) (0.702) (0.530) Spirits *** ***

0.532 0.233 0.065 -0.583 -2.998 -1.677 -0.399 0.311 0.757 0.740 Coffee (0.321)* (0.141)* (0.161) (0.065)*** (1.300)** (0.712)** (0.667) (0.220) (0.525) (0.397)*

2.578 -0.537 -0.744 -0.332 -0.026 -0.381 -4.332 -0.358 -1.071 1.138 (1.145) (0.273) (0.502) (0.215) (0.427) (0.154)** (2.820) (1.443) (1.401) (1.098) Tea ** **

Cocoa 1.071 -0.152 0.088 0.377 -3.698 -0.856 -0.723 0.688 -0.445 1.938 beverages (0.464)** (0.612) (0.400) (0.345) (1.576)** (1.211) (0.913) (0.564) (0.819) (0.864)**

0.412 -0.605 0.005 -0.178 -6.164 1.261 0.606 0.243 -1.060 2.692 Mineral water (0.742) (0.652) (0.309) (0.300) (2.568)** (1.139) (1.598) (0.595) (0.812) (0.507)***

-0.964 0.424 -0.146 0.156 0.013 0.194 0.004 -0.301 0.167 0.603 (0.248) (0.127) Non-alcoholic (0.240) (0.227) (0.098) (0.135) (0.331) (0.330) (0.437) (0.138)*** soft drinks *** ***

Fruit juices and -1.501 0.235 0.414 0.123 0.082 -0.731 -0.250 0.072 -0.187 0.286 vegetables (0.452) (0.263) (0.271) (0.163) (0.117) (1.072) (0.753) (0.244) (0.464) (0.323) juices ***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

149

Chapter IV

Chapter IV IV Chapter Chapter Table 20: Conditional, Marshallian price and income elasticities at stage 2, moderate

drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.053 0.102 0.052 -0.011 0.352 -0.063 -0.202 -0.024 0.006 1.402 Wine (1.248) (1.179) (0.653) (0.536) (0.598) (0.541) (0.405) (0.365) (0.212) (0.456)***

0.322 -0.900 0.215 -0.035 1.231 -0.795 -0.171 0.121 -0.059 1.141 Beer (0.916) (1.269) (0.495) (0.615) (1.250) (0.540) (0.419) (0.332) (0.245) (1.126)

0.076 0.111 -0.907 -0.083 -0.779 -0.408 -0.048 -0.106 -0.088 0.827 Spirits (0.822) (0.470) (0.392)** (0.229) (0.499) (0.300) (0.276) (0.244) (0.130) (0.460)*

0.013 0.010 0.067 -0.827 -1.709 -0.244 0.181 -0.067 0.110 0.075 Coffee (1.317) (1.551) (0.733) (0.638) (1.641) (1.180) (0.448) (0.421) (0.357) (0.567)

0.177 -0.015 0.002 -0.233 -1.323 -1.340 0.560 0.189 -0.149 0.314 Tea (2.117) (1.613) (1.048) (0.744) (1.482) (1.004) (0.647) (0.640) (0.412) (0.797)

Cocoa 0.404 -1.127 -0.065 -0.066 -2.224 1.837 -0.115 -0.222 0.373 -0.242 beverages (2.323) (2.405) (1.018) (1.427) (3.153) (3.374) (1.021) (1.169) (0.851) (2.003)

-0.521 -0.075 -0.141 -0.202 -0.619 -0.065 -0.418 0.105 0.372 0.822 Mineral water (1.464) (0.914) (0.631) (0.512) (0.737) (0.618) (1.194) (0.629) (0.578) (0.661)

Non-alcoholic -0.031 -0.076 -0.075 0.004 -0.649 0.002 0.095 -0.897 -0.008 0.636 soft drinks (1.312) (0.890) (0.609) (0.431) (0.594) (0.574) (0.512) (0.486)* (0.287) (0.546)

Fruit juices and -0.903 vegetables 0.180 -0.101 -0.078 -0.033 -2.189 0.321 0.250 -0.168 (0.363) 0.654 juices (0.847) (0.948) (0.414) (0.443) (0.851)*** (0.456) (0.614) (0.334) ** (0.427)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

150

Consumer demand for alcoholic beverages

Table 21: Conditional, Hicksian price and income elasticities at stage 2, moderate drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-0.479 0.250 0.262 0.114 0.385 -0.049 -0.105 0.098 0.086 1.402 Wine (1.413) (1.155) (0.606) (0.546) (0.599) (0.540) (0.415) (0.347) (0.200) (0.456)***

0.790 -0.780 0.385 0.067 1.258 -0.784 -0.092 0.220 0.005 1.141 Beer (1.218) (1.184) (0.534) (0.686) (1.275) (0.540) (0.380) (0.339) (0.210) (1.126)

0.415 0.199 -0.784 -0.009 -0.759 -0.400 0.009 -0.034 -0.041 0.827 Spirits (0.950) (0.452) (0.350)** (0.239) (0.505) (0.298) (0.284) (0.229) (0.119) (0.460)*

0.044 0.018 0.079 -0.821 -1.707 -0.243 0.187 -0.060 0.114 0.075 Coffee (1.412) (1.579) (0.685) (0.608) (1.646) (1.179) (0.462) (0.397) (0.354) (0.567)

0.306 0.018 0.049 -0.205 -1.315 -1.337 0.582 0.217 -0.131 0.314 Tea (2.395) (1.583) (0.960) (0.753) (1.483) (1.005) (0.669) (0.601) (0.390) (0.797)

Cocoa 0.304 -1.153 -0.101 -0.088 -2.229 1.834 -0.132 -0.243 0.359 -0.242 beverages (2.458) (2.478) (1.038) (1.426) (3.185) (3.373) (1.012) (1.160) (0.820) (2.003)

-0.184 0.012 -0.018 -0.128 -0.599 -0.057 -0.361 0.177 0.419 0.822 Mineral water (1.702) (0.872) (0.561) (0.517) (0.748) (0.615) (1.224) (0.590) (0.553) (0.661)

Non-alcoholic 0.230 -0.009 0.020 0.061 -0.633 0.008 0.139 -0.842 0.028 0.636 soft drinks (1.494) (0.870) (0.553) (0.429) (0.597) (0.573) (0.533) (0.453)* (0.274) (0.546)

Fruit juices and -0.866 vegetables 0.449 -0.031 0.019 0.026 -2.174 0.327 0.295 -0.111 (0.349) 0.654 juices (0.970) (0.937) (0.384) (0.449) (0.853)** (0.454) (0.629) (0.318) ** (0.427)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

151

Chapter IV

Chapter IV IV Chapter Chapter Table 22: Conditional, Marshallian price and income elasticities at stage 2, heavy

drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

-1.041 0.637 (0.216) 0.477 0.031 0.192 1.647 -1.381 -0.417 0.052 (0.307) 1.313 Wine *** (0.539) (0.238) (0.101)* (1.069) (0.802)* (0.243)* (0.159) ** (0.214)***

0.069 -0.379 -0.019 0.517 4.217 -3.155 -0.992 0.330 1.549 1.838 Beer (0.757) (1.932) (0.828) (0.546) (5.196) (4.126) (1.431) (0.418) (1.917) (0.996)*

-0.997 0.152 -0.718 (0.359) -0.255 -3.294 1.761 0.573 -0.028 -1.062 0.464 Spirits (0.384) (0.871) *** (0.216) (1.767)* (1.340) (0.448) (0.281) (0.558)* (0.311)

-0.186 -0.498 0.130 -0.990 -3.165 1.209 0.517 0.055 -0.853 0.635 Coffee (0.280) (0.778) (0.336) (0.295)*** (2.288) (1.723) (0.591) (0.192) (0.837) (0.438)

0.097 -1.332 0.143 -0.315 -2.553 3.412 1.206 -0.446 -0.835 0.336 Tea (0.591) (1.175) (0.593) (0.389) (3.894) (2.867) (0.802) (0.549) (1.028) (0.655)

Cocoa -0.090 -0.470 0.007 -0.279 -6.907 -1.608 0.150 0.032 -0.824 0.554 beverages (0.612) (1.092) (0.647) (0.611) (5.154) (3.763) (1.410) (0.571) (1.986) (0.825)

-0.127 -0.480 -0.142 -0.368 -2.163 1.586 -0.598 -0.305 0.044 0.518 Mineral water (0.365) (0.777) (0.383) (0.436) (2.183) (1.656) (0.781) (0.352) (1.094) (0.455)

-1.423 Non-alcoholic 0.282 -0.231 -0.016 -0.250 -2.590 1.936 0.395 (0.236) -0.895 0.506 soft drinks (0.333) (0.982) (0.404) (0.303) (2.422) (1.950) (0.655) *** (0.932) (0.491)

Fruit juices and vegetables 0.029 -1.007 -0.168 -0.686 -2.286 1.582 1.605 0.124 -2.118 0.514 juices (0.512) (1.272) (0.513) (0.576) (3.843) (3.073) (1.360) (0.343) (1.702) (0.782)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

152

Consumer demand for alcoholic beverages

Table 23: Conditional, Hicksian price and income elasticities at stage 2, heavy drinkers (at sample means)

Win Bee Sp Cof Tea CBe MWa NaS FJVJ Income

0.690 -0.402 0.605 0.266 0.273 1.666 -1.373 -0.354 0.136 (0.313) 1.313 Wine (0.311) (0.557) (0.204) (0.104)*** (1.072) (0.801)* (0.238) (0.150) ** (0.214)***

0.965 -0.200 0.310 0.632 4.243 -3.144 -0.905 0.449 1.625 1.838 Beer (1.202) (2.026) (0.670) (0.595) (5.210) (4.121) (1.390) (0.390) (1.953) (0.996)*

-0.914 0.378 -0.673 (0.315) -0.226 -3.288 1.764 0.595 0.002 -1.043 0.464 Spirits (0.511) (0.897) *** (0.219) (1.770)* (1.339) (0.441) (0.270) (0.564)* (0.311)

0.123 -0.436 0.244 -0.951 -3.156 1.213 0.547 0.096 -0.827 0.635 Coffee (0.448) (0.819) (0.275) (0.315)*** (2.294) (1.720) (0.572) (0.188) (0.853) (0.438)

0.261 -1.299 0.203 -0.294 -2.548 3.414 1.222 -0.424 -0.821 0.336 Tea (0.828) (1.232) (0.522) (0.413) (3.903) (2.864) (0.785) (0.525) (1.048) (0.655)

Cocoa 0.180 -0.416 0.106 -0.244 -6.899 -1.605 0.176 0.068 -0.802 0.554 beverages (0.870) (1.159) (0.569) (0.652) (5.164) (3.760) (1.375) (0.571) (2.013) (0.825)

0.125 -0.430 -0.049 -0.336 -2.155 1.589 -0.573 -0.271 0.065 0.518 Mineral water (0.466) (0.783) (0.343) (0.455) (2.189) (1.654) (0.765) (0.356) (1.109) (0.455)

-1.390 Non-alcoholic 0.529 -0.181 0.074 -0.218 -2.582 1.939 0.419 (0.227) -0.875 0.506 soft drinks (0.542) (1.027) (0.332) (0.327) (2.428) (1.947) (0.637) *** (0.949) (0.491)

Fruit juices and vegetables 0.279 -0.956 -0.076 -0.654 -2.278 1.585 1.630 0.157 -2.097 0.514 juices (0.822) (1.339) (0.410) (0.619) (3.854) (3.068) (1.328) (0.348) (1.732) (0.782)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Note: Win: Wine, Bee: Beer, Sp: Spirits, Cof: Coffee, Tea: Tea, CBe: Cocoa beverages, MWa: Mineral Water, NaS: Non-alcoholic soft drinks, FJVJ: Fruit juices and vegetables juices

153

Chapter IV

Chapter IV IV Chapter Chapter Table 24: Parameter estimates of the budget share equation, Stage 1

Coeff. for equation beverages Coeff. for equation other products

Coeff. for equation food (bootsrap (bootsrap standard errors in and services (bootsrap standard

Independent variables standard errors in parentheses) parentheses) errors in parentheses)

0.805 (0.184)*** 0.125 (0.109) 0.070 (0.167) Constant

0.005 (0.013) -0.003 (0.009) -0.002 (0.016) Logprice food

-0.003 (0.009) -0.005 (0.012) 0.008 (0.008) Logprice beverages

-0.002 (0.016) 0.008 (0.008) -0.006 (0.019) Logprice other products and services

-0.052 (0.001)*** -0.004 (0.000)*** 0.056 (0.001)*** Log (expenditure/a(p))

0.002 (0.001)*** 0.000 (0.000) -0.002 (0.001)*** (Log (expenditure/a(p)))^2

0.048 (0.001)*** 0.003 (0.000)*** -0.051 (0.001)*** Household size in adult equivalents

0.004 (0.001)*** -0.001 (0.000)*** -0.003 (0.001)*** Young children (<=5 years, yes or no)

0.001 (0.000)*** 0.000 (0.000)*** -0.001 (0.000)*** Age of the household‘s reference person

University degree of the household‘s reference person 0.001 (0.000) -0.001 (0.000)*** 0.000 (0.001) (yes or no)

0.001 (0.000)*** 0.000 (0.000)*** -0.001 (0.000)*** Month

0.000 (0.000)*** 0.000 (0.000) 0.001 (0.000)*** Year

0.015 (0.001)*** 0.003 (0.000)*** -0.018 (0.001)*** Residuals

R2 0.86 0.47 --

-test for the whole demand system (Wald test) 240,000***

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

154

Table 25: Parameter estimates of the budget share equation, Stage 2, all households

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

Independent variables parentheses

-0.126 (0.315) 0.136 (0.259) 0.156 (0.112) -0.072 (0.151) 0.659 (0.205)*** -0.584 (0.363) -0.422 (0.223)* 0.011 (0.099) -0.213 (0.151) Constant

0.081 (0.108) 0.002 (0.083) 0.071 (0.022)*** -0.049 (0.052) -0.013 (0.036) 0.068 (0.064) -0.126 (0.052)** -0.031 (0.039) -0.003 (0.035) Logprice wine

0.002 (0.083) 0.063 (0.062) -0.028 (0.022) 0.052 (0.039) -0.019 (0.034) -0.085 (0.050)* -0.051 (0.040) 0.014 (0.029) 0.053 (0.036) Logprice beer

0.071 (0.022)*** -0.028 (0.022) 0.019 (0.011) -0.020 (0.010)** -0.024 (0.016) 0.003 (0.022) -0.016 (0.018) -0.006 (0.009) 0.002 (0.014) Logprice spirits

-0.049 (0.052) 0.052 (0.039) -0.020 (0.010)** 0.062 (0.026)** -0.022 (0.021) -0.005 (0.035) 0.000 (0.027) 0.010 (0.016) -0.028 (0.025) Logprice coffee

-0.013 (0.036) -0.019 (0.034) -0.024 (0.016) -0.022 (0.021) -0.007 (0.044) -0.085 (0.047)* 0.121 (0.024)*** 0.030 (0.020) 0.018 (0.019) Logprice tea

0.068 (0.064) -0.085 (0.050)* 0.003 (0.022) -0.005 (0.035) -0.085 (0.047)* 0.124 (0.059)** 0.058 (0.031)* -0.023 (0.028) -0.054 (0.027)** alcoh for demand Consumer Logprice cocoa beverages

-0.126 (0.052)** -0.051 (0.040) -0.016 (0.018) 0.000 (0.027) 0.121 (0.024)*** 0.058 (0.031)* 0.077 (0.037)** -0.027 (0.017) -0.035 (0.030) Logprice mineral water

-0.031 (0.039) 0.014 (0.029) -0.006 (0.009) 0.010 (0.016) 0.030 (0.020) -0.023 (0.028) -0.027 (0.017) -0.008 (0.015) 0.042 (0.017)** Logprice non-alcoholic soft drinks

Logprice fruit juices and vegetables -0.003 (0.035) 0.053 (0.036) 0.002 (0.014) -0.028 (0.025) 0.018 (0.019) -0.054 (0.027)** -0.035 (0.030) 0.042 (0.017)** 0.005 (0.019) juices

0.271 (0.093)*** -0.178 (0.022)*** 0.059 (0.030)** -0.134 (0.044)*** -0.074 (0.028)*** 0.165 (0.019)*** 0.110 (0.024)*** -0.084 (0.012)*** -0.106 (0.014)*** Log (expenditure/a(p))

0.059 (0.012)*** -0.024 (0.005)*** -0.021 (0.005)*** -0.026 (0.006)*** 0.020 (0.005)*** -0.012 (0.004)*** -0.009 (0.003)*** -0.003 (0.002)* -0.005 (0.003)* (Log (expenditure/a(p)))^2

-0.197 (0.007)*** 0.082 (0.012)*** -0.044 (0.015)*** 0.108 (0.005)*** 0.063 (0.007)*** -0.070 (0.024)*** -0.077 (0.009)*** 0.093 (0.007)*** 0.065 (0.006)***

Household size in adult equivalents olic bevrages olic

155

156

0.006 (0.006) -0.022 (0.007)*** 0.007 (0.008) -0.020 (0.005)*** -0.009 (0.005)* 0.058 (0.008)*** -0.011 (0.006)* -0.028 (0.004)*** 0.004 (0.005)

Young children (<=5 years, yes or no) IV Chapter

0.001 (0.000)*** 0.001 (0.000)*** 0.000 (0.000) 0.006 (0.000)*** 0.001 (0.000)*** 0.001 (0.000)** -0.001 (0.000)*** -0.004 (0.000)*** -0.002 (0.000)*** Age of the household‘s reference person

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

Independent variables parentheses

University degree of the household‘s 0.022 (0.006)*** 0.026 (0.008)*** -0.011 (0.008) -0.035 (0.005)*** 0.040 (0.006)*** -0.003 (0.008) -0.047 (0.006)*** -0.053 (0.004)*** 0.078 (0.005)*** reference person (yes or no)

-0.002 (0.001)** 0.002 (0.001)*** 0.001 (0.001) 0.000 (0.001) 0.000 (0.000) -0.003 (0.001)*** -0.003 (0.001)*** 0.000 (0.000) 0.000 (0.001) Month

0.000 (0.000) 0.000 (0.000) 0.000 (0.000)** 0.000 (0.000)*** 0.000 (0.000) 0.000 (0.000)** 0.000 (0.000) 0.000 (0.000) 0.000 (0.000)*** Year

-0.052 (0.012)*** 0.232 (0.019)*** 0.001 (0.026) 0.070 (0.009)*** 0.013 (0.011) -0.169 (0.030)*** -0.172 (0.016)*** 0.064 (0.007)*** 0.045 (0.009)*** Residuals

0.508 (0.008)*** 0.130 (0.024)*** -0.001 (0.021) 0.487 (0.015)*** 0.193 (0.013)*** 0.081 (0.018)*** 0.441 (0.012)*** 0.285 (0.009)*** 0.358 (0.013)*** pdf

R2 0.53 0.18 0.20 0.35 0.20 0.08 0.31 0.43 0.34

-test for the whole demand system 13,998*** (Wald test) *, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Table 26: Parameter estimates of the budget share equation, Stage 2, light drinking household

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

parentheses Independent variables

0.846 (0.565) -0.021 (0.653) -0.529 (0.276)* -0.189 (0.366) 2.835 (0.939)*** -1.492 (1.590) 0.102 (1.572) 0.506 (0.760) 0.463 (0.184)** Constant

-0.057 (0.127) -0.042 (0.064) 0.044 (0.025)* -0.028 (0.029) -0.131 (0.096) 0.321 (0.210) 0.204 (0.138) -0.201 (0.119)* -0.110 (0.058)* Logprice wine

-0.042 (0.064) 0.089 (0.111) 0.031 (0.045) 0.101 (0.064) -0.096 (0.038)** -0.048 (0.155) -0.166 (0.231) 0.034 (0.095) 0.097 (0.052)* Logprice beer

0.044 (0.025)* 0.031 (0.045) 0.067 (0.025)*** 0.043 (0.034) -0.030 (0.036) -0.079 (0.037)** -0.226 (0.109)** 0.045 (0.029) 0.105 (0.043)** Logprice spirits

-0.028 (0.029) 0.101 (0.064) 0.043 (0.034) 0.092 (0.052)* -0.123 (0.040)*** 0.016 (0.075) -0.166 (0.150) 0.036 (0.050) 0.029 (0.049) Logprice coffee

-0.131 (0.096) -0.096 (0.038)** -0.030 (0.036) -0.123 (0.040)*** 0.075 (0.111) 0.125 (0.127) 0.466 (0.130)*** -0.101 (0.070) -0.186 (0.072)***

Logprice tea Consumer demand for alcoh for demand Consumer 0.321 (0.210) -0.048 (0.155) -0.079 (0.037)** 0.016 (0.075) 0.125 (0.127) -0.235 (0.436) -0.364 (0.347) 0.184 (0.238) 0.080 (0.024)*** Logprice cocoa beverages

0.204 (0.138) -0.166 (0.231) -0.226 (0.109)** -0.166 (0.150) 0.466 (0.130)*** -0.364 (0.347) 0.086 (0.477) 0.117 (0.233) 0.048 (0.065) Logprice mineral water

-0.201 (0.119)* 0.034 (0.095) 0.045 (0.029) 0.036 (0.050) -0.101 (0.070) 0.184 (0.238) 0.117 (0.233) -0.073 (0.128) -0.040 (0.022)* Logprice non-alcoholic soft drinks

Logprice fruit juices and vegetables -0.110 (0.058)* 0.097 (0.052)* 0.105 (0.043)** 0.029 (0.049) -0.186 (0.072)*** 0.080 (0.024)*** 0.048 (0.065) -0.040 (0.022)* -0.023 (0.040)

juices

-0.019 (0.129) -0.143 (0.114) -0.091 (0.107) -0.129 (0.096) 0.092 (0.160) 0.151 (0.138) 0.452 (0.109)*** -0.110 (0.039)*** -0.182 (0.076)** Log (expenditure/a(p))

0.010 (0.011) -0.010 (0.011) -0.010 (0.008) -0.009 (0.007) 0.013 (0.008) -0.009 (0.010) 0.005 (0.004) 0.002 (0.002) -0.002 (0.003) (Log (expenditure/a(p)))^2

0.039 (0.015)*** 0.070 (0.024)*** -0.020 (0.016) 0.052 (0.014)*** 0.080 (0.010)*** -0.137 (0.029)*** -0.263 (0.033)*** 0.145 (0.014)*** 0.083 (0.038)** bevrages olic Household size in adult equivalents

157

158

-0.027 (0.010)*** -0.001 (0.008) 0.002 (0.009) 0.000 (0.006) -0.015 (0.005)*** 0.051 (0.008)*** -0.001 (0.006) -0.036 (0.006)*** 0.002 (0.006)

Young children (<=5 years, yes or no) IV Chapter

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

Independent variables parentheses

0.000 (0.000) 0.000 (0.000) 0.001 (0.000) 0.007 (0.000)*** 0.001 (0.000)*** 0.001 (0.000)*** 0.000 (0.000) -0.004 (0.000)*** -0.001 (0.000)*** Age of the household‘s reference person

University degree of the household‘s 0.062 (0.008)*** 0.027 (0.010)*** 0.011 (0.011) -0.072 (0.007)*** 0.041 (0.005)*** -0.013 (0.008) -0.059 (0.006)*** -0.068 (0.005)*** 0.089 (0.007)*** reference person (yes or no)

0.001 (0.001) 0.001 (0.001) 0.000 (0.001) -0.001 (0.001) -0.001 (0.001) -0.002 (0.001)** 0.000 (0.001) 0.001 (0.001) 0.000 (0.001) Month

-0.001 (0.000)*** 0.000 (0.000) 0.000 (0.000)** 0.000 (0.000)*** -0.001 (0.000)*** 0.002 (0.000)** 0.002 (0.001)*** -0.001 (0.000)*** -0.001 (0.000)** Year

0.345 (0.021)*** 0.038 (0.040) -0.018 (0.034) -0.024 (0.019) 0.045 (0.014)*** -0.318 (0.036)*** -0.398 (0.051)*** 0.141 (0.015)*** 0.112 (0.057)** Residuals

0.383 (0.013)*** 0.390 (0.055)*** -0.054 (0.039) 0.635 (0.019)*** 0.212 (0.017)*** 0.084 (0.024)*** 0.555 (0.024)*** 0.352 (0.018)*** 0.375 (0.020)*** pdf

R2 0.34 0.14 0.11 0.36 0.20 0.08 0.31 0.45 0.34

-test for the whole demand system 5910*** (Wald test) *, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Table 27: Parameter estimates of the budget share equation, Stage 2, moderate drinking households

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

Independent variables parentheses

-0.024 (1.011) -0.332 (1.407) -0.101 (0.559) 0.013 (0.697) -1.043 (1.558) 0.650 (1.435) 0.184 (0.485) 0.019 (0.466) 0.147 (0.369) Constant

-0.032 (0.406) 0.001 (0.204) 0.045 (0.154) -0.012 (0.112) 0.013 (0.084) 0.039 (0.140) -0.068 (0.156) -0.012 (0.131) 0.025 (0.061) Logprice wine

0.001 (0.204) -0.012 (0.369) 0.033 (0.130) 0.025 (0.146) 0.011 (0.048) -0.059 (0.131) 0.001 (0.079) 0.014 (0.070) -0.013 (0.038) Logprice beer

0.045 (0.154) 0.033 (0.130) 0.010 (0.066) -0.016 (0.053) -0.004 (0.033) -0.016 (0.048) -0.017 (0.063) -0.019 (0.047) -0.016 (0.022) Logprice spirits

-0.012 (0.112) 0.025 (0.146) -0.016 (0.053) 0.031 (0.080) -0.011 (0.024) -0.006 (0.068) -0.020 (0.042) 0.007 (0.040) 0.002 (0.023) Logprice coffee

0.013 (0.084) 0.011 (0.048) -0.004 (0.033) -0.011 (0.024) 0.049 (0.061) -0.083 (0.075) 0.026 (0.031) 0.013 (0.032) -0.013 (0.025) Logprice tea

0.039 (0.140) -0.059 (0.131) -0.016 (0.048) -0.006 (0.068) -0.083 (0.075) 0.146 (0.162) -0.020 (0.051) -0.017 (0.056) 0.015 (0.034) alcoh for demand Consumer Logprice cocoa beverages

-0.068 (0.156) 0.001 (0.079) -0.017 (0.063) -0.020 (0.042) 0.026 (0.031) -0.020 (0.051) 0.061 (0.134) 0.010 (0.065) 0.026 (0.057) Logprice mineral water

-0.012 (0.131) 0.014 (0.070) -0.019 (0.047) 0.007 (0.040) 0.013 (0.032) -0.017 (0.056) 0.010 (0.065) 0.011 (0.061) -0.008 (0.030) Logprice non-alcoholic soft drinks

Logprice fruit juices and vegetables 0.025 (0.061) -0.013 (0.038) -0.016 (0.022) 0.002 (0.023) -0.013 (0.025) 0.015 (0.034) 0.026 (0.057) -0.008 (0.030) -0.019 (0.031) juices

0.190 (0.225) 0.060 (0.204) -0.052 (0.109) -0.142 (0.079)* -0.040 (0.048) -0.120 (0.128) -0.018 (0.073) -0.040 (0.065) -0.033 (0.040) Log (expenditure/a(p))

0.002 (0.021) -0.019 (0.021) 0.005 (0.018) 0.007 (0.011) 0.001 (0.004) 0.013 (0.030) -0.001 (0.006) -0.002 (0.006) 0.001 (0.004) (Log (expenditure/a(p)))^2

-0.148 (0.031)*** 0.067 (0.021)*** -0.036 (0.018)** 0.057 (0.021)*** 0.015 (0.009)* -0.017 (0.033) 0.016 (0.014) 0.096 (0.014)*** 0.028 (0.012)**

Household size in adult equivalents olic bevrages olic

159

160

0.038 (0.012)*** 0.013 (0.011) -0.009 (0.010) -0.016 (0.007)** -0.006 (0.006) 0.022 (0.008)*** -0.021 (0.009)** -0.031 (0.009)*** 0.006 (0.006)

Young children (<=5 years, yes or no) IV Chapter

0.003 (0.000)*** -0.002 (0.000)*** -0.001 (0.000)*** 0.002 (0.000)*** 0.000 (0.000) 0.001 (0.000)* 0.000 (0.000) -0.002 (0.000)*** -0.001 (0.000)*** Age of the household‘s reference person

Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for equation

wine (bootstrap beer (bootstrap spirits (bootstrap coffee (bootstrap tea (bootstrap cocoa beverages mineral water non-alcoholic soft fruit juices and

standard errors in standard errors in standard errors in standard errors in standard errors in (bootstrap standard (bootstrap standard drinks. (bootstrap vegetables juices

parentheses) parentheses) parentheses) parentheses) parentheses) errors in errors in standard errors in (bootstrap standard

parentheses parentheses parentheses errors in

Independent variables parentheses

University degree of the household‘s 0.007 (0.016) -0.013 (0.013) 0.024 (0.013)* -0.017 (0.010) 0.011 (0.005)** -0.011 (0.011) -0.017 (0.008)** -0.018 (0.007)*** 0.024 (0.006)*** reference person (yes or no)

-0.002 (0.001) -0.007 (0.001)*** 0.008 (0.001)*** 0.000 (0.001) 0.000 (0.000) -0.001 (0.001) -0.001 (0.001) 0.000 (0.001) -0.001 (0.001) Month

0.000 (0.001) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.001 (0.001) 0.000 (0.001) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) Year

-0.082 (0.061) 0.043 (0.031) -0.081 (0.032)** 0.097 (0.034)*** 0.009 (0.014) -0.021 (0.045) 0.016 (0.024) 0.062 (0.024)*** 0.003 (0.019) Residuals

0.431 (0.029)*** 0.309 (0.049)*** 0.214 (0.024)*** 0.230 (0.026)*** 0.032 (0.020) -0.002 (0.022) 0.109 (0.025)*** 0.132 (0.018)*** 0.122 (0.024)*** pdf

R2 0.27 0.16 0.19 0.13 0.06 0.04 0.11 0.13 0.10

-test for the whole demand system 1657*** (Wald test)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Table 28: Parameter estimates of the budget share equation, Stage 2, heavy drinking households

Coeff. for equation Coeff. for Coeff. for equation Coeff. for Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for Coeff. for

wine (bootstrap equation beer spirits (bootstrap equation coffee tea (bootstrap cocoa beverages mineral water equation non- equation fruit

standard errors in (bootstrap standard errors in (bootstrap standard errors in (bootstrap (bootstrap alcoholic soft juices and

parentheses) standard errors parentheses) standard errors parentheses) standard errors in standard errors in drinks. vegetables juices

in parentheses) in parentheses) parentheses parentheses (bootstrap (bootstrap

standard errors standard errors in

Independent variables in parentheses parentheses

-0.339 (0.708) -1.522 (0.724)** 0.406 (0.661) -0.597 (0.432) -4.509 (1.757)*** 4.270 (2.229)* 1.371 (0.888) -0.071 (0.498) -1.983 (1.108)* Constant

-0.155 (0.127) -0.150 (0.121) 0.147 (0.102) 0.012 (0.036) 0.019 (0.029) 0.021 (0.035) 0.013 (0.033) 0.059 (0.043) 0.033 (0.025) Logprice wine

-0.150 (0.121) -0.248 (0.208) 0.133 (0.104) 0.040 (0.048) 0.010 (0.044) 0.064 (0.077) 0.042 (0.059) 0.093 (0.057) 0.016 (0.067) Logprice beer

0.147 (0.102) 0.133 (0.104) -0.111 (0.100) -0.012 (0.024) -0.016 (0.027) -0.025 (0.025) -0.036 (0.023) -0.040 (0.032) -0.040 (0.022)* Logprice spirits

0.012 (0.036) 0.040 (0.048) -0.012 (0.024) 0.010 (0.022) -0.001 (0.011) -0.004 (0.017) -0.010 (0.014) -0.001 (0.020) -0.034 (0.021) Logprice coffee

0.019 (0.029) 0.010 (0.044) -0.016 (0.027) -0.001 (0.011) -0.001 (0.048) -0.008 (0.054) 0.005 (0.020) -0.018 (0.018) 0.010 (0.019) Logprice tea alcoh for demand Consumer

0.021 (0.035) 0.064 (0.077) -0.025 (0.025) -0.004 (0.017) -0.008 (0.054) 0.025 (0.091) -0.034 (0.031) -0.015 (0.026) -0.024 (0.045) Logprice cocoa beverages

0.013 (0.033) 0.042 (0.059) -0.036 (0.023) -0.010 (0.014) 0.005 (0.020) -0.034 (0.031) -0.020 (0.036) -0.027 (0.021) 0.067 (0.028)** Logprice mineral water

0.059 (0.043) 0.093 (0.057) -0.040 (0.032) -0.001 (0.020) -0.018 (0.018) -0.015 (0.026) -0.027 (0.021) -0.048 (0.025)* -0.003 (0.026) Logprice non-alcoholic soft drinks

Logprice fruit juices and vegetables 0.033 (0.025) 0.016 (0.067) -0.040 (0.022)* -0.034 (0.021) 0.010 (0.019) -0.024 (0.045) 0.067 (0.028)** -0.003 (0.026) -0.025 (0.036)

juices

0.229 (0.120)* 0.237 (0.164) -0.185 (0.097)* -0.046 (0.036) -0.038 (0.032) -0.039 (0.045) -0.042 (0.032) -0.063 (0.045) -0.051 (0.043) Log (expenditure/a(p))

-0.009 (0.003)*** -0.015 (0.006)*** 0.009 (0.006) 0.002 (0.002) 0.002 (0.002) 0.002 (0.003) 0.001 (0.002) 0.003 (0.002) 0.003 (0.002)

(Log (expenditure/a(p)))^2 olic bevrages olic

161

162

-0.082 (0.019)*** -0.034 (0.011)*** -0.004 (0.008) 0.024 (0.011)** 0.000 (0.005) 0.021 (0.018) 0.016 (0.007)** 0.055 (0.007)*** 0.022 (0.009)**

Household size in adult equivalents IV Chapter

0.007 (0.015) -0.001 (0.018) -0.004 (0.012) -0.002 (0.007) -0.001 (0.005) 0.037 (0.011)*** -0.005 (0.008) -0.015 (0.009) 0.006 (0.006) Young children (<=5 years, yes or no)

Coeff. for equation Coeff. for Coeff. for equation Coeff. for Coeff. for equation Coeff. for equation Coeff. for equation Coeff. for Coeff. for

wine (bootstrap equation beer spirits (bootstrap equation coffee tea (bootstrap cocoa beverages mineral water equation non- equation fruit

standard errors in (bootstrap standard errors in (bootstrap standard errors in (bootstrap (bootstrap alcoholic soft juices and

parentheses) standard errors parentheses) standard errors parentheses) standard errors in standard errors in drinks. vegetables juices

in parentheses) in parentheses) parentheses parentheses (bootstrap (bootstrap

standard errors standard errors in

Independent variables in parentheses parentheses

0.001 (0.000)*** -0.002 (0.000)*** 0.000 (0.000) 0.001 (0.000)*** 0.000 (0.000) 0.001 (0.000)* 0.000 (0.000)*** -0.002 (0.000)*** 0.000 (0.000) Age of the household‘s reference person

University degree of the household‘s 0.040 (0.014)*** -0.034 (0.013)*** -0.004 (0.010) -0.016 (0.006)** 0.007 (0.004)* -0.009 (0.007) -0.009 (0.006) -0.021 (0.005)*** 0.020 (0.005)*** reference person (yes or no)

-0.002 (0.001)* -0.003 (0.001)*** 0.007 (0.001)*** -0.001 (0.001) 0.000 (0.000) -0.002 (0.001) 0.000 (0.001) -0.001 (0.001)* 0.000 (0.001) Month

0.000 (0.000) 0.000 (0.000) 0.000 (0.000)* 0.000 (0.000)* 0.002 (0.001)*** -0.002 (0.001)* -0.001 (0.000) 0.000 (0.000) 0.001 (0.001)** Year

0.081 (0.048)* -0.040 (0.017)** -0.019 (0.016) -0.006 (0.015) -0.014 (0.007)* -0.008 (0.018) 0.002 (0.012) -0.007 (0.010) -0.026 (0.007)*** Residuals

0.336 (0.028)*** 0.160 (0.027)*** 0.269 (0.027)*** 0.111 (0.022)*** 0.002 (0.012) 0.038 (0.020)* 0.051 (0.018)*** 0.081 (0.016)*** 0.065 (0.024)*** pdf

R2 0.80 0.31 0.50 0.35 0.18 0.09 0.35 0.41 0.32

-test for the whole demand system 1984*** (Wald test)

*, ** and *** denote significance at 10%, 5% and 1% level, respectively.

Consumer demand for alcoh for demand Consumer olic bevrages olic

163

Chapter V

Chapter IV Chapter Chapter V

Endogeneity in Censored Demand Systems

Sandro Steinbach and Matteo Aepli

ETH Zurich

Meat and milk demand elasticities demand milk and Meat

Manuscript under review, Review of Economics of the Household

164

Endogeneity in censored demand systems

Abstract

Every policy impact analysis depends on reliable supply and demand elasticities. Demand elasticities are usually determined by estimating demand systems with rolling cross-sectional survey data. However, these estimates are erroneous if endogeneity is present in the data. Our study examines different strategies to cope with endogeneity in demand system estimations. We apply the quadratic almost ideal demand system to a rolling cross-sectional household dataset with monthly expenditure data for bread and cereal products in Switzerland for 2004–2009. Our results show that endogeneity is an issue that needs to be addressed. The prime strategy to deal with unobserved heterogeneity is the month fixed-effects specification. Not controlling for endogeneity can induce a severe bias in elasticity estimates.

Meat and milk demand elasticities demand milk and Meat

165

Chapter V

Chapter IV Chapter 1 Introduction

The estimation of demand elasticities with household data has gained tremendously in importance over the last few years. However, the validity of such estimates has a crucial dependence on the underlying data. Studies use four types of data, namely, cross-sectional, time-series, rolling cross-sectional, and panel data. Cross-sectional studies rely on household surveys that represent a single snapshot in time (e.g. Zheng & Henneberry, 2011; Wood et al., 2012). They allow differences between subjects to be compared but do not provide definite information about cause-and-effect relationships. Other studies use time-series data (Moschini & Moro, 1994; Holt, 1998). Such data allow changes in characteristics of subjects to be detected. Therefore, time- series data make the establishment of cause-and-effect relationships more likely. In recent years, rolling cross-sectional and panel household data have been established as the gold standard for demand elasticity estimations (Thiele, 2008; 2010; Schroeck,

2013). These allow changes in variables over time and differences between subjects elasticities demand milk and Meat to be examined. Elasticity estimates based on such data are more representative of the entire population and the influence of outliers is limited because of the large sample size. While panel data refer to multidimensional data that involve frequent measurement of the same subjects over time, rolling cross-sectional data rely on random samples at different points in time that are representative of the entire population. Most institutions that conduct household surveys collect rolling cross- sectional data because panel household data are difficult to obtain.21

Elasticity estimates with rolling cross-sectional data are more reliable than estimates with cross-sectional and time-series data. However, the use of rolling cross-sectional data exaggerates the problem of unobserved heterogeneity in the form of endogeneity, which is the correlation between an explanatory variable and the error term. It is a special case of unobserved heterogeneity, which results in erroneous

21 The collection of panel data is expensive and time consuming. Moreover, households‘ willingness to participate over longer time periods is often limited. 166

Endogeneity in censored demand systems

statistical inference, if in addition to the observed independent variables, there are other variables that are unobserved, but correlated with the observed variables. Such unobserved variables, for instance, are household demand decisions that depend not only on economic and socio-demographic factors, but also on psychological and lifestyle factors. Moreover, food demand is often characterized by seasonality, and follows certain patterns over time. Unlike panel data, rolling cross-sectional data rely on samples that vary over time; this represents an additional source of heterogeneity that has to be taken into account to obtain reliable parameter estimates and standard errors (Luttmer, 1999).

Several strategies have been used to cope with unobserved heterogeneity. However, surprisingly, most studies do not control for this source of parameter bias at all (e.g. Lambert et al., 2006; Tefera et al., 2012). We identified two different strategies to deal with unobserved heterogeneity. A first set of measures can be classified as factor

models. For instance, Mittal (2010) included a time trend for year, and Aepli and elasticities demand milk and Meat Kuhlgatz (2014) used time trends for year and month. Their approaches crucially rely on the assumption that time variation follows a linear expansion path. A second set of measures can be classified as fixed-effects (FE) specifications. For instance, Abdulai (2002) used month-FE to account for seasonality, and Blow et al. (2012) used year-FE to account for variation over time. Although both studies identified time variation as a problem, the selected approaches to cope with the problem are limiting. This is because unobserved heterogeneity is neither specific to a year nor to a season in the year. Instead, variation is possible over both dimensions. This is why we recommend the more flexible month-FE as the prime strategy to deal with unobserved variation.

We employ the quadratic almost ideal demand system (QUAIDS) to study the consequences of unconsidered endogeneity for demand system estimations. The demand system was developed by Banks et al. (1997) and is widely used to study price and income responses (e.g. Gil & Molina, 2009; Obayelu et al., 2009). It is a rank three demand system that allows general income response and approximate nonlinear

167

Chapter V

Chapter IV Chapter Engel curves to be estimated. Aggregation across consumers is possible with preferences of the generalized Gorman polar form because the population of utility maximizers is treated as if it only consists of a single representative individual. We apply QUAIDS to a rolling cross-sectional household dataset with monthly observations for bread and cereal products from the Swiss household survey for 2004–2009. To illustrate the consequences of ignoring unobserved heterogeneity, we estimate four different specifications and compare their performances. Here, the baseline QUAIDS (1), where unobserved heterogeneity is ignored, is compared with (2) the factor specification (Mittal, 2010; Aepli & Kuhlgatz, 2014), (3) the year-FE specification (Blow et al., 2012), and (4) the recommended month-FE specification.

The remainder of the paper is organized as follows: In section 2, we provide a review of the QUAIDS model. Section 3 details the methodology. Here, we discuss the problem of endogeneity and present strategies to deal with it, address the problem of

censored data, describe our econometric approach, introduce the dataset, and discuss elasticities demand milk and Meat performance measures. The final section provides concluding remarks.

2 The quadratic almost ideal demand system

QUAIDS shares the attractive properties of the almost ideal demand system (AIDS) by Deaton and Muellbauer (1980) which is applied in several studies e.g. in Balli et al. (2010) or Apps et al. (2010). It is capable of approximating any demand system arbitrarily to first order, aggregating perfectly over consumers, satisfying the axioms of choice, and imposing homogeneity and Slutsky symmetry. Both models are set apart by a quadratic log-income term. While AIDS includes a linear log-income term, QUAIDS adds a quadratic term to capture effects of nonlinear Engel curves (Lewbel, 1995). The demand system can be expressed as follows:

∑ * + , * +- (1) ( ) ( ) ( )

168

Endogeneity in censored demand systems

where n is the number of products, indexes i and j indicate specific products, is the budget share for product , m is total expenditure, and p is a price vector. We define the trans-log price aggregator ( ) and the Cobb-Douglas price aggregator ( ) as

( ) ∑ ∑ ∑ (2)

( ) ∏ (3) which makes the system of equations highly nonlinear in i and j. In order to make the model consistent with demand theory, the following parameter restrictions are imposed:

Homogeneity: ∑ (4.1)

Symmetry: (4.2)

elasticities demand milk and Meat Adding up: ∑ ; ∑ ; ∑ ; ∑ (4.3)

The fourth restriction is negativity, which is verified by signs of Hicksian own-price elasticities. The negativity restriction is necessary to guarantee the convexity of the cost function. Notably, censoring of the dependent variable commonly occurs when demand systems are estimated at the disaggregated level. We apply the method outlined in Shonkwiler and Yen (1999), who suggest using a two-step approach that accounts for censoring of the dependent variable. As a consequence, it is not possible to implement adding-up with parameter restrictions. We discuss the identification

strategy in greater detail in the econometric section (3.3).

3 Methodology

3.1 Unobserved heterogeneity in the form of endogeneity

Household survey data contain different sources of unobserved heterogeneity, which represent variations between households. First, household demand decisions depend 169

Chapter V

Chapter IV Chapter not only on economic and sociodemographic factors but also, for instance, on psychological and lifestyle factors. Second, food demand follows certain patterns over time (e.g. seasonality, time trends, shocks). Third, rolling cross-sectional data are drawn from varying samples. If an unobserved variable is correlated with explanatory variables, parameter estimates and standard errors are biased. This problem is referred to as endogeneity in the form of omitted variable bias.22 To obtain reliable parameter estimates, these sources of unobserved heterogeneity have to be taken into account.

3.2 Accounting for unobserved heterogeneity

Several strategies are commonly used in demand system estimations to account for unobserved heterogeneity. We compare the factor specification and the year-FE specification with the alternative month-FE specification; our findings are then related

to the base specification, where unobserved heterogeneity is ignored. The factor Meat and milk demand elasticities demand milk and Meat model includes two factor variables, namely, linear time trends for months and for years (Mittal, 2010; Aepli & Kuhlgatz, 2014).The year-FE specification controls for unobserved heterogeneity with a dummy for each year of the studied period (Blow et al., 2012); however, these specifications only partially account for unobserved heterogeneity. We therefore present the month-FE specification as an alternative. This contains dummies for each month, which account for a large share of unobserved heterogeneity. The three approaches are implemented by an adjustment of the error

term as follows:

Factor specification: = (5.1)

Year-FE specification: = (5.2)

Month-FE specification: = (5.3)

22 Another source of endogeneity is measurement error, which we assume is not a problem because household surveys are usually conducted with high data collection standards. 170

Endogeneity in censored demand systems

where and are factor variables for year and month, respectively, are year dummies, and are month dummies. The corrected error term is denoted by .

A further source of heterogeneity is variation among households. We consider this source of heterogeneity by including a set of household characteristics, which are incorporated by adjusting the correct error term as follows:

∑ (6) where represents household characteristics and represents respective parameter estimates. We include the following four household characteristics: (1) the household size in adult equivalents (Hagenaars et al., 1994), (2) the age of the household‘s reference person, (3) the presence of young children (≤5 years), and (4) the educational level of the household‘s reference person.

3.3 Econometric approach and elasticity calculation elasticities demand milk and Meat

Censoring due to zero consumption has both economic and econometric consequences; not accounting for it biases the parameter estimates, resulting in incorrectly computed elasticities. To account for zero consumption, Shonkwiler and Yen (1999) developed a two-step estimation procedure which is applied e.g. in Tekg (2012). The binary purchase decision is modeled in a first step, whereby the estimates from the first step are used indirectly in a second step to adjust for censoring. The first step is defined as follows23:

( ) , (7)

{ (8)

23 In contrast to univariate specifications for every product group, multivariate specification allows for dependencies between purchasing decisions of different product groups. 171

Chapter V

Chapter IV Chapter where and are dependent variables for product . The latter is a binary variable

representing the purchasing decision, while represents the corresponding latent variable, and and stand for explanatory variables. In a first step, we regress logarithmized prices and household characteristics related to the binary purchasing decision.24 These estimates are used to calculate the standard normal cumulative distribution function (cdf) ( ) and the standard normal probability density function (pdf) ( ). In a second step, cdf and pdf are introduced to the budget share equations to correct for censoring:

( ) ( ) ( ) , ( ) (9) where represents the covariance vector between error terms in the budget share equation and the multivariate probit model (7). Several econometric problems arise with this specification. First, the right-hand side ( ) ( ) ( )

does not obligatorily add up to unity; therefore, the error terms do not add up to zero Meat and milk demand elasticities demand milk and Meat across equations (Yen et al., 2002). To deal with this problem, we estimate the full system of equations instead of restricting parameters. Second, Tauchmann (2005) drew attention to heteroskedasticity, which results from introducing cdf and pdf in the second step. We correct for heteroskedasticity while estimating the variance- covariance matrix with the Huber-White-Sandwich estimator. Third, the expenditure variable is endogenous due to correlation with the error term. We correct for this additional source of endogeneity by including the residuals of the first stage as

additional right-hand-side variables in the second step.

The estimates from both steps are used to calculate conditional income elasticities with the following formula25:

( ) (10)

24 The errors and are multivariate-normal distributed, with a zero mean and a variance-covariance matrix V with diagonal elements of 1 and off-diagonal elements of . 25 The elasticities are conditional with respect to total expenditure for bread and cereal products. 172

Endogeneity in censored demand systems

26 where is the first difference of the budget share equation with respect to ( ). We consider the full effect on demand to compute price elasticities. The price effect consists of an indirect and a direct effect. The indirect effect reflects the decision concerning whether or not to purchase, and the direct effect reflects the consumption level. We calculate the Marshallian price elasticity as follows (Zheng & Henneberry, 2010):

( ) ( ) (11)

where denotes the first difference of the budget share equation with respect to 27 ( ). We denote the price parameter of the th product with respect to the th product from the multivariate probit estimation with , and the Kronecker delta with

, which is equal to unity if and zero otherwise. Lastly, the Hicksian price

elasticities are derived from the Slutsky identity as follows: Meat and milk demand elasticities demand milk and Meat

(12) where the Hicks decomposition of a demand change is calculated with reference to the Marshallian demand and the income effect.

3.4 Data description

We use a rolling cross-sectional dataset from the Swiss household expenditure survey for 2004–2009. The sample is representative for Switzerland, and the Swiss Federal

Statistical Office (SFSO) collects these data with a monthly periodicity. To account for regional differences, data are collected in seven major regions, with more than 20,000 surveyed households. The SFSO gathers detailed information on expenditures for various products and services. For most food products, information on consumed

26 , * +- ( ) ( )

27 ( ∑ ) , * +- ( ) ( ) 173

Chapter V

Chapter IV Chapter quantities is available at the household level. The survey also provides information on

income and household characteristics. We summarize the expenditure data in Table 1.

Table 1: Descriptive statistics for Swiss household expenditures on bread and cereal products in 2004–2009

Standard Share of zero Mean deviation consumption Total expenditure 8,763.29 5,255.45 0,0.00 Food 0,620.51 0,147.24 0,0.19 Bread and cereal products 0,105.84 0,073.04 0,0.55 Rice 0,002.53 00,05.63 ,68.03 Bread 0,031.59 00,24.92 0,5.73 Pasta 0,009.60 0,013.08 ,33.64 Sandwiches and other cereal products 0,058.75 0,050.93 ,03.32 Wheat flour 00,01.56 00,04.94 ,76.18 Other kinds of flour 000,2.40 00,07.21 ,71.58 Total number of households 19,593000

Notes. Expenditures in current Swiss Francs. elasticities demand milk and Meat

The share of food in total expenditures is relatively low in Switzerland compared to other high-income countries. Averaged over all households and years, we find that Swiss households spend 7.1% of their income on food. Households in other high- income countries spend substantially more on food. For instance, for 2009, the share of food in total expenditures was 11.2% in Germany and 13.8% in France (Destatis, 2010). Bread and cereal products have a large share of about a sixth of overall food expenditures in Switzerland. We consider the following expenditures from the bread

and cereal products category: rice, bread, pasta, sandwiches and other prepared products, wheat flour, and other kinds of flour. Among them, sandwiches and other prepared products represent the major expenditure, reaching about 50% of total expenditures for the bread and cereal products category. Notably, the large share of zero consumption shows that correcting for censoring is vital when it comes to obtaining reliable parameter estimates.

174

Endogeneity in censored demand systems

Price data are not collected within the household survey. We follow the approach outlined in Aepli and Finger (2013) and calculate monthly market prices for every product group at the regional level. We divide Switzerland into three regions, which represent economic and cultural zones (South, Central and East, and West).28 Unit values are computed by dividing expenditures for every household and product group by consumed quantity, adjusting for differences in quality with income and household characteristics. We normalize market prices based on the first observation month. Because data for consumed quantity are not available at the household level for sandwiches and other prepared products, we compute the series using the national- level Swiss consumer price index (CPI) for the product category (SFSO, 2013). We have developed a novel approach that uses expenditure-weighted regional variation in market prices for other bread and cereal products to obtain regional CPIs for sandwiches and other prepared products. The approach relies on the assumption that

the regional CPI variation is equal to the mean variation of other bread and cereal Meat and milk demand elasticities demand milk and Meat products.

3.5 Performance evaluation

We use five performance measures to evaluate strategies that deal with unobserved heterogeneity. The Akaike information criterion (AIC) is used to capture the information lost via each method in representing the underlying data generation process. We correct the AIC for the finite sample size and the large number of parameters. The AIC penalizes the number of parameters less than the Bayesian

information criterion (BIC) does. Hence, we also report test results for BIC. To measure the gain in explanatory power over the baseline specification, we calculate McFadden‘s R-squared, which is based on the log-likelihood kernels of the baseline and alternative specification. The corrected McFadden‘s R-squared is calculated to measure gains in efficiency due to the inclusion of additional explanatory variables. Lastly, we compare the fit of each specification over the baseline specification with the

28 We do not study food demand at the level of the seven major regions because the share of censored observations would be 100% for some regions in some months. 175

Chapter V

Chapter IV Chapter likelihood-ratio (LR) test, where the null model (baseline specification) is assumed to be a special case of the other model (alternative specification). The LR test compares log likelihoods of the two models and tests if the difference is statistically significant. If it is, then the less restrictive model fits the data substantially better than the more restrictive model. The test statistic is asymptotically chi-squared distributed, with a degree of freedom equal to the difference in the number of parameters between the two models. The test fails to reject the null hypothesis if the LR test statistic reports lower values than the critical chi-squared value at the 95% confidence level.

4 Results and discussion

We compare test statistics of the first and the second step for the four specifications in Table 2. AIC and BIC show that the month-FE specification fits the data best in the first step. We fail to reject the null hypothesis in each case where the baseline specification is equal to the alternative specifications. The month-FE specification has elasticities demand milk and Meat the largest chi-squared statistic (1,927.9), followed by the year-FE specification (1,297.0) and the factor specification (839.9). The difference in McFadden‘s R-squared confirms these results, with the month-FE specification providing the largest gain in fit. The more flexible month-FE specification far outperforms the factor specification, which indicates a problem with unobserved heterogeneity.

176

Endogeneity in censored demand systems

Table 2: Goodness-of-fit of each specification in the first and second step

First step estimation Second step estimation (1) (2) (3) (4) (1) (2) (3) (4) AIC 102,796.1 101,960.2 101,511.1 101,013.0 370,940.1 359,574.5 362,509.8 358,872.5 BIC 102,932.2 102,115.7 101,705.5 101,848.2 371,076.1 359,729.9 362,704.1 359,707.1 2 ΔR McFadden 0.000 0.008 0.013 0.019 0.000 0.031 0.023 0.033 2 Adj. R McFadden 0.000 0.008 0.012 0.017 0.000 0.031 0.023 0.032 LR test 00,000.0 00,839.9 01,297.0 01,927.9 00.000.0 11,369.4 08,442.3 12,212.4 (1.000) (0.000) (0.000) (0.000) (1.000) (0.000) (0.000) (0.000) Notes. (1) indicates the baseline specification, (2) the factor specification, (3) the year-FE specification, and (4) the month-FE specification l. The LR test reports chi-squared statistics with p-values in parentheses. ΔR2 McFadden is the increase in R2 with respect to model (1).

The test statistic for the second step conveys the same picture. The month-FE specification outperforms the other specification in fitting the data. We find a large increase in AIC and BIC due to the inclusion of month-FE. The LR test results show

that each alternative specification is statistically significantly different from the baseline Meat and milk demand elasticities demand milk and Meat specification. Notably, the factor specification performs nearly as well as the month-FE specification, while the year-FE specification is clearly outperformed by both specifications. Our results show that including variables that control for unobserved heterogeneity is a necessary step. This is because all specifications are statistically significantly different from the baseline specification.

Table 3 compares the significance level of parameter estimates, which deal with unobserved heterogeneity, for each specification. We find that almost all estimates are

significant at the 1% confidence level for the month-FE specification in the second step. The month-FE specification outperforms the year-FE specification in the relative number of significant estimates. Notably, a similar picture emerges from the first step, which we present in Table 5 in the Appendix. The significance levels of the parameter estimates for the three alternative specifications show that unobserved heterogeneity is present in the data and needs to be addressed.

177

Chapter V

Chapter IV Chapter Table 3: Number of significant parameter estimates for factors and FE at different

significance levels of the second step

(1) (2) (3) (4) <1% 000 007 012 421 <5% 000 001 000 000 <10% 000 001 006 001 >10% 000 003 018 010 Notes. See Table 2 for the definition of the models.

In Figure 1, we compare the significance level of parameter estimates for (A) price and income and (B) household characteristics. The best fit in terms of the significance level is achieved with the month-FE specification in the second step. The results for the first step are reported in Figure 2 in the Appendix. Notably, the month-FE specification is the only alternative specification that outperforms the baseline specification in terms of

significant price and income parameters. Because these parameters are crucial inputs Meat and milk demand elasticities demand milk and Meat for policy analysis, we find support for our hypothesis that controlling for unobserved heterogeneity is necessary. Estimates for the household characteristics show a similar picture. The month-FE specification outperforms all other specifications in terms of significant parameter estimates.

178

Endogeneity in censored demand systems

Figure 1: Number of significant parameter estimates for price and income parameters and household characteristics at different significance levels in the second step

60 30

50 25

40 20

30 15

20 10

10 5

0 0 (1) (2) (3) (4) (1) (2) (3) (4) <1% <5% <10% >10% <1% <5% <10% >10%

(A) Significance of price and income parameters. (B) Significance of household characteristics. Meat and milk demand elasticities demand milk and Meat Notes. See Table 2 for the definition of the models.

To illustrate the benefits of a correct control for unobserved heterogeneity, we present income elasticity estimates in Table 4. The results show endogeneity in the data. This is because the parameter estimates vary with different strategies to deal with unobserved heterogeneity. Parameter estimates for bread and wheat flour are particularly affected. The magnitude of income elasticity for bread changes from 0.569 in the baseline specification to 1.170 in the month-FE specification. For wheat flour,

we see that income elasticity changes from 0.864 in the baseline specification to 1.753 in the month-FE specification. The elasticities turn from being inelastic to elastic, which indicates that using uncorrected estimates to evaluate policies in equilibrium models can induce a huge bias. We report own- and cross-price elasticities in Tables 6–9 in the Appendix. Overall, our findings for price elasticities convey a similar picture, confirming that a better control for unobserved heterogeneity is necessary to obtain reliable elasticity estimates.

179

Chapter V

Chapter IV Chapter Table 4: Income elasticity estimates

(1) (2) (3) (4)

0.515 0.396 0.827 0.387 e1 (0.094)*** (0.462)*** (0.084)*** (0.378)*** 0.569 1.253 0.964 1.170 e2 (0.019)*** (0.095)*** (0.053)*** (0.078)*** 0.942 1.386 0.523 0.749 e3 (0.046)*** (0.161)*** (0.050)*** (0.131)*** 1.355 0.854 1.229 1.061 e4 (0.016)*** (0.059)*** (0.040)*** (0.049)*** 0.864 1.962 0.808 1.753 e5 (0.092)*** (0.333)*** (0.094)*** (0.276)*** 1.240 0.683 1.127 0.519 e6 (0.092)*** (0.416)*** (0.101)*** (0.356)*** Notes. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. We present elasticities for rice (1), bread (2), pasta (3), sandwiches and other prepared products (4), wheat flour (5), and other kinds of flour (6).

5 Concluding remarks

The estimation of demand systems with rolling cross-sectional household data has elasticities demand milk and Meat become increasingly popular. Such data allow changes in variables over time and differences between subjects to be examined. However, elasticity estimates are highly erroneous if unobserved heterogeneity in the form of endogeneity is present in the data. Correlation between an explanatory variable and the error term is problematic because it leads to biased parameter estimates and standard errors. This study examined different strategies to deal with unobserved heterogeneity. To compare the empirical performance of common strategies, we applied QUAIDS to a rolling cross-

sectional household dataset with monthly expenditure data for bread and cereal products in Switzerland for 2004–2009. We found that endogeneity is an issue that needs to be addressed. Our estimates showed that the month-FE specification outperforms the factor specification and the year-FE specification in fitting the data. The correct strategy to deal with unobserved heterogeneity in the form of endogeneity is the month-FE specification, which provides the most flexibility while correcting for unobserved heterogeneity. This is important because reliable elasticity estimates are only obtained if the estimates are corrected for endogeneity. 180

Endogeneity in censored demand systems

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Aepli, M., & Finger, R. (2013). Determinants of sheep and goat meat consumption in Switzerland. Agricultural and Food Economics. doi:10.1186/2193-7532-1-11

Aepli, M., & Kuhlgatz, C. (2014). Meat and milk demand elasticities for Switzerland: a three stage budgeting quadratic almost ideal demand system. Submitted.

Apps, P., & Rees, R. (2010). Family labor supply, taxation and saving in an imperfect capital market. Review of Economics of the Household, 8(3), 297-323.

Balli, F., & Tiezzi, S. (2010). Equivalence scales, the cost of children and household consumption patterns in Italy. Review of Economics of the Household, 8(4), 527-549.

Banks, J., Blundell, R., & Lewbel, A. (1997). Quadratic Engel curves and consumer

demand. Review of Economics and Statistics, 79(4), 527–539. Meat and milk demand elasticities demand milk and Meat

Blow, L., Lechene, V., & Levell, P. (2012). Using the CE to model household demand. NBER Chapters No. 12676, National Bureau of Economic Research, Massachusetts, USA.

Deaton, A., & Muellbauer, J. (1980). An almost ideal demand system. American Economic Review, 70(3), 312–326.

Destatis. (2010). German Federal Statistical Office, Wiesbaden. https://www.destatis.de/DE/ZahlenFakten/LaenderRegionen/Internationales/Thema/Ta bellen/Basistabelle_KonsumN.html. Accessed 31 Jan, 2014.

Gil, A.I., & Alberto Molina, J. (2009). Alcohol demand among young people in Spain: an additive QUAIDS. Empirical Economics, 36(3), 515–530.

Hagenaars, A.J.M, De Vos, K., & Zaidi, M.A. (1994). Poverty Statistics in the Late 1980s: Research Based on Micro-Data. Kontoret for De Europæiske Fællesskabers Officielle Publikationer, Luxembourg, Luxembourg.

Holt, M. (1998). Autocorrelation specification in singular equation systems: a further look. Economic Letters, 58(2), 135–141.

181

Chapter V

Chapter IV Chapter Lambert, R., Larue, B., Yélou, C., & Criner, G. (2006). Fish and meat demand in Canada: regional differences and weak separability. Agribusiness, 22(2), 175–199.

Lewbel, A. (1995). Utility functions and global regularity of fractional demand systems. International Economic Review, 36(4), 943–961.

Luttmer, E.F.P. (1999). Measuring Poverty Dynamics and Inequality in Transition Economies: Disentangling Real Events from Noisy Data. Policy research working paper, The World Bank.

Mittal, S. (2010). Application of the QUAIDS model to the food sector in India. Journal of Quantitative Economics, 8(1), 42–54.

Moschini, G., & Moro, D. (1994). Autocorrelation specification in singular equation systems. Economic Letters, 46(4), 303–309.

Obayelu, A.E., Okoruwa, V.O., & Ajani, O.I.Y. (2009). Cross-sectional analysis of food demand in the North Central, Nigeria: The quadratic almost ideal demand system (QUAIDS) approach. China Agricultural Economic Review, 1(2), 173–193.

Schroeck, R. (2013). Quality and endogeneity issues in demand systems: a elasticities demand milk and Meat comparative estimation of price and expenditure elasticities of the demand for organic and conventional vegetables in Germany. German Journal of Agricultural Economics, 62(1), 18–38.

SFSO (2013). Consumer Price Index. Swiss Federal Statistical Office, Neuchâtel, Switzerland.

Shonkwiler, J.S., & Yen, S.T. (1999). Two-step estimation of a censored system of equations. American Journal of Agricultural Economics, 81(4), 972–982.

Tauchmann, H. (2005). Efficiency of two-step estimators for censored systems of equations: Shonkwiler and Yen reconsidered. Applied Economics, 37(4), 367–374.

Tefera, N., Demeke, M., & Rashid, S. (2012). Welfare impacts of rising food prices in rural Ethiopia: a quadratic almost ideal demand system approach. Selected paper prepared for presentation at the International Association of Agricultural Economists (IAAE), Foz de Igua-cu, Brazil.

Tekgü , H. (2012). Separability between own food production and consumption in Turkey. Review of Economics of the Household, 10(3), 423-439.

182

Endogeneity in censored demand systems

Thiele, S. (2010). Erhöhung der Mehrwertsteuer für Lebensmittel: Budget- und Wohlfahrtseffekte für Konsumenten. Jahrbücher für Nationalökonomie und Statistik, 230(1), 115–130.

Wood, B.D.K., Nelson, C.H., & Nogueira, L. (2012). Poverty effects of food price escalation: the importance of substitution effects in Mexican households. Food Policy, 37(1), 77–85.

Yen, S.T., Kan, K., & Su, S.J. (2002). Household demand for fats and oils: two-step estimation of a censored demand system. Applied Economics, 34(14), 1799–1806.

Zheng, Z., & Henneberry, S.R. (2010). An analysis of food grain consumption in urban Jiangsu province of China. Journal of Agricultural and Applied Economics, 42(2),337– 355.

Zheng, Z., & Henneberry, S.R. (2011). Household food demand by income category: Meat and milk demand elasticities demand milk and Meat evidence from household survey data in an urban Chinese province. Agribusiness, 27(1), 99–113.

183

Chapter V

Chapter IV Chapter Appendix

Table 5: Number of significant parameter estimates for factors and FE at different significance levels of the first step

(1) (2) (3) (4) <1% 000 007 025 057 <5% 000 000 000 045 <10% 000 000 000 021 >10% 000 005 005 303 Notes. See Table 2 for the definition of the models.

Figure 2: Number of significant parameter estimates for price and income parameters

and household characteristics at different significance levels in the first step Meat and milk demand elasticities demand milk and Meat 45 45 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0

(1) (2) (3) (4) (1) (2) (3) (4) <1% <5% <10% >10% <1% <5% <10% >10%

(A) Significance of price and income parameters. (B) Significance of household characteristics. Notes. See Table 2 for the definition of the models.

184

Endogeneity in censored demand systems

Table 6: Own- and cross-price elasticities for the base specification

(1) (2) (3) (4) (5) (6)

-1.335 -0.110 -0.021 0.100 0.035 0.186 (0.093)*** (0.047)** (0.030) (0.033)*** (0.073) (0.081)** -0.851 -1.222 0.464 0.697 -0.288 0.975 (0.289)*** (0.113)*** (0.112)*** (0.077)*** (0.233) (0.240)*** 0.248 0.149 -1.442 0.153 0.473 -0.258 (0.045)*** (0.033)*** (0.048)*** (0.022)*** (0.063)*** (0.042)*** 1.083 1.131 0.804 -0.954 0.697 1.837 (0.301)*** (0.124)*** (0.129)*** (0.088)*** (0.275)** (0.474)*** -0.057 -0.107 0.155 0.056 -1.186 0.187 (0.066) (0.037)*** (0.034)*** (0.027)** (0.104)*** (0.077)** 0.183 0.180 -0.272 -0.049 0.332 -1.189 (0.079)** (0.049)*** (0.030)*** (0.034) (0.418) (0.103)*** Notes. We present elasticities for rice (1), bread (2), pasta (3), sandwiches and other prepared products (4), wheat flour (5), and other kinds of flour (6).

Meat and milk demand elasticities demand milk and Meat

Table 7: Own- and cross-price elasticities for the factor specification

(1) (2) (3) (4) (5) (6)

-1.267 0.200 0.135 -0.123 0.875 -0.052 (1) (0.273)** (0.076)** (0.098) (0.049)* (0.250)** (0.256) 0.243 -0.442 0.659 0.178 0.766 0.307 (2) (0.369) (0.147)** (0.151)** (0.100) (0.432) (0.318) 0.091 0.028 -1.137 0.218 0.174 -0.158 (3) (0.096) (0.066) (0.053)** (0.045)** (0.217) (0.089) 0.567 0.299 0.675 -0.294 -0.934 0.709 (4) (0.374) (0.159) (0.164)** (0.110)** (0.473)* (0.314)* -0.030 0.201 0.533 -0.220 0.094 -0.140 (5) (0.394) (0.097)* (0.142)** (0.062)** (0.350) (0.363) 0.008 0.133 -0.043 -0.017 0.383 -1.118 (6) (0.135) (0.040)** (0.048) (0.027) (0.129)** (0.108)** Notes. See Table 6 for abbreviations.

185

Chapter V

Chapter IV Chapter Table 8: Own- and cross-price elasticities for the year-FE specification

(1) (2) (3) (4) (5) (6)

-1.016 0.062 0.117 0.028 -0.743 0.309 (1) (0.061)** (0.023)** (0.050)* (0.016) (0.202)** (0.053)** 0.190 -0.758 0.058 0.498 0.401 0.396 (2) (0.124) (0.081)** (0.115) (0.047)** (0.154)** (0.132)** 0.082 0.122 -0.965 0.175 0.269 -0.151 (3) (0.067) (0.035)** (0.070)** (0.026)** (0.082)** (0.078) 0.152 0.390 0.922 -0.364 0.646 0.157 (4) (0.128) (0.071)** (0.184)** (0.048)** (0.161)** (0.175) 0.272 0.133 0.144 -0.175 -1.467 -0.146 (5) (0.088)** (0.047)** (0.089) (0.039)** (0.106)** (0.096) 0.176 0.122 -0.167 0.008 -0.081 -1.001 (6) (0.044)** (0.017)** (0.041)** (0.013) (0.047) (0.055)** Notes. See Table 6 for abbreviations.

Table 9: Own- and cross-price elasticities for the month-FE specification elasticities demand milk and Meat

(1) (2) (3) (4) (5) (6)

-1.500 0.269 -0.731 -0.013 1.379 -0.160 (1) (0.379)*** (0.133)** (0.151)*** (0.080) (0.426)*** (0.360) -0.598 -0.498 -0.760 0.430 1.786 0.251 (2) (0.632) (0.283)* (0.221)*** (0.149)*** (0.846)** (0.516) 0.438 -0.335 -1.773 0.447 -0.494 0.637 (3) (0.619) (0.177)* (0.225)*** (0.107)*** (0.585) (0.611) 1.651 -0.042 3.163 -0.631 -3.497 1.047 (4) (1.254) (0.524) (0.475)*** (0.270)** (1.643)** (0.964)

-0.228 0.249 -0.095 -0.117 0.180 -0.625 (5) (0.515) (0.163) (0.193) (0.089) (0.568) (0.447) -0.204 0.369 -0.597 -0.077 1.266 -1.235 (6) (0.462) (0.137)*** (0.150)*** (0.087) (0.464)*** (0.419)*** Notes. See Table 6 for abbreviations.

186

Endogeneity in censored demand systems

Meat and milk demand elasticities demand milk and Meat

187

Chapter VI

Chapter IV Chapter Chapter VI

General findings and conclusions

Meat and milk demand elasticities demand milk and Meat

188

Conclusions

1 General findings and answers to the research questions

This dissertation is part of a research project funded by the Swiss Federal Office for Agriculture. The goal was to estimate final demand elasticities for different food and beverage product groups to improve the integration of Switzerland into the Common Agricultural Policy Regionalised Impact (CAPRI) model, which is an EU-wide partial equilibrium model consisting of a supply-and-demand module. This model allows the estimation of, for example, the effects of a free trade agreement between Switzerland and the European Union with respect to agricultural products or other policy scenarios.

The results of the various demand analyses in this dissertation and our findings and conclusions, which are presented in Chapters 1 to 5, allow us to draw some general conclusions and provide brief answers to the research questions in the form of a summary of the findings in those chapters.

1.1 The role of economic and other determinants

Our results, especially for the coefficients of determination and those of the statistical test for the parameter estimations, indicate that price and income are still major determinants of food and beverage demand in Switzerland. Nevertheless, demand is influenced by a complex range of variables. Popkowski et al. (1998), for example, determined that several aspects of consumers‘ choice processes, such as different tastes, preferences, and habits, contribute to heterogeneity in consumer demand, which they termed ―noise.‖ We addressed this issue by introducing household characteristics, a month factor, and a year factor in our meat and milk and alcohol demand analyses (Chapters 3 and 4). We further developed a new approach by replacing the two factor variables with a dummy variable for each month—the month FE model presented in Chapter 5—allowing us to filter the noise in the data more effectively. The significance of the factor variables, the dummy variables, and the household characteristics shows that variables other than prices and income exert a

189

Chapter VI

Chapter IV Chapter clear influence.29 This has weighty implications for, for example, agricultural policy measures based on market models such as computable general equilibrium (CGE) or partial equilibrium (PGE) models, which are used widely in policy analysis. These models allow the estimation of different ex-ante policy scenarios for supply and demand, but to simplify the models, they are normally driven only by economic determinants and neglect other determinants, such as household characteristics, on the demand side. On the basis of our findings, we can assume that the inclusion of other economic determinants may increase the accuracy of the model outputs significantly and consequently the accuracy of the impact estimation of policy measures.

1.2 Increased price elasticity

Demand analyses for food and beverages in Switzerland are scarce. This study

therefore fills a relevant research gap by providing current price elasticity estimates of Meat and milk demand elasticities demand milk and Meat meat and milk products (Chapter 3), beverages (Chapter 4), and bread and cereal products (Chapter 5). Due to the scarcity of literature with elasticity estimations, a complete comparison of elasticities over different time periods was not possible. Nevertheless, we were able to compare our findings regarding meat and milk products with those of Abdulai (2002), who used data from 1998. As mentioned in Chapter 3, our estimations, especially those of own-price elasticities for meat and milk products, were generally lower (more elastic, negative own-price elasticities decrease) than the estimations of Abdulai (2002), suggesting that consumers have become more price

elastic in recent years. While our analysis employed a more refined version of the QUAIDS than Abdulai (2002) and the differences in estimates could be the result of

29 For bread and cereal products, we compared the FE model with the other approaches of month and year dummy variables. We found that the difference in the elasticity estimates was not too big. Nevertheless, we would recommend using the FE model for further research. 190

Conclusions

different methods,30 there are also good reasons to expect that consumer demand has become more elastic over time.

Over the last 20 years, price transparency along the food supply chain has gained importance not only in Switzerland, but throughout Europe (e.g., EU Commission, 2009). Various studies have called for greater price transparency in Switzerland, especially at the wholesale level (see, e.g. Aepli & Joerin, 2011) The increased transparency at the retail market during the last years is mainly due to the market entry of the two discounters Aldi and Lidl in 2005 and 2009, respectively, and developments on the internet and on the mobile market, for example, in the form of so- called price-finding apps (Credit Suisse, 2014). This has probably led to greater price awareness in Switzerland – at least for some consumers. Regmi et al. (2001) compared several countries and showed that own-price elasticity for food becomes less elastic with higher income levels (i.e., negative own-price elasticities increase and move closer to zero). Our research raises the question of whether price elasticity does indeed follow a linear trend with higher incomes or rather charts a parabolic trend, increasing with higher income until a certain level of prosperity and then decreasing, in the highest income countries, such as Switzerland, due to consumers‘ better price awareness.

To the best of our knowledge, there is no literature on this issue, and we therefore recommend further theoretical and empirical research. Our assumption, shown in Figure 7, must be validated through more research focused on elasticity estimations for ultra-high income countries during different time periods. This will determine if other countries behave like Switzerland with respect to the development of elasticities, particularly as a high-income country becomes an ultra-high income country. Our assumption regarding price elasticity development for high/ultra-high income countries

30 Abdulai (2002) also used the QUAIDS but did not control for either censoring or the different qualities of products bought by the households in his study. We believe that the deviation in elasticities in comparison to our findings was only partially caused by the application of different statistical procedures. Rather, changes in consumer behavior, for which there may be several reasons, account for the bulk of the deviation. 191

Chapter VI

Chapter IV Chapter would not contradict the existing theory on consumer demand (e.g., Engel‘s law) but

instead supplement it.

Figure 7: Own-price elasticity among different income classes with respect to countries

Switzerland 1998 Switzerland

2004-2009

price elasticityprice

- own

Middle ic Ultra-high ic Increasing

Low ic High ic Meat and milk demand elasticities demand milk and Meat

Income Increasing Economic welfare Price transparency Price awareness ic = income countries

Source: Partly based on Regmi et al. (2001)

1.3 Heterogeneity of consumer behavior

Another issue that was discussed in Chapter 4 was the question of constant elasticity among different household types. As this dissertation aims to generate the basis for the implementation of the market model in the CAPRI, the assumption of constant elasticity is justified as partial equilibrium models do not normally consider different elasticities among different household segments. Nevertheless, a more differentiated view with respect to consumer segments could be useful. This is especially the case in the context of a tax, for example, on alcohol, fat, or sugar (see Chapter 4). To show the importance of a differentiated contemplation, we calculated the elasticities for three

192

Conclusions

household segments for (alcoholic) beverages. We were able to show that constant elasticity over different households does not hold for Swiss households, and that heavy drinking households are less price-sensitive, especially with respect to wine and beer, than moderate or light drinking households. Therefore, increasing the tax for alcoholic beverages would not lead to a significant steering effect; rather, it would only bring about a fiscal effect. This example shows clearly how important a differentiated perspective is. In addition to the perspective of the precision of policy measures, a differentiated estimation of elasticities, for example, among different household types, also has a number of major implications for general marketing because consumer heterogeneity has become an important issue in marketing and is the basis for not only market segmentation, but also targeting and positioning (Kamakura et al., 1996), allowing for more purposeful marketing measures.

2 Answers to the research questions

In Chapter 1, we defined the research questions, which were subsequently answered in Chapters 3 to 5. We present here first a short summary of the literature study on the methodology and the further development of existing methods (Chapters 1 and 2) and afterwards also of each research question.

In Chapter 1, we presented a model analysis as the basis for our decision to use the QUAIDS model, which was applied in Chapters 3 to 5. Several demand models have been used in the literature of which the most popular are the AIDS, LA/AIDS, and QUAIDS. For many years, the latter has been used rarely due to its non-linearity in parameters and the resulting computational effort required. We decided to use the QUAIDS because of the advantages it offers with respect to non-linear Engel curves. Our estimations showed that the assumption of linear Engel curves made by the AIDS, LA/AIDS, or other models, such as the Rotterdam model, is often not fulfilled in the case of Switzerland, depending on the product group. Therefore, our empirical work supported the decision with respect to the model, which we made on the basis of the

193

Chapter VI

Chapter IV Chapter theoretical properties and other empirical work regarding Switzerland and other developed countries. The main challenges were the occurrence of zero consumption, especially at the disaggregated level (Chapters 3 to 5), missing price information (Chapter 2), and endogeneity caused by, for example, omitted variables (noise in the data) (Chapter 5). We adopted and combined existing procedures to correct the QUAIDS to obtain unbiased estimates. However, these econometric procedures come at a cost: the adding-up restriction does not hold anymore. There was thus a methodological trade-off between the correct economic assumptions and the adequate, i.e., unbiased, econometric measurement. Given that our data had a high share of zero-values, we decided to employ the zero-corrected QUADS to analyze the Swiss data. Although the adding-up restriction was not implemented in the model, a test based on the estimates for meat and milk products showed that it was often fulfilled at least for those product groups.

The proposed procedures in Chapters 2 to 5 were not specific to the QUAIDS, but to a elasticities demand milk and Meat certain extent could also be applied in other demand models, or even in the broad field of econometric regression techniques. Furthermore, we adapted a recently published technique to generate market prices according to unit values (Chapter 2), which is useful not only for demand models, but also for trade models, etc., where the price variety among regions is of interest or useful in the models for obtaining a higher price variance.

Based on the literature review and the methodological developments we estimated the

models and answered the research question.

Research question 1: To what extent do consumers respond to price and income changes with respect to milk, meat, bread, and cereal products?

194

Conclusions

The reactions of consumers to price and income changes differ depending on the product group. From the perspectives of demand theory and previous empirical work, most of the elasticities are in the range of what we expect.

Most of the milk products are substitutes for one another. This finding is in line with that of Jaquet et al. (2000), with some exceptions, such as eggs, which are complementary goods to whole milk and milk drink/skim milk. This elasticity needs to be carefully interpreted because for this product group, we use the consumer price index with a relatively low variance in prices. For whole milk and curd, yogurt, and cheese, the compensated own-price elasticity is lower than that in Jaquet et al. (2000). This change indicates that consumers have become price sensitive in recent years, a result that accords with that of a recently published study in Switzerland in which Swiss consumers were found to have become price sensitive, especially to fluid milk prices (Schwarzenbach et al., 2013). We have discussed this issue in Section 1.2.

For meat products, quite similar results are obtained. Most of the product groups are substitutes, a finding that confirms those of other studies in Switzerland (see, e.g., Bernegger & Strasser, 1986 or Jaquet et al., 2000). The own-price elasticities are much lower than those estimated in Jaquet et al. (2000). This observation holds especially true for sausages, sausage products, pork, veal, and poultry, and it confirms the assumption that price sensitivity has increased in recent years; this trend is also evident, for example, in the increasing shopping tourism of Swiss residents abroad (Anwander Phan-Huy, 2006).

For bread and cereal products, no comparable estimates for Switzerland or for neighboring countries can be found at this level of disaggregation. With respect to the time fixed-effect model, the income elasticities are all positive and significantly different from zero. Our findings are mostly in line with those for aggregated analysis for Germany, in which the income elasticities for bread and cereal products range from 0.5 to 1 (Grings, 1993; Hoffmann, 2003; Thiele, 2008; Schröck, 2012). Our estimated

195

Chapter VI

Chapter IV Chapter own-price elasticities are all negative and range from -0.5 to -1.7, with the exception of

that for wheat flour.

The values of the elasticities have been provided in Chapter 4. All income elasticities are positive, whereas the own-price elasticities are mostly negative. These results are fully in line with those of previous literature (see Gallet, 2007, for a meta-analysis). The magnitude of the cross-price elasticities indicates that alcoholic and non-alcoholic

beverages are partly substitutes and partly complementary goods. With respect to the Meat and milk demand elasticities demand milk and Meat differences among the three household types, we have observed higher own-price elasticities, especially for beer and wine for low or moderate drinkers compared with heavy drinkers. Our results are in line with those of previous literature. Ayyagari et al. (2013), for example, found that heavy drinkers in the US are almost unresponsive to prices, whereas light or moderate drinkers are responsive to prices. This finding is also supported by the meta-analysis of Nelson (2013).

In contrast to that of beer and wine, the elasticity of spirits is more or less constant across the three segments. This result is slightly contradictory to that of Heeb et al.

(2003) who found that low drinkers are more elastic than heavy drinkers with respect to spirits. The possible explanations are that the analysis of Heeb et al. (2003) is based on another time period and that they only focus on one event in the past, whereas we use a large data set over a time period of six years. Furthermore, Heeb et al. (2003) have excluded other product groups (alcoholic or non-alcoholic beverages) in their model, whereas we estimate a full demand system for various beverages; in our model, parameters are simultaneously estimated, and the demand for a certain 196

Conclusions

product category is allowed to be influenced also by the price of other product categories. This result is a clear advantage over those of previous literature, in which single equations are often used to estimate elasticities (Nelson, 2013). For the further advantages of demand models compared with single equations models, please refer to Chapter 1.

Our research shows that the effectiveness of a high alcohol tax for Switzerland is questionable, and this issue is currently being discussed at a political level. Our results show that this policy will have a rather small effect on heavy drinkers, but it will strongly influence the behavior of low and moderate drinkers. A high tax can therefore fail to reduce the negative externalities related to alcohol consumption, but it leads to a loss in consumer surplus.

Our estimations are true, at least for the range of price variation in the data. If the tax is set massively higher than the current level, consumer elasticities will maybe change, and these elasticities will probably differ from those in our estimations. Furthermore, Section 1.1 mentioned that the demand for (alcoholic) beverages is influenced by determinants other than prices and income. This concern needs to be analyzed in further studies.

3 Contributions and limitations

3.1 Data material

The Swiss household expenditure survey is widely used for several purposes, such as weighting product groups in the Consumer Price Index or providing information on the demographic composition of the Swiss population. Furthermore, the data from the survey are regularly used for research. However, as it must fulfill different requirements, the survey has limitations for food and beverage demand analysis (for a description of the data set, we refer to Chapter 1). The limitations are as follows:

197

Chapter VI

Chapter IV Chapter  Missing price information: The household survey data on expenditure does provide data on expenditure on the product groups, household characteristics,

and in particular, the quantities of food and beverages bought. However, prices are not collected, which is not unusual for household expenditure surveys (see, e.g., Thiele, 2008, for the same limitations of the German household expenditure survey). To overcome the missing price information, we developed a model to calculate quality adjusted market prices for every household, presented in Chapter 2.  Cross-sectional data: The survey has a periodicity of one month, and every household is only recorded once. Therefore, it is not possible to conduct an analysis of the behavior of the households over time (panel data analysis). The provided elasticities could only reflect the short-term reactions of the households with respect to price and income changes. In contrast to a cross-

sectional data set, a panel data set would allow the estimation of adjustment Meat and milk demand elasticities demand milk and Meat processes through dynamic demand systems. However, those studies are very rare because panel data over a longer time period are scarce.  Time delay: Due to aggregation and other treatment of the data, the single- household data of the survey could only be obtained with a time delay. Therefore, we used the data from 2000 to 2009. More recent data were not available at the time when the project was initiated. We assume that this has not significantly affected the estimates. Nevertheless, we would recommend a repetition of the estimations in a few years with newer data to check whether

consumer behavior has changed.  Aggregation level: To minimize the reporting burden on households, the disaggregation of the product groups is partially limited (see Chapter 1). For the purposes of policy decision-making, which was the main focus of our analysis, the data on the most disaggregated product groups provided by the household expenditure survey were sufficient. For further demand analysis, for example, as the basis for marketing decision-making, data on single products would be

198

Conclusions

more appropriate. For instance, data on expenditure on meat does not provide information regarding different pieces of meat, such as fillet or minced meat, within the beef product group. Further disaggregation would allow the estimation of the substitution effects within a product group or across different product groups, such as substitution elasticity between fillet of beef and fillet of pork.

Despite these relatively minor limitations, the data of the Swiss household expenditure survey is an appropriate data set for estimating price and income responses because of the precision with which the data are collected and subsequently prepared, and the diverse variables that are recorded. The only alternative for Switzerland is the household panel of Nielsen AC, which has some advantages with respect to aggregation levels and price information. However, those data are normally very expensive and are only available for a relatively short time period (two to three years). Using appropriate methods, for example, by generating missing prices or applying a censored QUAIDS, we were able to overcome the survey data limitations for the most part.

3.2 Method and estimations

The proposed model discussed in Chapters 3 to 5 combines different state-of-the-art methods to take into account: (a) the censoring of the dependent variable due to zero consumption, especially at the most disaggregated level, (b) the possible endogeneity of the expenditure variable in the model, (c) biased unit values due to quality effects, which were corrected by a new method to adjust the unit values and retrieve market prices, and (d) noise in the data. We used a combination of recent and widespread methods to estimate the QUAIDS and overcome these challenges. The combination of different methods obviously led to an improvement in the estimations. However, the extent of the improvement with respect to reduced model forms is mainly dependent on the data material. For example, a high share of zero consumption would lead to a

199

Chapter VI

Chapter IV Chapter higher bias, while a low share of zero consumption would allow us to estimate the system without taking into account the censoring problem. Therefore, the advantage of our model in comparison with the reduced models is varying size depending on the product groups in the model.

Although we tried to address almost all the specificities of the data and to use the most suitable model to obtain accurate estimates, the interpretation of the elasticities has some limitations:

 From an econometric perspective, the elasticity estimations are only valid within the variance of prices and income. While the income variance is relatively high due to the cross-sectional structure of the data set, the price variance is limited. Especially in the case of a free trade agreement for agricultural products with the European Union or with other economic areas, we anticipate a sharp shift in

consumer prices. For a free trade agreement with the EU, we expect prices for Meat and milk demand elasticities demand milk and Meat food to decrease on average about 10% (Bösch et al., 2011), supposing even higher reductions, of up to 40% (SFOA, 2008) for some product groups with highly restrictive trade barriers, e.g., meat. Due to a lack of price variation in this extent, price elasticities after trade liberalization could differ slightly from our estimations.  In addition to the economic determinants of food and beverage consumption, household characteristics and factors other than price and income can influence consumption decisions. We tried to consider as many of those effects

as possible with the data material by implementing household characteristics

and a factor for month and for year (Chapter 3 and 4) or dummy variables for months (Chapter 5) into the model. Our results revealed that the coefficient of determination is relatively high for the models with aggregated product groups, and somewhat lower for the models concerning the most disaggregated product level. For instance, education levels, which play a major role in the consumption of, for example, unhealthy food products, were considered using a dummy

200

Conclusions

variable to determine whether or not a university degree had been obtained. This variable possibly only caught a partial effect of education. Furthermore, a detailed analysis would be needed to estimate the whole effect of education on food and beverage consumption, which was not within the scope of our studies.  By implementing a variable for month and a variable for year, we considered at least part of the seasonality and trends in food and beverage consumption. With the approach presented in Chapter 5, we were able to filter out almost all the noise in the data. However, we did not consider non-linear trends or cross- effects, for example, between seasonality and prices.31 We intentionally did not implement such variables because interpreting those effects is often quite unclear, and implementing these variables would have reduced the degree of freedom drastically and increased the computation time for the non-linear equation system.

4 Further areas of application and implications for agribusiness

In addition to the use of elasticities for policy analysis, for example, within partial equilibrium models such as the CAPRI or general equilibrium models such as the GTAP, our estimates also lead to an improved understanding of consumers. We are aware that the most disaggregated level in our analysis is still too highly aggregated for specific marketing purposes. However, our results provide an overview for a general understanding of consumers, particularly their reactions to price changes. We did only discuss the most important product groups for Swiss agribusiness in this thesis. In addition, we also calculated the elasticities for other product groups, such as various fruits and vegetables or fats and oils, which are presented in the final report of the project, entitled ―Food Demand Analysis in Switzerland,‖ for the Swiss Federal Office for Agriculture. Together with this report, the current thesis will serve as the

31 For example, it is possible – in the case of meat during barbecue season – that the seasonality effect depends on whether prices are high or low. 201

Chapter VI

Chapter IV Chapter basis for further in-depth studies, particularly those that provide the basis to develop

efficient and effective marketing measures.

Furthermore, the environment for agribusiness companies in Switzerland has become increasingly uncertain in recent decades. This issue is related to, for example, the high price volatility of the world markets and the limited market intervention of the government. This situation in turn leads to a high need for information to improve the predictability of the markets. Consumers are part of the market, so our results contribute to future planning with good forecasts and thus reduce uncertainty in the field of agribusiness.

5 Future research and outlook32

The estimation results and limitations discussed above indicate that further research

activities should focus on estimating dynamic demand systems. While static models elasticities demand milk and Meat assume that consumers react immediately to price and income changes, dynamic models consider delayed and incomplete reactions. Inertia and habits play an important role. Therefore, Ray (1984) developed a dynamic generalization of the AIDS (e.g., applied by Liao & Chern, 2007). The challenge of the dynamic AIDS or the dynamic QUAIDS is the lagged data, which are missing in the Swiss household expenditure survey. With repeated cross-sectional data, a direct estimation is not possible. However, although controversial, there are methods for creating a pseudo- panel based on cross-sectional data (e.g., Varbeck, 2008). Nevertheless, we did a preliminary (unpublished) study, based on estimations of the Mahalanobis distance, to check whether a pseudo-panel creation would be possible. The method is not mature enough because a large number of households had to be assigned to only one or a few households. Despite these challenges, the idea of a dynamic model should still be pursued.

32 Some lines of this chapter are based on a report that was submitted to the Swiss Federal Office for Agriculture in June 2012 (Aepli, 2012). The report is unpublished as of now (May 2014). 202

Conclusions

As the data set includes additional information for every household, one might consider estimating income elasticities for different household types to show whether elasticities vary across household types for, for example, alcohol consumption (Chapter 4). In addition to the analysis of beverages, a further analysis of different household segments by income class would enable the estimation of whether the assumption of a constant elasticity function has been violated. Instead of estimating a separate model for every household segment, an adjustment of the QUAIDS specification allowing for interaction effects between household segments and the parameters for expenditure and the quadratic expenditure term would permit the estimation of elasticities for different household segments in one model. This would be particular interesting with respect to computable general equilibrium models because it allows, for example, to estimate the effect on food security of different household segments with respect to changing prices or pricing policies in, for example, developing countries. Nevertheless, from an econometric perspective, there are some limitations. An estimation of own- and cross-price parameters with allowances for differences between household segments would lead to a strong loss of degrees of freedom. Therefore, the estimation of cross-effects is only realistic for income elasticity.

From a methodological perspective, there is a need for further research on how to implement the adding-up restriction while correcting for censoring. As mentioned above, estimations on a more disaggregated level would be interesting, but this implies that the share of zero consumption will increase strongly. Neglecting the censoring of the dependent variable in the QUAIDS would therefore lead to significant distortions. We first tried to develop a new approach to apply the adding-up condition while implementing the consistent two-step estimation procedure used by Shonkwiler and Yen (1999). It is due to difficulties in computation that this thesis did not report the results derived from this approach. Tiffin and Arnoult (2010) implemented the adding- up restriction while controlling for the censoring of the data by using Bayes-based simulations, and Fourmouzi et al. (2012) estimated a censored AIDS by using a 203

Chapter VI

Chapter IV Chapter likelihood approach. Both approaches are accepted in the literature, but they are difficult to implement – especially the likelihood approach – because local optima in the likelihood surface can occur and cause subsequent problems in the estimation. These two procedures have hardly been used in recent years, presumably due to the high complexity of implementation. Once the computational issues have been solved, the methodological approach that we suggest should lead to broader application in research by reducing the complexity of the method. We are therefore currently working on a follow-up study that will successfully implement this approach for practical application.

Meat and milk demand elasticities demand milk and Meat

204

Conclusions

References

Abdulai, A. (2002). Household demand for food in Switzerland: A Quadratic Almost Ideal Demand System. Swiss Journal of Economics and Statistics, 138(1), 1-18.

Anwander Phan-Huy, S. (2006). Das nahe Ausland als Konkurrent im Schweizer Detailhandel. Die Volkswirtschaft, 6, 24-27.

Ayyagari, P., Deb, P., Fletcher, J., Gallo, W., & Sindelar, J.L. (2013). Understanding Heterogeneity in Price Elasticities in the Demand for Alcohol for Older Individuals. Health Economics, 22(1), 89-105.

Bernegger, U., & Strasser, M. (1986). Analyse der Fleischnachfrage in der Schweiz. Scientific report, ETH Zurich, Zurich.

Bösch, I., Weber, M., Aepli, M., & Werner, M. (2011). Folgen unterschiedlicher Öffnungsszenarien für die Schweizer Nahrungsmittelindustrie. Scientific report, ETH Zurich, Zurich.

Credit Suisse (2014). Retail Outlook 2014. Credit Suisse Economic Research, Zurich.

EU Comission (2009). Improving price transparency along the food supply chain for consumer and policy makers. Commission of the European communities, Brussels.

Forumouzi, V., Genius, M., & Midmore, P. (2012). The demand for organic and conventional produce in London, UK: A system approach. Journal of Agricultural Economics, 63(3), 677-693.

Gallet, C.A. (2007). The Demand for Alcohol: A Meta-analysis of elasticities. Australian Journal of Agricultural and Resource Economics, 51(2), 121-135.

Grings, M. (1993) Die Nachfrage nach Nahrungsmitteln in Ost- und Westdeutschland. Vergleichende Analyse auf Grundlage eines ökonometrischen Modells. Habilitation, Georg-August-Universität, Göttingen.

Hoffmann, C. (2003) Die Nachfrage nach Nahrungs- und Genussmitteln privater Haushalte vor dem Hintergrund zukünftiger Rahmenbedingungen. Studien zur Haushaltsökonomie, Peter Lang, Frankfurt am Main.

205

Chapter VI

Chapter IV Chapter Jaquet, P., Abdulai, A., & Rieder, P. (2000). Empirische Analyse des Nahrungsmittelverbrauchs in der Schweiz: Ein dreistufiges LA/AIDS Modell. ETH

Zurich, Zurich.

Kamakura, W.A., Kim. B-D., Lee, J. (1996). Modeling Preference and Structural Heterogeneity in Consumer Choice. Marketing Science, 15(2), 152-172.

Liao, H., & Chern, W. S. (2007). A dynamic analysis of food demand patterns in China. Selected Paper at the AAEA Annual Meeting in Portland.

Nelson, J.P. (2013) Does heavy drinking by adults respond to higher alcohol prices and taxes? A survey and assessment. Economic Analysis and Policy, forthcoming.

Popkowski Leszczyc, P.T.L., & Bass, F.M. (1998). Determining the effects of observed and unobserved heterogeneity on consumer brand choice. Applied Stochastic Models and Data Analysis, 14(2), 95-115.

Ray, R., (1984). A dynamic generalisation of the Almost Ideal Demand System. Economic Letters, 14(2/3), 235-239.

Regmi, A., Deepak, M.S., Seale, J.L., & Bernstein, J. (2001). Cross-country analysis of elasticities demand milk and Meat food consumption patterns. In: Changing Structure of Global Food Consumption and Trade, Economic Research Service, USDA, Washington.

Schröck, R. (2012) Wie preissensibel reagieren deutsche Verbraucher? Aktuelle Schätzungen von Preis- und Ausgabenelastizitäten auf der Basis von Haushaltspaneldaten. Presentation at the GEWISOLA annual meeting, 26-28 September 2012.

Schwarzenbach, R., Blättler, T., Durgiai, B., & Haller, T. (2013). Marktöffnung ―Weisse Linie‖. Scientific report, Bern University of applied Sciences, Bern.

SFOA (2008). Verhandlungen Schweiz-EU für ein Abkommen im Bereich der öffentlichen Gesundheit (GesA). Swiss Federal Office for Agriculture, Bern.

Shonkwiler, J.S., & Yen, S.T. (1999). Two-step estimation of a censored system of equations. American Journal of Agricultural Economics, 81(4), 972–982.

Tiffin, R., & Anroult, M. (2010). The demand for a healthy diet: estimating the almost ideal demand system with infrequency purchase. European Review of Agricultural Economics, 37(4), 501-521.

206

Conclusions

Varbeck, M. (2008). Pseudo panels and repeated cross-sections. in The econometrics of panel data: Fundamentals and recent developments in theory and practice, Mátyás L. and P. Sevestre (ed.), Springer-Verlag, Berlin and Heidelberg.

207

Chapter VI

Chapter IV Chapter

Meat and milk demand elasticities demand milk and Meat

208

Acknowledgements

Acknowlegdements

I would like to thank the many people who have supported me during my doctoral studies.

First of all, I especially thank examiner Prof. Dr. Michael Siegrist for the opportunity to write this thesis and for his helpful support and assistance throughout this time.

Furthermore, I would like to express my gratitude to Dr. Michael Weber for his support with respect to the organization of my dissertation, the project of the Swiss Federal Office for Agriculture, and for agreeing to be my co-examiner; he has also done much to see that this dissertation could be completed.

Special thanks also to my second co-examiner and co-author of my second paper Dr. Christian Kuhlgatz from the Thünen-Institute in Braunschweig. His helpful advice and scientific support were absolutely essential to the success of this dissertation.

Other thanks go to Prof. Dr. Robert Finger from the University of Bonn and to Sandro Steinbach from the Agricultural Economics Group at ETH for being co-authors of my papers. Furthermore, I would like to thank Prof. Dr. Thomas Heckelei and Dr. Wolfgang Britz from the University of Bonn for their input, especially at the beginning of the dissertation project.

I would also like to thank to Dr. Peter Stalder for his support during the first part of the thesis.

Many thanks to the Swiss Federal Office for Agriculture for financially supporting this dissertation and to my colleagues in the Agricultural Economics group, especially to Dr. Michel Dumondel and Dr. Robert Jörin. Both have made important contributions and helped make this dissertation possible. I also offer my sincere gratitude to my student assistants: Benjamin Rohrer, Jonathan Dürr, Annina Christoffel, Andreas Zweifel and Jonas Anderegg.

209

Chapter VI

Chapter IV Chapter I would also like to express my deepest gratitude to my family and friends for

supporting me during this stimulating but often stressful time.

Meat and milk demand elasticities demand milk and Meat

210

Acknowledgements

211

Curriculum vitae

Chapter IV Chapter Curriculum vitae

PERSONAL INFORMATION

Aepli Matteo Federico

Born on May 9, 1988 Citizen of Lucerne and Niederhelfenschwil Nationality Swiss

WORK EXPERIENCE

Current, May 2013 Lecturer in statistics, institute of higher vocational education Strickhof

May 2012 – Jun 2013 Project leader regional project ‗Mehrwert Holz‘, Pro Holz Schwyz. Project to further strengthen the wood value chain. Responsible for marketing/networking and communication.

Current, Mar 2011 Research fellow, Institute for Environmental Decisions (IED), ETH Zurich / Agricultural Economics Group (former head: Prof. Dr. Bernard Lehmann), since Sep 2011: PhD-student

Mar 2008 – Feb 2011 Scientific-support assistant to Mr. Dr. Robert Jörin (Agricultural Trade), Institute for Environmental Decisions IED ETH Zurich / Agri-Food & Agri-Environmental Economics Group (Prof. Dr. Bernard Lehmann)

UNIVERSITY STUDIES AND FURTHER TRAINING

Feb 2011 – May 2013 DAS in applied statistics (advanced studies diploma), Department of Mathematics, Seminar for Statistics, ETH Zurich

Feb 2010 – Feb 2011 MSc in Agricultural Science, Department of Agricultural and Food Science (D-Agrl), ETH Zurich, Major in Food and Environmental Economics, Minor in Ruminant Science (attending several courses in other departments e.g. in supply chain management or marketing)

Oct 2006 – Jan 2010 BSc in Agricultural Science, Department of Agricultural and Food Science (D-Agrl), ETH Zurich, focus on agricultural economics

212

Curriculum vitae

AWARDS

Dr.-Berthold-Pohl-Stipendium Prize Winner 2011 (best academic and scientific work in agriculture and forestry within Germany, Austria, Italy and Switzerland)

Young researcher award 2011 3rd place, SSA Swiss society for Agriculture Economics and Rural Sociology, awarded scientific article ‗Intra-industrieller Handel und Wettbewerbsfähigkeit der Schweizer Nahrungsmittel- industrie‘

213