Determinants of beef demand in developing countries and the environmental Kuznets curve

Determinants of beef demand in developing countries and the

environmental Kuznets curve

University of Groningen Faculty of Economics and Business

Master Thesis International Economics and Business

Name Student: Prijs, M.A. Student ID number: s2175576

Student email: m.a.prijs@student.rug.nl Date Thesis: 19-06-2017

Name Supervisor: Jepma, C.J.

Abstract

I. Introduction

In Al Gore's well-known "An inconvenient truth", directed by Davis Guggenheim (2006), the problem of global warming is addressed. The mechanism behind man-induced climate change is quite simple: mankind needs energy to life comfortably and produce manufactures, the energy comes – among other things – from fossil fuels. Due to the usage of fossil fuels, the concentration of CO2 and other green house gasses1 in the atmosphere increases. This warms the earth, which induces climate change. One of the industries largely responsible for GHG emissions is the meat industry. In fact, the meat industry contributes more to climate change than, for example, traffic (Steinfeld, 2006; Goodland & Anhang, 2009). If you think about it this makes sense, animals need to grow, be fed, cared for, and transported. After the animal has been turned into meat fit for consumption, the meat has to be transported and kept cool. This process is very energy intensive. To make matters worse, ruminant meats like cattle release a lot of methane during their lifetime, a GHG more potent than CO2. Goodland and Anhang (2009) calculated that livestock produced GHG emissions equivalent to over 32,564 million tons of CO2 in 2004; over 51% of the world's total GHG emissions. Because of the meat industry's huge carbon footprint, there is an on-going discussion in the Netherlands about introducing a meat tax which would help the environment. The rationale is that if meat is more expensive, people will consume less meat and therefore the meat industry will produce less. This, in turn, will have a positive impact on the environment.

In this paper I am not concerned with the ethical question of whether a meat tax would be justified as a measure to combat climate change. Rather, the focus in this paper is on demand for beef in developing countries and the possible existence of an environmental Kuznets curve for beef demand. Climate change is a worldwide problem, which should be tackled on an international scale. If the Netherlands and other advanced economies start eating less meat – irrespective of whether this is caused by a tax on meat, awareness of the meat industry's carbon footprint, or any other reason – and developing countries like China and Brazil start demanding more meat as their income per capita increases, then limiting meat consumption in the Netherlands would be fighting a running battle – although arguably every little bit helps.

complementary research questions:

1. What are the determinants of beef demand in developing countries, and which of these determinants are most important?

2. Is there evidence of an environmental Kuznets' curve for beef consumption? 3. How does income inequality impact the demand for beef in developing countries?

To answer these research questions different multiple regression models are constructed, which are tested on a sample of 14 developing countries over the period 1983 until 2013.

In light of the two global trends, the importance of these research questions is quite clear. They also fill a research gap. To the best of my knowledge, no-one has ever investigated the existence of an environmental Kuznets curve for beef consumption. Nor has anyone looked at the relationship between income inequality and beef demand. Furthermore, although there are many studies investigating demand for meat in developed countries, there are next to none that focus specifically on developing countries.

In section 2 of this paper, the existing literature on the subject of determinants of meat demand and the environmental Kuznets curve will be reviewed. Furthermore, several testable hypothesis are developed in this section. Then, in section 3, empirical models to test these hypotheses are built and the methodology and data will be described. In section 4, the results of estimating the models are shown and briefly discussed. In the fifth section, the limitations of this study will be discussed, and lastly some conclusions will be drawn in section 6.

II. Literature review

As mentioned in the previous section, the research questions this paper addresses are: "What are the determinants of meat demand in developing countries, and which of these determinants are most important?", "Is there evidence of an environmental Kuznets' curve for meat consumption?", and “How does income inequality impact the demand for beef in developing countries?”. Before the hypotheses which correspond to these research questions are developed, the existing literature on determinants of beef demand and the environmental Kuznets curve will be reviewed. Then, the broader purpose of this paper will be explained and I will position this paper in the literature. First, I will go into the existing literature on determinants of meat demand.

buy beef. Their findings of specific beef quantity demand drivers are represented in table 1.

Whereas Schroeder et al. (2013) examine the determinants of the demand for beef in the United States, Oyewumi and Jooste (2006) focus on demand for pork in Bloemfontijn (South Africa). They use ten explanatory variables (demand determinants) – identified by previous studies as important determinants for meat demand – which may impact the consumption of pork in South Africa. These determinants include economic factors like income, the relative price of pork, the price of other meat types, and expenditure on meat, as well as non-economic factors such as race, gender, religion, quality, place of purchase, and value adding. Of these determinants, Oyewumi and Jooste (2006) found the household monthly income, the relative price of pork, the current household monthly expenditure on meat, preference for value-added pork products, the price of substitutes, and pork quality to be of significant importance for pork demand in South Africa. In the next section, when I will build a model to answer the research questions, Oyewumi and Jooste (2006) will be revisited. For now, it will serve to state that they found that "non-economic factors are becoming increasingly more important when consumers have to make purchasing decisions regarding pork."

According to Van Zyl et al (1992), who present a model within which price interactions and market price leadership can be analyzed [for meat], beef has been the dominant red meat in South Africa since at least 1980-1981. They state that "increased prices and decreased consumption of beef can be expected to result in increased demand, and hence higher prices, for substitute products such as mutton, pork, chicken, fish and eggs." Of course, the opposite would also be true if beef prices fall (Van Zyl et al., 1992). Hence, just like Tryfos and Tryfanopoulos (1973), Van Zyl et al. (1992) hypothesize the importance of meats as each others substitutes. There is, however, a difference between Van Zyl et al (1992) and Tryfos and Tryfanopoulos (1973). While both note the importance of substitution effects, Tryfos and Tryfanopoulos (1973) find that the substitution effects are not completely symmetric. In particular, they state that the consumption of beef, veal and (domestic) lamb is not affected by changes in the prices of other meats. On the other hand, the demand for pork is sensitive to changes in the prices of beef, veal and lamb. The authors illustrate this point with an example. According to them, "a rise in the price of pork may not lead to an increase in the consumption of beef, but only to a decrease in pork consumption; on the other hand, an increase in beef prices will result in an increase in pork consumption and a decrease in beef consumption." (Tryfos & Tryfanopoulos, 1973).

Liu and Deblitz (2007) specifically investigated the determinants of beef consumption in China. Beef is relatively new to China and is slowly gaining ground. Traditionally, pork was the preferred choice of meat in Chinese households. Because beef is slowly but steadily becoming more popular, Liu and Deblitz (2007) focus on factors affecting beef purchasing decisions in China. They did a survey and found that the main limitations for beef consumption concerned the relatively high price of beef and the unfamiliarity with its cooking method. What is particularly interesting about the paper by Liu & Deblitz (2007), is that they discuss the historical perception of beef in China. It turns out, cattle has long been protected by policy makers in China. In fact, slaughtering cattle has long been illegal and whoever did, or knew about it, would be punished by the government. Even today, farmers in rural China believe that if they slaughter cattle or eat beef, they will face a bad fate (Liu & Deblitz, 2007). This example shows the importance of cultural perception for beef demand.

So far, I have reviewed literature on both meat demand in general and beef demand. However, while it is informative to look at broader determinants of meat demand, the focus in this paper is on the determinants of consumer demand for beef. There are several reasons for this. For starters, Van Zyl et al (1992) state that beef is the most consumed meat in South Africa, and Tryfos and Tryfanopoulos (1973) find that – in Canada – demand for other meats responds to price changes of beef. This is further evidence of beef's dominance. In fact, we can reasonably expect that beef is one of the most popular meats in most countries. Moreover, as one of the ruminant meats, beef has the largest environmental (carbon) footprint (Edwards-Jones et al., 2009; Ratnasiri & Bandara, 2017). Therefore, from an environmental perspective beef is interesting as well, and understanding the determinants of beef demand in developing countries and their relative importance might have policy relevance. From the aforementioned papers we can safely infer the importance of several determinants of beef demand.

Hypothesis 1: Price is the most important determinant of beef demand

Hypothesis 2: The price of other meats has an effect on the quantity of beef demanded

In 1955, Simon Kuznets published a now famous paper in which he addresses the relationship between a county's economic development and the amount of income inequality in that country. He found an inverted U-shape relationship between inequality (on the vertical axis) and economic growth (on the horizontal axis). Such a curve implies that as the economy develops – indicated by an increase in Gross Domestic Product2 _{per capita – the level of inequality first increases, as the} economy moves from an agricultural to an industrial society, and then, as the transition to a service-based society is made, the level of inequality in the economy falls. In other words, according to Kuznets (1955) as GDP per capita increases, the level of inequality first increases and then decreases after GDP per capita has reached a certain 'tipping point'. The existence of the relationship proposed by Kuznets has been much debated. Recently, hugely influential books written in the wake of the Great Recession by Thomas Piketty (Capital in the twenty-first century, 2014) and Joseph Stiglitz (The great divide: Unequal societies and what we can do about them, 2015) have shown that – at least in the United States – inequality has only increased instead of decreased over the past couple decades.

In this paper I am not concerned with the existence of the Kuznets curve, but rather with the possible existence of an environmental Kuznets curve for beef demand. The environmental Kuznets curve hypothesis was first introduced by Grossman and Krueger (1991), who investigated the environmental impacts of a North American Free Trade Agreement (NAFTA). They found, among other things, that the concentration of the two pollutants sulfur dioxide and smoke increases with per capita GDP at low levels of national income, but decreases with GDP growth at high levels of income. The environmental Kuznets curve hypothesis was born. Stern (2004) explains the environmental Kuznets curve, henceforth EKC, quite well. The EKC is a hypothesized relationship between environmental degradation and income per capita. Much like with the 'original' Kuznets curve, the EKC relationship implies that the environmental impact indicator is an inverted U-shaped function of income per capita. To be more explicit, in the early stages of economic growth environmental degradation and pollution increase, but beyond some level of income per capita – the so called tipping point, which will vary for different indicators – the trend reverses. Hence, if the EKC hypothesis holds, at high levels of income economic growth leads to environmental improvement (Stern, 2004; Moosa, 2017).

Much like the Kuznets curve, the existence of the EKC is much debated and empirical results have been mixed. On the one hand, Apergis and Ozturk (2015) tested the EKC hypothesis in fourteen Asian countries and confirmed that there is indeed evidence of the EKC for the relationship between CO2 emissions and income in these Asian countries when controlling for population density, land, industry shares in GDP and institutions. Al Mulali et al. (2015), too, find evidence of the EKC for Latin American and Caribbean countries. Furthermore, Moosa (2017) also provides evidence of an inverted U-shaped relationship between income and CO2 emissions in Australia, although he clearly stresses that his results cannot be generalized.

in Cambodia, and Gill et al (2017) are convinced the empirical EKC literature is not econometrically sound and state that the relationship between income and many sources of environmental degradation has not been tested yet. A 'negative' result was also obtained by Stern (2004), who concludes that the statistical analysis on which the EKC is based is not robust and that there is hardly any evidence for an inverted U-shaped pathway that countries follow as their income rises. Moreover, Stern (2004) completely agrees with Copeland and Taylor (2004), who state that “... both the theoretical and empirical work on the EKC leads us to be sceptical of a simple and predictable relationship between pollution and per capita income.”

Still, a theoretical case can be made for the EKC hypothesis. The inverted U-shaped curve can be explained by the interaction between the so called time and scale effects (Moosa, 2017; Stern, 2004). The scale effect is caused by growth in the scale of the economy, which is expected to produce proportional growth in environmental degradation (Moosa, 2017). The time effect implies the decline of pollution over time in the absence of economic growth (Stern, 2014). Hence, if the scale effect dominates the time effect then the curve has a positive slope and if the time effects dominates the scale effect then the curve has a negative slope.

According to Moosa (2017), the theoretical rationale behind the EKC proposition – and hence the reason why the scale effect becomes weaker at higher levels of income per capita – is that higher levels of per capita income are associated with structural change in the economy. As the economy shifts from industrial towards information-intensive industries and services, the environmental awareness and enforcement of environmental regulation also increase. Furthermore, countries with high levels of GDP per captita often have access to better technology and can afford a higher level of environmental expenditure (Moosa, 2017). He puts forward four possible – and imperfect yet plausible – explanations as to why the impact of economic growth on the environment becomes less severe as income rises. Drawing from Beckerman (1992), Moosa (2017) entertains the thought that environmental quality can be considered a normal good, which would mean that as a person's income rises he or she will start demanding more environmental quality. A second possible explanation is that as the level of income rises, the economy will start using less pollution-intensive technology (Grossman & Krueger, 1995; Moosa, 2017). Thirdly, the share of pollution-intensive sectors in total output goes down as the share of service sectors goes up (which is usually associated with rising income), and finally, that the tendency op people to breed declines as they become richer, thus alleviating pressure on the environment (Moosa, 2017).

The third explanation provided by Moosa (2017) – the relative importance of pollution-intensive industries as opposed to service sectors – deserves a closer look. The second wave of globalization might be a possible source of the existence of the EKC in advanced economies, because as their income per capita rises, labour intensive firms had an incentive to relocate certain activities to low-wage countries. More oft than naught, these activities entailed tasks which use more pollution-intensive technology. This idea was already present in Stern (2004), citing Arrow et al (1995) and Stern et al (1996), who argued that the EKC type relationship might to some extent be the result of the effects of trade on the distribution of polluting industries. According to Hecksher-Ohlin trade theory, countries specialize in the production of goods that are intensive in the production factors they are endowed with in relative abundance. Part of the reduction in environmental degradation levels in developed countries and increases in environmental degradation in developing countries may reflect this specialization (Arrow et al, 1995; Stern et al, 1996; Stern, 2004).

hypothesis, no-one has ever investigated the inverted U-shape relationship between income and beef demand – to my knowledge. Yet, there are compelling reasons why such a relationship might exist. If we adopt Beckerman's (1992) view of environmental quality as a normal good – meaning that demand for environmental quality goes up as income rises – and we take into account that the production of beef has a massive environmental footprint, then we might infer that demand for beef goes down as income increases. Granted, this causal relationship might only exist if the public is aware of beef's carbon footprint. As we know from Ratnasiri and Bandara (2017), Australian demand for ruminant meat (like beef) is decreasing, whereas the demand for non-ruminant meat (i.e. meat with a relatively smaller (negative) impact on the environment has been increasing). One could infer from this that there is a growing awareness about the negative impact of beef consumption on the environment among the general public. Another telling reason why one might expect awareness to be on the rise, is that several countries – like Denmark and Sweden – have introduced an environmental tax on meat and dairy products (Edjabou & Smed, 2013; Jensen et al, 2015; Sall & Gren, 2015) whilst other countries are debating implementing such a measure. Furthermore, the expectation that consumers care about the carbon footprint of consumers goods does not only come from extending the conceptual model of Beckerman (1992). Shewmake et al (2015) propose a model to systematically estimate how consumers will respond to carbon footprint labels on consumer goods. One of their several findings was that carbon labels on alcohol and meat would achieve a large decrease in carbon emissions. Hence, we know from Beckerman (1992) that consumers care more about environmental quality as their income rises, and we know from Shewmake et al (2015) that if consumers are aware of meat's massive environmental footprint they are likely to demand less meat. Moreover, according to Ratnasiri and Bandara (2017) the Australian per capita consumption of ruminant meat (for example beef and lamb) has declined over the last two decades, whilst the non-ruminant meat consumption (for example chicken, pork and kangeroo) has continued to increase. Actually, Ratnasiri and Bandara (2017) already touched upon the relationship between the EKC and meat consumption, stating that the per capita GHG emissions [in Australia] are likely to decrease due to the increased inclusion of non-ruminant meat in Australian diets at the expense of ruminant meat (i.e. beef). Thus, as the consumer becomes more aware of the carbon footprint of ruminant meat, demand is likely to increasingly shift to non-ruminant meat. Since we are already observing this in an advanced economy like Australia, one might expect awareness to (soon) be on the rise on a global scale.

Hypothesis 4: There exists an inverted U-shape relationship between income per capita and demand for beef, with demand for beef first increasing as income increases and later decreasing at high levels of income per capita.

per capita (H4). However, if most of the wealth is concentrated at the top of the income distribution, then income per capita might not matter much for beef consumption. If only the bourgeoisie is rich (and increasingly getting richer), then a country's per capita income might be misleadingly high, for the poor might not even able to afford beef. Therefore, one might expect within country income inequality to affect the country's demand for beef.

Hypothesis 5: Within country income inequality has a negative effect on the quantity of beef demanded in a country.

III. Methodology and data

The focus in this paper is on developing countries because of their large populations and increasing income per capita. As a result of the increasing purchasing power of these countries' inhabitants, their demand for luxury goods and (imported) meat might increase and indirectly have a substantial impact on the environment. The countries investigated in my analysis are the BRICS countries (Brazil, Russia, India, China and South Africa) and the Next Eleven countries, as identified by O'Neill (2001). The Next Eleven countries are Bangladesh, Egypt, Indonesia, Iran, Mexico, Nigeria, Pakistan, the Philippines, Turkey, South Korea, and Vietnam.

The data is retrieved from multiple sources, including the World Bank, the Association of Religion Data Archives (ARDA), and the Organisation for Economic Cooperation and Development (OECD). I combined the data obtained from these different sources into one comprehensive dataset suitable for my purposes. The determinants of beef demand and whether or not there is evidence of the EKC are tested over the period 1983 to 2013. This period is selected for two reasons. First, over this period the emerging economies really started to emerge, so it is likely that their per capita beef consumption increased (heavily). The second reason is more pragmatic: this was the only period for which there was sufficient data available to investigate the research questions. Unfortunately, Russia had to be dropped from the sample due to a lack of data availability. India is also dropped from the sample, for another reason. I will go into details about dropping India later in this section. To get a better feel for the data and, more importantly, to indicate what kind of numbers we are dealing with here, the variables are defined in appendix A, and the descriptive statistics are provided in appendix B.

As this paper tries to answer three distinct research questions, it seems only fitting that I will build different estimation models in order to deal with these questions. The first one is a multiple regression equation with the purpose of determining the determinants of beef demand and their relative importance. The second model will use a multiple regression equation to estimate the environmental Kuznets curve for beef demand and the influence of income inequality. In order to formulate these estimation equations it is important to know which variables are to be included and which variables to control for. Luckily, most of the relevant variables have already been discussed in the previous section of this paper.

III.i Determinants of beef demand

that is consumed, per capita, in developing countries is used as a proxy for the quantity of beef demanded. Hence, the implicit assumption is that all beef that is demanded, is also consumed. This assumption is necessary because demand curves cannot be observed directly, but need to be inferred from consumption behaviour. To be clear, the purpose here is not to model the demand curves. Instead, the goal is to find out which determinants of meat demand are important in emerging economies like the Next Eleven and the BRICS countries.

An undoubtedly important factor for beef demand is a person's income – measured as GDP per capita. Beef is relatively expensive and, thus, it makes sense that persons with higher incomes have a larger budget constraint and can therefore afford to buy more beef. Of course, the mere fact that they can consume more beef does not mean that they do. Yet, one can see how (a certain level of) income is necessary for consumption of any kind.

Multiple studies have consistently emphasized the importance of a number of the same determinants of beef/meat demand. The price and price of substitutes have been found to be particularly important. (Tryfos & Tryfonopoulos, 1973; Van Zyl et al., 1992; Schroeder et al., 2000; Oyewumi & Jooste, 2006; Schroeder et al., 2013). The substitutes that are controlled for in the estimation equation are pork, poultry, and sheep. Both the world price of these meat products and the quantity of these meats demanded (per capita) in each developing country are included in the analysis.

Finally, even though Oyewumi and Jooste (2006) did not find religion to be a significant factor influencing household consumption of pork in Bloemfontijn (South Africa), one might expect religion to be a more important determinant of meat demand at higher levels of aggregation. For example, a predominantly Islamic country will probably have lower per capita demand for pork than a Christian country with a comparable number of inhabitants. In the model, determinants of beef demand in developing countries and their relative importance are estimated. Due to religion possibly being a factor influencing demand for beef, I will control for religion by introducing a religion dummy, and I will drop India from the sample. India is not included in the model because its inhabitants are predominantly Hindu (and Hindu's view cows as sacred and therefore do not eat beef). A religion dummy is included because Jewish people and Islamic people do not eat pork. These people might therefore be inclined to consume more beef. It should be noted that none of the countries included are predominantly Jewish, but there are quite a few Islamic countries. The religion dummy will be one if a country is predominantly Islamic, and zero otherwise. So, the countries for which the religion dummy will take the value of one are: Bangladesh, Egypt, Indonesia, Iran, Pakistan and Turkey.

The initial random effects estimation equation is:

ln

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

otherwise. Lastly, ε is an errorterm.

Before I can estimate this regression equation, possible statistical problems need to be addressed because these statistical problems might lead to biased results or non-sensical outcomes. The most probable statistical problems are heteroskedasticity, multicollinearity, and endogeneity.

If there is heteroskedasticity in the data, this means that the variances of all observations are not random. Basically, if there is heteroskedasticity higher values of a certain variable coincide with higher variability of the random errors. This violates one of the conditions of multiple regression analysis, namely that the standard errors are random. (Hill, Griffiths & Lim, 2012). I conduct a Breusch-Pagan test to test for heteroskedasticity. The Breusch-Pagan (/ Cook-Weisberg) test for heteroskedasticity tests the null hypothesis of constant variances against the alternative hypothesis that the variances are non-constant and therefore not random. For the variables included in the estimation equation above, the Breusch-Pagan test returns a chi-square value of 4.52, with 1 degree of freedom. This coincides with a P-value of 0.0335. Hence, the null hypothesis of constant variances is rejected at the 95% confidence level. In other words, there is indeed evidence of heteroskedasticity in the data. Because the Breusch-Pagan test assumes that heteroskedasticity is a linear function of the independent variables, a White's test for heteroskedasticity is conducted. The White's test returns a chi-square value of 234.56, with 43 degrees of freedom. The P-value obtained by doing the White's test is 0.00. This rules out, without any reasonable doubt, that the variances are constant. Appendix C provides an easy overview of the results of these two statistical tests. Due to the clear evidence of heteroskedasticity, the model is estimated using robust standard errors.

Another possible statistical issue is multicollinearity. Essentially, multicollinearity means that two (independent) variables are highly correlated with each other and therefore the explanatory power of one of these variables undermines the explanatory power of the other (Moore et al., 2011). To test for multicollinearity a table with pairwise correlations is constructed (see appendix D) to identify which variables are susceptible to problems of multicollinearity. For the estimation regarding the determinants of beef demand, there is a strong pairwise correlation between the (natural logarithm of the) world price of beef (lnPwBeef) and the (ln) world price of sheep (lnPwShp), and between lnPoulq and lnGDPpc. For the other variables in the model estimating the determinants of beef demand there is but weak evidence of multicollinearity being an issue (appendix D.1). Including the natural logarithm of the per capita consumption of poultry and the natural logarithm of the world price of sheep meat as a control variable might significantly impede the explanatory power of the natural logarithm of respectively GDP per capita and the world price of beef.

Durbin-Wu-Hausman augmented regression test for endogeneity can be found in appendix E. The test shows that there is an endogenous relationship between the quantity of beef consumed per capita and the world price of beef. To address the endogeneity problem, the multiple regression model will be estimated using fixed effects. An additional benefit of using the fixed effects model to test for hypotheses 1, 2, and 3 is that this model specification automatically controls for (unobserved) cultural differences between countries. The selection of a fixed effects model does mean that the religion dummy, R, is dropped from the estimation equation because fixed effects estimations already control for constant differences between panel data groups.

Taking these statistical issues into account, the initial random effects model needs to be adjusted. To tackle the issues of endogeneity and heteroskedasticity, the model is estimated as a fixed effects model with robust standard errors instead of a normal random effects model. To deal with possible multicollinearity, four largely similar models are estimated.. The first model includes all variables (except for R, of course), the second model includes lnPwShp but not lnPoulq, the third model includes lnPoulq but not lnPwShp, and the last model excludes both these variables.

Model 1:

ln

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

Model 2:

ln

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

Model 3:

ln

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

Model 4:

ln

(

)

(

)

(

)

(

)

(

)

(

)

+β₆ln

(

)

III.ii Environmental Kuznets curve

The initial model with which the last two hypotheses will be tested looks as follows:

ln

(

)

(

)

(

)

2

+β₃Gini_t+β₅ln

(

)

(

)

+β₇ln

(

)

Clearly, the difference between the estimation equations investigating the determinants of beef demand in developing countries and the estimation equations that test for the EKC for beef demand, is that the estimation equation regarding the determinants of beef demand aggregate all countries in order to find some universal truths. The EKC for beef demand, in contrast, is estimated on the individual country level using time series data. There are two reasons for estimating the EKC for beef demand on the individual country level. Firstly, the only way to test for the effect of within country income inequality on beef consumption is to estimate the environmental Kuznets curve hypothesis on the individual country level. Secondly, by estimating the EKC equation this way some issues regarding aggregation bias (Baek, 2015) are automatically addressed. It is easy to illustrate why aggregation bias might exist; a positive result in Bangladesh might be offset by a negative result in Brazil. Hence, on the international level of aggregation there might be no evidence of the EKC, while the EKC hypothesis might still hold in some countries. Therefore, the EKC for beef demand will be estimated in each country separately.

Of course, the statistical problems of heteroskedasticity, endogeneity, and multicollinearity should be taken into account here too. Based on appendix C, it can safely be assumed that there is an issue with heteroskedasticity, and appendix D.2 shows that there is multicollinearity between lnPoulq and lnGDPpc. The estimation will be conducted using robust standard errors to deal with the problem of heteroskedasticity. Furthermore, multicollinearity will be dealt with in largely the same way as I dealt with it before. Two distinct models will be estimated for each country, one in which lnPoulq is included as a control variable, and one in which this variable is excluded. Theoretically, endogeneity is extremely unlikely because the prices of the different meats (the causes- and primary suspects of endogeneity in the estimation of the determinants of beef demand) are not included in the models here. Therefore, this equation is estimated using a random effects model. Another reason to select a random effects model over a fixed effects model is that the test is done on the individual country level, which means that it is not necessary to control for constant differences between countries (as these countries are not compared in the analysis anyway).

Model 5: ln(Beefq)t=β0+β1ln(GDPpc)t+β2ln(GDPpc)t 2 +β₃ln(Porkq)t+β4ln(Poulq)t+β5ln(Shpq)t+εt Model 6: ln(Beefq)t=β0+β1ln(GDPpc)t+β2ln(GDPpc)t 2 +β₃ln(Porkq)t+β4ln(Shpq)t+εt

Thus, model 5 and model 6 are specifically designed to test for H4 on the individual country level. Unlike the first three hypotheses, – which are tested using four variations of a the same multiple regression model, wherein no particular variation specifically tests for any specific hypothesis – hypothesis 4 and 5 need to be estimated separately. The two estimation equations to test for the effect of income inequality on beef consumption per capita (H5) are:

Model 7:

ln(Beefq)t=β0+β1ln(GDPpc)t+β2ln(GDPpc)t 2

+β3Ginit+β4ln(Porkq)t+β5ln(Poulq)t

+β₆ln

(

)

Model 8:

ln(Beefq)t=β0+β1ln(GDPpc)t+β2ln(GDPpc)t

2

+β₃Gini_t+β₄ln(Porkq)t+β5ln(Shpq)t+εt

As one can clearly see; model 5 and model 7 include lnPoulq as a variable, whereas model 6 and model 8 do not.

In the next section, the results from estimating models 5 through 8 on the individual country level are shown. Therefore, the models are named after the country they estimated. For example, Bangladesh 1 refers to the estimates for Bangladesh using model 5, and Turkey B refers to the estimates for Turkey using model 8.

IV. Results

To test the first three hypotheses, the four fixed effects models discussed in the previous section are estimated. The results are shown below in table 2:

Table 2: Determinants of beef demand in developing countries

Variable Model 1 Model 2 Model 3 Model 4

LnPwShp -0.219* (0.021) 0.042 (0.592)

Constant 1.755 0.925 2.720 0.682

Observations 427 427 427 427

Note: * p < 0.05; ** p < 0.01

A couple things about these results stand out, but before discussing the (statistical) significance of individual estimates it is informative to look at the explanatory power of the different models. In appendix F a comprehensive overview is given of the different explanatory powers, the R-squared, of model 1 through 4. With roughly 56%, Model 1 has the highest explanatory power. This can probably be explained by the fact that it has the most variables. It is well known that the R-squared (as a measure of the model's explanatory power) has the tendency to increase whenever more variables are included in the analysis. Hence, even though model 1 has the highest explanatory power, this does not necessarily mean that model 1 is automatically the best model.

From table 2 a couple conclusions can be drawn, regardless of which model best describes the data. For purposes of structure, the findings from model 1 will be discussed first, followed by model 2 and so on.

The only variables that significantly affect beef demand in model 1 are the world price of pork (p-value: 0.042), the quantity of poultry consumed (p-value: 0.001), and the world price of sheep meat (p-value: 0.021). Of these variables, the world price of pork and the quantity of poultry consumed have a positive impact on the quantity of beef consumed, whereas the world price of sheep meat has a negative impact on the quantity of beef consumed. An interesting result from model 1 is that both the world price of beef and gross domestic product per capita are statistically insignificant. This is probably caused by multicollinearity, especially considering that the world price of sheep meat and the quantity of poultry consumed (per capita) are statistically significant while these were the two main suspects of collinearity.

In model 2, the model without lnPoulq, there is only one variable statistically significant: lnGDPpc. So, according to model 2 the only variable that significantly impacts the quantity of beef demanded is income per capita. The other variables included in the model do not have a significant influence on the quantity of beef demanded.

Model 3 is the model without the world price of sheep meat (lnPwShp). The variables with a significant effect on the quantity of beef demanded – in model 3 – are the world price of beef, the world price of pork and the quantity of poultry demanded. As expected, the world price of beef has a negative impact on the quantity of beef demanded. This makes sense theoretically, as a price increase of a normal good would naturally lead to a decrease in demand for that good. According to model 3, the world price of pork and the quantity of poultry consumed (per capita) both have a positive effect on the quantity of beef demanded.

As explained in the methodology section, hypothesis 4 and 5 are estimated on the individual country level. The results obtained by estimating hypothesis 4, using model 5 and model 6, are represented in table 3 through 9. Aside from the tables, the (quadratic) relationship between the quantity of beef consumed and GDP per capita (and implicitly (GDP per capita)2_{) is plotted. These} graphs can be found in appendix G. For Brazil, China, Iran, and the Philippines an inverted U-shaped relationship can be observed, which takes the shape one would expect if the EKC for beef demand exists. For other countries, the relationship between the quantity of beef consumed and income per capita does not resemble the hypothetical EKC shape in the slightest. The individual estimates from model 5 and model 6 are discussed for each country individually, right after the corresponding table.

Table 3: Environmental Kuznets curve for beef demand (Bangladesh and Brazil; H4) Variable Bangladesh 1 Bangladesh 2 Brazil 1 Brazil 2

LnGDPpc 0.937 (0.257) 0.778 (0.287) 3.091*** (0.000) 3.336*** (0.000) LnGDPpc2 -0.068 (0.272) -0.057 (0.301) -0.178*** (0.000) -0.191*** (0.000) LnPorkq -0.001 (0.851) 0.001 (0.704) 0.280* (0.014) 0.315*** (0.000) LnPoulq -0.038 (0.272) 0.047 (0.610) LnShpq -0.191* (0.030) -0.181 (0.027) 0.010 (0.876) -0.013 (0.833) Constant -3.296 -2.709 -10.978** -12.035*** Observations 31 31 29 29 R2 _{0.5734 0.5645 0.9215 0.9205} Note: * p < 0.05; ** p < 0.01; *** p < 0.001

Table 3 shows the result of estimating model 5 and 6 for Bangladesh and Brazil. For Bangladesh, it is clear that the only variable that significantly impacts the quantity of beef demanded is the quantity of sheep meat demanded. If we drop the quantity of poultry demanded as a control variable, then the quantity of sheep meat demanded – too – becomes statistically insignificant. The results for Brazil look more promising for the existence of an environmental Kuznets curve for beef demand. In both model 5 and model 6, the variables lnGDPpc and lnGDPpc2 are highly (statistically) significant. Furthermore, the signs of the coefficients are exactly what one expect them to be if such a thing as an EKC for beef exists. The quantity of pork demanded also appears to be an important driver of demand for beef in Brazil.

Table 4: Environmental Kuznets curve for beef demand (China and Egypt; H4)

Variable China 1 China 2 Egypt 1 Egypt 2

LnGDPpc -0.066 (0.881) 0.934 (0.385) -0.196 (0.874) 1.797 (0.225) LnGDPpc2 -0.009 (0.753) -0.077 (0.250) 0.011 (0.889) -0.104 (0.315) LnPorkq 0.536* (0.036) 2.305*** (0.000) -0.058 (0.215) -0.076 (0.162)

LnPoulq 0.906*** (0.000) 0.606* (0.012)

Constant -1.918 -9.362 1.301 -5.875

Observations 30 30 31 31

R2 _{0.9970 0.9853 0.8044 0.7515}

Note: * p < 0.05; ** p < 0.01; *** p < 0.001

In China, it appears as though income per capita does not affect the quantity of beef demanded at all. Only demand for pork and demand for poultry seem to have a statistically significant (positive) effect on the quantity of beef demanded. When the quantity of poultry demanded is dropped (model 6), the only variable that still significantly affects beef demand is the demand for pork. The results for Egypt do not provide much evidence of the existence of an environmental Kuznets curve for beef demand either. In model 5, only lnPoulq and lnShpq are statistically significant. In model 6, none of the variables are.

Table 5: Environmental Kuznets curve for beef demand (Indonesia and Iran; H4)

Variable Indonesia 1 Indonesia 2 Iran 1 Iran 2

LnGDPpc -0.095 (0.855) 0.794* (0.048) -0.133 (0.922) -0.116 (0.929) LnGDPpc2 0.012 (0.733) -0.045 (0.102) -0.011 (0.902) 0.009 (0.912) LnPorkq 0.088 (0.269) 0.106 (0.193) 0.029 (0.236) 0.032 (0.140) LnPoulq 0.149* (0.015) 0.024 (0.809) LnShpq -0.128 (0.071) -0.173 (0.059) 0.172 (0.403) 0.138 (0.340) Constant 0.057 -3.231 1.610 1.704 Observations 31 31 29 29 R2 _{0.8101 0.7824 0.1040 0.1018} Note: * p < 0.05

In Indonesia, we see that lnPoulq is significant at first (model 5) whilst lnGDPpc is insignificant. However, after the variable lnPoulq is dropped (model 6), income per capita suddenly does become a significant driver of beef demand. In Iran, non of the variables included in either model 5 or 6 appear to affect the quantity of beef demanded in any meaningful way.

Table 6: Environmental Kuznets curve for beef demand (Korea and Mexico; H4)

Variable Korea 1 Korea 2 Mexico 1 Mexico 2

R2 _{0.9251 0.8855 0.6071 0.4793} Note: * p < 0.05; ** p < 0.01; *** p < 0.001

For South Korea there is some evidence of an environmental Kuznets curve for beef demand, especially after lnPoulq is dropped from the model. Furthermore, the quantity of pork consumed per capita also seems to significantly affect the amount of beef consumed per capita. Mexico seems to be an entirely different story. The only variable that significantly impacts the quantity of beef consumed in Mexico seems to be the quantity of poultry consumed (model 5).

Table 7: Environmental Kuznets curve for beef demand (Nigeria and Pakistan; H4)

Variable Nigeria 1 Nigeria 2 Pakistan 1 Pakistan 2

LnGDPpc -0.488 (0.443) -0.837 (0.119) 0.184 (0.897) 1.259 (0.241) LnGDPpc2 0.038 (0.399) 0.062 (0.120) 0.012 (0.906) -0.064 (0.435) LnPorkq -1.174*** (0.000) -1.256*** (0.000) -0.009 (0.465) -0.016 (0.219) LnPoulq -0.653 (0.288) 0.087 (0.312) LnShpq 0.094 (0.814) 0.367 (0.108) 0.162** (0.009) 0.172** (0.005) Constant 2.149 2.967 -0.473 -4.214 Observations 31 31 31 31 R2 _{0.8708 0.8608 0.8905 0.8849} Note: * p < 0.05; ** p < 0.01; *** p < 0.001

The only variable that affects the quantity of beef demanded in Nigeria is the quantity of pork consumed there. This variable is highly significant in both models. There seems to be no evidence of an EKC for beef demand in Nigeria. For Pakistan, the only variable that affects the quantity of beef consumption is the quantity of sheep meat that is consumed per capita. Hence, in Pakistan – too – there apears to be no evidence of an inverted U-shape relationship between the quantity of beef demanded and income per capita.

For the Philippines, model 6 provides some evidence of an EKC for beef demand. Both lnGDPpc and lnGDPpc2 are statistically significant here and have the expected signs, and in both models lnPorkq is statistically significant for the Philippines. The result for South Africa is quite strange. In model 5, GDP per capita does significantly affect the quantity of beef consumed but the coefficient is negative. Meanwhile lnGDPpc2 is statistically significant and has a positive sign. This result indicates that, for South Africa, the demand for beef first falls as income increases and then, after a certain tipping point, the demand for beef starts to increase again. In this model, the quantity of poultry demanded also has a (statistically significant) negative impact on the quantity of beef demanded. After dropping lnPoulq, the effects of GDP per capita and its quadratic counterparts are no longer statistically significant, while the demand for sheep meat is.

Table 9: Environmental Kuznets curve for beef demand (Turkey and Vietnam; H4)

Variable Turkey 1 Turkey 2 Vietnam 1 Vietnam 2

LnGDPpc -3.882 (0.206) -3.788 (0.104) -3.933*** (0.000) -3.876*** (0.000) LnGDPpc2 0.249 (0.189) 0.243 (0.082) 0.358*** (0.000) 0.352*** (0.000) LnPorkq -0.006 (0.901) -0.003 (0.933) 0.134 (0.540) 0.107 (0.471) LnPoulq -0.033 (0.929) -0.030 (0.833) LnShpq 0.976 (0.111) 1.005* (0.020) 0.026 (0.867) 0.033 (0.817) Constant 14.888 14.556 10.974 10.885 Observations 31 31 29 29 R2 _{0.2125 0.2122 0.9600 0.9599} Note: * p < 0.05; ** p < 0.01; *** p < 0.001

In model 5, none of the variables for Turkey significantly affects the quantity of beef demanded by the Turks. In model 6, the quantity of sheep meat consumed per capita is the only variable for which the null hypothesis of no effect on the quantity of beef consumed per capita is rejected. The other variables do not appear to significantly influence the consumption of beef in Turkey. Vietnam is an entirely different story. In both models, both lnGDPpc and lnGDPpc2 are highly statistically significant. The relationship between income per capita and the quantity of beef consumed (per capita) resembles the same relationship found in South Africa (model 5).

Table 10: Environmental Kuznets curve for beef demand (Brazil and Mexico; H5).

Variable Brazil A Brazil B Mexico A Mexico B

LnGDPpc 2.747* (0.046) 3.248*** (0.000) 0.898 (0.776) 1.268 (0.661) LnGDPpc2 -0.157* (0.047) -0.185*** (0.000) -0.055 (0.786) -0.067 (0.692) Gini 0.007 (0.231) 0.005 (0.160) 0.000 (0.980) -0.006 (0.545) LnPorkq 0.262 (0.129) 0.321** (0.001) -0.052 (0.778) -0.087 (0.489) LnPoulq 0.067 (0.635) 0.163 (0.501) LnShpq 0.012 (0.854) -0.011 (0.864) 0.071 (0.633) 0.059 (0.674) Constant -10.006 -12.023 -1.661 -3.132 Observations 25 25 14 14 R2 _{0.9342 0.9327 0.6757 0.6564} Note: * p < 0.05; ** p < 0.01; *** p < 0.001

For Brazil, GDP per capita and lnGDPpc2 are statistically significant in both model 7 and model 8. In model 8, lnPorkq is also statistically significant at the 99% confidence level. For Mexico, none of the variables are statistically significant. The variable Gini is statistically insignificant for Brazil as well as Mexico in both models.

Table 11: Environmental Kuznets curve for beef demand (Pakistan and Philippines, H5) Variable Pakistan A Pakistan B Philippines A Philippines B LnGDPpc -0.785 (0.762) -2.292 (0.328) -2.130 (0.785) 4.145 (0.446) LnGDPpc2 0.101 (0.606) 0.190 (0.290) 0.084 (0.861) -0.281 (0.432) Gini -0.014 (0.466) 0.020 (0.244) 0.029 (0.630) 0.002 (0.971) LnPorkq -0.008 (0.661) 0.003 (0.909) 0.385 (0.601) 1.088** (0.002) LnPoulq -0.534 (0.076) 1.240 (0.333) LnShpq -0.591 (0.197) -0.435 (0.281) 1.088 (0.531) -0.527 (0.453) Constant 3.797 8.236 8.265 (0.781) -17.530 Observations 11 11 10 10 R2 _{0.9534 0.9222 0.9731 0.9616} Note: * p < 0.05; p < 0.01

Table 12: Environmental Kuznets curve for beef demand (Turkey, H5)

Variable Turkey A Turkey B

LnGDPpc -9.984* (0.045) -9.752* (0.047) LnGDPpc2 0.620* (0.032) 0.590* (0.036) Gini 0.078 (0.255) 0.086 (0.234) LnPorkq -0.053 (0.423) 0.021 (0.714) LnPoulq -0.512 (0.095) LnShpq -0.245 (0.873) -0.048 (0.972) Constant 39.281 38.023 (0.098) Observations 14 14 R2 _{0.6149 0.5442} Note: * p < 0.05

Turkey is the last country for which hypothesis 5 is tested. In both models, both lnGDPpc and lnGDPpc2 appear to have a statistically significant effect on the quantity of beef consumed (per capita) at the 95% significance level. However, the Gini coefficient once again does not affect the quantity of beef consumed in any significant way. Therefore, there seems to be no evidence supporting H5 in any of the countries for which the hypothesis is tested.

V. Discussion

The analysis conducted in this paper returned some mixed results. Particularly for hypothesis 4, the existence of an EKC for beef demand, different countries returned different results. The possible reasons for these mixed results are diverse and will largely be explained in the remainder of this section. As explained in section 2, the focus of this study was on demand for beef. The reasons for this included that beef has a larger environmental footprint than other meats (Ratnasiri & Bandara, 2017), and because different types of meat are each others substitutes (Tryfos & Tryfanopoulos, 1973; Van Zyl et al., 1992; Oyewumi & Jooste, 2006). However, just because an EKC tyoe relationship for beef is not found in most countries included in the analysis, does not mean this relationship does not exist for meat in general. Therefore, it is interesting to test the environmental Kuznets curve hypothesis for the total per capita meat consumption in developing countries. After all, the environmental impact of eating any kind of meat is still greater than of a comparable vegetarian diet. To check for the existence of the environmental Kuznets curve for meat demand, a new variable was constructed: Meatq, which indicates the aggregate yearly per capita consumption of beef, pork, poultry and sheep meat (in kilograms). The descriptive statistics of this new variable, disaggregated on the country level, can be found in appendix H. Just like with model 5 through 8, the environmental Kuznets curve will be estimated with a random effects model using time series data on the country level.

Model 9, the environmental Kuznets curve for meat demand: ln(Meatq)t=β0+β1ln(GDPpc)t+β2ln(GDPpc)t

The results obtained by estimating this random effects model can be found in table 13 through 16 below. The results will not be discussed in as much detail as the results of model 5 and model 6. Table 13: EKC for meat demand (Bangladesh, Brazil, China and Egypt)

Variable Bangladesh Brazil China Egypt

lnGDPpc 2.410*** (0.000) 3.307*** (0.001) 4.141** (0.006) 1.392 (0.123) LnGDPpc2 -0.168*** (0.000) -0.169** (0.003) -0.262** (0.009) -0.066 (0.279) Constant -7.245 -11.792 -12.451 -3.806 Observations 31 31 31 31 R2 _{0.9638 0.7790 0.6994 0.8827} Note: * p < 0.05; ** p<0.01; *** p<0.001

Table 14: EKC for meat demand (Indonesia, Iran, Korea, Mexico)

Variable Indonesia Iran Korea Mexico

lnGDPpc 2.543*** (0.000) -3.075* (0.035) 0.306 (0.592) -4.286*** (0.000) lnGDPpc2 -0.156*** (0.000) 0.203* (0.025) 0.015 (0.641) 0.278*** (0.000) Constant -8.017 14.574 -0.723 19.760 Observations 31 29 31 31 R2 _{0.8653 0.3736 0.9507 0.8707} Note: * p < 0.05; ** p<0.01; *** p<0.001

Table 15: EKC for meat demand (Nigeria, Pakistan, Philippines, South Africa)

Variable Nigeria Pakistan Philippines South Africa

lnGDPpc 0.582*** (0.000) 1.518 (0.235) 7.062*** (0.000) -6.271*** (0.000) lnGDPpc2 -0.041*** (0.000) -0.098 (0.315) -0.455*** (0.000) 0.404*** (0.000) Constant -0.196 -3.388 -24.075 27.737 Observations 31 31 31 31 R2 _{0.3822 0.6517 0.9010 0.8900} Note: * p < 0.05; ** p<0.01; *** p<0.001

Table 16: EKC for meat demand (Turkey and Vietnam)

Variable Turkey Vietnam

One can clearly see that as the scope is broadened from beef to meat in general, income per capita becomes an important driver of demand. The only countries for which income per capita is not found to have a statistically significant influence on the consumption of meat per capita are Egypt, South Korea, and Pakistan. For the other countries, lnGDPpc and lnGDPpc2 are consistently significant on – at least – the 95% confidence level, although often at even higher confidence levels. Granted, the coefficients do not always have the signs one would expect if the EKC were to hold, but at least income per capita is found to be a very important driver of demand for meat. To further investigate the possible existence of an EKC for meat demand, the quadratic relationship between lnMeatq and lnGDPpc (and therefore implicitly lnGDPpc2 too) is plotted in appendix I. Looking at the graphs in appendix I, a clear inverted U-shaped relationship between per capita meat consumption and income per capita can be observed for China, Nigeria, and the Philippines. Furthermore, the curve for Bangladesh, Brazil, and Indonesia look as though they might reach the tipping point soon. For the rest of the countries in the sample, it looks as though an increase in income will only cause more demand for meat – and this situation is unlikely to change. Thus, the evidence for the existence of an environmental Kuznets curve continues to be mixed even when different types of meat are all thrown together into one big basket labelled 'meat'.

There are other possible reasons the results obtained in the previous section were mixed, including inherent limitations of the data and methodology used. These limitations will now be discussed.

The first limitation is related to the empirical models with which the determinants of beef demand are investigated. These models do not take the quality- and safety of meat into account. In some countries, the quality or safety of meat might be so low that it has a strong negative impact on consumer demand. Consumers value the safety and quality of the meat products they consume (Schroeder et al., 2013; Oyewumi & Jooste, 2006). Because the quality and safety of beef are not observed in the models, there might be a strong bias against the consumption of beef resulting from a lack of product quality/safety. This problem is somewhat solved by using a fixed effects model for the first four estimation equations. However, ideally it would have been better to be able to quantify food quality/safety and internalize it in the model. This was not feasible, and the suspicion of bias lingers. That food quality/safety cannot be directly observed is also problematic on the country level, where the threat of this unobserved variable distorting the outcome is arguably even larger. Unfortunately, on the individual country level it was not possible to solve this problem. How much, if at all, this unobserved variable distorts the outcome is impossible to say.

A third limitation is related to the data itself. To estimate the determinants of beef demand, the world price of beef, pork, poultry, and sheep meat have been included as control variables. These world prices do not allow for local variability, and do not take purchasing power into account. Ideally, it would have been preferred to use producer prices, corrected for purchasing power and denoted in a single currency (ideally US dollars). Unfortunately, such data simply did not exist. The only data available on producer prices were denoted in local currency and therefore non-comparable across countries and over time. The use of the world price instead of the (local) producer price might lead to bias for countries who are (historically) endowed with large stocks of a certain kind of animal. It is quite possible that lamb is (slightly) cheaper in Bangladesh than in Mexico, irregardless of international trade. These price differences might stem from variability in local demand, and the fact that the retailer will slap on an extra margin over the world price for important meat. Taking all these limitations into account, one may doubt the choice for using the world price as an indicator. However, the world price was used because it was the best way to ensure international comparability, and because – in theory – the world price is not influenced greatly by local demand characteristics. The idea was that using the world price would limit endogeneity, because the local quantity demanded is determined by the price, but the world price is not determined by the local quantity demanded.

A fourth limitation is related to the countries that are included in the analysis. The analysis focusses on developing countries because as these countries get richer they are likely to start consuming more beef (or meat in general), which might accelerate climate change because of beef's large GHG footprint. Yet, the focal point of hypothesis 4 was the existence of an inverted U-shaped relationship – the EKC – for beef demand. The mixed results that were obtained while conducting this analysis, might well be caused because (some of) the developing countries have not yet reached the so-called 'tipping point', the level of income per capita after which demand starts to decrease. Just because, say, Bangladesh has not reached a level of GDP per capita after which demand for beef starts to decrease does not mean they never will. Future research might focus on the existence of an environmental Kuznets curve for beef demand in general, by including advanced economies as well as developing ones.

purely from an academic point of view, to research why the consumption of different meats significantly impacts the demand for beef. For example, in Nigeria, Korea and the Philippines, the per capita consumption of pork influences the per capita consumption of beef (model 5 and 6). Whereas, in Bangladesh (model 5), Pakistan (model 5 and 6), and South Africa (model 6) the per capita consumption of sheep meat influences the per capita consumption of beef. It would be interesting to find out what causes these differences. It makes you wonder if these differences are determined by local customs, historical availability, or whether something else is going on.

VI. Conclusions

In this paper, five hypotheses were formulated to investigate the determinants of beef demand in developing countries and their relative importance, and the existence of an environmental Kuznets curve for beef demand. Because of the global trend towards increasing income inequality, an hypothesis concerning the impact of income inequality on beef demand was also included. These five hypotheses were:

1. Price is the most important determinant of beef demand.

2. The price of other meats has an effect on the quantity of beef demanded.

3. Income per capita is an important determinant of the quantity of beef demanded by an individual.

4. There exists an inverted U-shape relationship between income per capita and demand for beef, with demand for beef first increasing as income increases and later decreasing at high levels of income per capita.

5. Within country income inequality has a negative effect on the quantity of beef demanded in a country.

The empirical evidence paints a mixed picture, but still a couple of conclusions can be drawn about these hypotheses. For purposes of clarity, a comprehensive overview of which models support which hypotheses can be found in appendix J. Now, the conclusions regarding each of the five hypotheses will be discussed in turn. Lets start with the first hypothesis.

The only model in which direct evidence supporting H1 was found, was model 3. In the other models, the world price of beef played virtually no role. However, due to collinearity concerns it cannot be completely ruled out that the price of beef did not play an important part in model 1, especially because its counterpart – the world price of sheep meat – was statistically significant in model 1. Furthermore, the sign of the world price of beef was consistently negative, as expected. Based on the mixed empirical evidence, it can be asserted that the price of beef is an important determinant of beef demand although it is unlikely to be the most important factor. Hence, H1 is not accepted.

For the second hypothesis; the price of other meats has an effect on the quantity of beef demanded, the evidence is mixed as well. Model 1 and model 3 provide clear evidence for H2, but model 2 and model 4 do not. Moreover, the type of meat makes a difference too. It looks as though the price of pork and sheep meat might influence the quantity of beef demanded, whereas the price of poultry does not. Still, in light of the empirical evidence obtained in model 1 and model 3, combined with the phrasing of hypothesis 2, H2 cannot be rejected.

per capita does not affect beef demand. Model 2 and model 4, on the other hand, did find income per capita to have a significant effect on the amount of beef demanded. If you recall that the variables lnGDPpc and lnPoulq exhibited strong pairwise correlation (see appendix D.1) and that model 2 and model 4 were the two models in which lnPoulq was dropped as a variable, then it becomes clear to see that GDP per capita must indeed impact the demand for beef. Therefore, H3 is accepted.

Due to the country level scope of the analysis, and the different results obtained for different countries, it is hard to boil hypothesis 4 down to a simple 'accepted' or 'rejected'. Instead, the countries for which evidence of an environmental Kuznets curve for beef demand was found – and the countries for which such a relationship was not found – will be discussed separately. About most countries, I can be brief. For Bangladesh, China, Egypt, Indonesia, Iran, Mexico, Nigeria, Pakistan, and Turkey no evidence was found supporting hypothesis 4 in either model 5 and model 6. A negative result was also obtained for South Africa (model 5) and Vietnam (model 5 and model 6). Yet, the results obtained for these two countries are interesting because both lnGDPpc and lnGDPpc2 were found to be statistically significant, but the coefficients turned out to have the wrong signs. An inverted U-shape relationship was hypothesized, but the relationship between income per capita and the quantity of beef consumed per capita for South Africa and Vietnam was found to resemble a U. This implies that, as income increases beef consumption first decreases and then, after a tipping point, starts increasing again. For South Korea and the Philippines a mixed result was obtained. No EKC relationship was found for both countries using model 5, although this may have been caused by collinearity as lnPoulq is statistically significant in model 5 and not incorporated in model 6. In model 6, a positive result was obtained for the EKC for beef demand in both South Korea and the Philippines. Evidence of an environmental Kuznets curve for beef demand was consistently found for Brazil, for which a positive result was obtained in both model 5 and model 6 on the 99.9% significance level. Hence, H4 is rejected for all countries except Brazil, South Korea and the Philippines.

References

Al-Mulali, U., Tang, C.F. & Ozturk, I. (2015). Estimating the environment Kuznets curve

hypothesis: Evidence from Latin America and the Caribbean countries. Renewable and Sustainable Energy Reviews, 50, 918-924.

Apergis, N. & Ozturk, I. (2015). Testing environmental Kuznets curve hypothesis in Asian countries. Ecological Indicators, 52, 16-22.

Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C., Holling, C.S., Jansson, B-E., Levin, S., Maler, K-G., Perrings, C. & Pimentel, D. (1995). Economic growth, carrying capacity, and the environment. Ecological Applications, 6(1), 13-15.

Beckerman, W. (1992). Economic growth and the environment: Whose growth? Whose environment? World Development, 20, 481-496.

Baek, J. (2015). Environmental Kuznets curve and CO2 emissions: The case of Arctic countries. Energy Economics, 50, 13-17.

Copeland, B.R. & Taylor, M.S. (2004). Trade, growth and the environment. Journal of Economic Literature, 42, 7-71.

Edjabou, L.D. & Smed, S. (2013). The effects of using consumption taxes on foods to promote climate friendly diets – The case of Denmark. Food Policy, 39, 84-96.

Edwards-Jones, G., Plassmann, K. & Harris, I.M. (2009). Carbon footprinting of lamb and beef production systems: Insights from an empirical analysis of farms in Wales, UK. The Journal of Agricultural Science, 147(6), 707-719.

Gill, A.R., Viswanathan, K.K. & Hassan, S. (2017). Is environmental Kuznets curve stille relevant? International Journal of Energy Economics and Policy, 7(1), 156-165.

Goodland, R. & Anhang, J. (2009). Livestock and climate change: What if the key actors in climate change are... cows, pigs, and chickens? World Watch Magazine 22(6).

Grossman, G.M. & Krueger, A.B. (1991). Environmental impacts of a North American free trade agreement. NBER working paper 3914.

Grossman, G.M. & Krueger, A.B. (1995). Economic growth and the environment. The Quarterly Journal of Economics, 110(2), 353-377.

Hill, R.C., Griffiths, W.E. & Lim, G.C. (2012). Principles of econometrics, Hoboken, NJ: John Wiley & Sons.

Jensen, J.D., Smed, S., Aarup, L., & Nielsen, E. (2015). Effects of the Danish saturated fat tax on the demand for meat and dairy products. Public Health Nutrition, 19(17), 3085-3094.

Kuznets, S. (1955). Economic growth and income inequality. The American Economic Review, 45(1), 1-28.

Liu, H. & Deblitz, C. (2007). Determinants of beef consumption in China, Working paper 41. Moore, D.S., McCabe, G.P., Alwan, L.C., Craig, B.A. & Duckworth, W.M. (2011). The practice of statistics for business and economics, New York, NY: W.H. Freeman and Company.

Moosa, I.A. (2017). The econometrics of the environmental Kuznets curve: An illustration using Australian CO2 emissions. Applied Economics, 1-19.

OECD/FAO. (2016). “Meat”, in Agricultural Outlook 2016-2025, OECD Publishing, Paris. O'Neill, J. (2001). Building better global economics BRICs. Goldman Sachs Global Economics Paper, 66.

Oyewumi, O.A. & Jooste, A. (2006). Measuring the determinants of pork consumption in Bloemfontein, Central South Africa. Agrekon, 45(2), 185-197.

Ozturk, I. & Al-Mulali, U. (2015). Investigating the validity of the environmental Kuznets curve hypothesis in Cambodia. Ecological Indicators, 57, 324-330.

Piketty, T. & Goldhammer, A. (2014). Capital in the twenty-first century, Cambridge, MA: The Belknap Press of Harvard University Press.

Ratnasiri, S. & Bandara, J. (2017). Changing patterns of meat consumption and greenhouse gas emissions in Australia: Will kangaroo meat make a difference? PloS ONE, 12(2), 1-13.

Sall, S. & Gren, I-M. (2015). Effects of an environmental tax on meat and dairy consumption in Sweden. Food Policy, 55, 41-53.

Schroeder, T.C., Marsh, L.T. & Mindert, J. (2000). Beef demand determinants.

Schroeder, T.C. , Tonsor, G. & Mindert, J. (2013). Beef demand: Recent determinants and future drivers.