Agricultural Support Measures Related to
Domestic Agricultural Production for
OECD Member Countries.
H.L. Visser (1779702)1 Supervisor: M.A. Allers University of Groningen
August 31 2012
Abstract: This research tries to estimate the impact of a change in agricultural support on the change in production quantities for fourteen OECD member countries. Using a system of five equations, where each equation represents a commodity group, and Seemingly Unrelated Regression (SUR) analysis does not confirm such a relation.
Keywords: agriculture, policy, production
JEL-code: Q11, Q18
1
2
1.
Introduction
Over the years there has been a lot of discussion about agricultural policies worldwide and their effect on competition and welfare. Since the 1930s and 1940s these policies were implemented with the objective to maintain a stable agricultural sector. However, these policies affect production and prices of several agricultural commodities not only on the national market but also on the world market. Already in 1986 with the start of the Uruguay Round, the WTO members tried to reach an agreement on this subject and did with the Agreement on Agriculture. Currently the Doha round, which started in 2001, is continuing this process.2 Also, in 2003 the European Union reformed its Common Agricultural Policy (CAP) by decoupling subsidies from particular crops. Further reform is expected in the future for example the abolishment of the milk quota in 2015.
The tendency towards reform was mainly initiated in the 1980s when it was recognized that the policies led to large disruptions in the agricultural market. High costs and disturbed trading relations were direct reasons to alter the stance towards policies. Initially subsidies were granted with the objective to secure food supply and reduce dependency on external resources. Providing farmers with a reasonable income and encouraging economic development in the agricultural sector were some additional arguments (Legg, 2003).
This research tries to answer the question whether agricultural support in the OECD member countries influences production quantities of several commodities in those countries. This question is relevant in the light of the current discussion about agricultural support. Furthermore, because the agricultural sector provides in people’s basic needs it influences their real income through production and prices. Understanding the impact of support can contribute to the debate and clarify the advantages and disadvantages of domestic agricultural support.
The academic discussion mainly covers the welfare effects of abolishing agricultural support on different levels and also investigates which kind of support is most detrimental to welfare. For example, Hoekman, Ng, and Olarreaga (2004) find that reductions in border protection of OECD countries will benefit developing countries more than equivalent cuts in subsidies. While previous work mainly focuses on changes from a global view using general equilibrium models, where policy in one country affects welfare in another, this paper tries to
3
identify the effect of variation in domestic agricultural support on output. Besides, many articles deal with a theoretical analysis in which agricultural policy is altered in a major way, reducing all support. This paper deals with the relation between agricultural policies and market outcomes for a number of OECD countries. The reason for focusing on OECD countries is that agricultural policy in these countries is similar. The agricultural sector is protected at high costs, while in many developing countries this sector is heavily taxed (Bale and Lutz, 1981).
Variation in support measures might induce changes in production, after accounting for other explanatory variables. To estimate this relationship a Seemingly Unrelated Regression (SUR) analysis is used with a system of equations. Five equations are estimated, for Cereals, Eggs, Meat, Milk and Other commodities. Production quantity of the commodity is the dependent variable and price support for the good, which has a direct link with output, as well as general support, which does not directly depend on output, are the main explanatory variables. Additional variables are the lag of GDP per capita, total population, a measure for extreme weather events, an index of world energy prices, and the lag of production. The regressions are estimated in first differences
The results do not show a clear relation between variation in agricultural support measures and production quantities. Besides, the predictive power, as measured by R-squared, does not appear to be strong for all equations. Also, expanding the basic model, for example using more detailed data on general support, gives different results depending on the specification. Furthermore, there is a difference in effects on the separate commodity groups in terms of significance and sign.
While examining the results one has to be aware of changes in the environment, which could not be included in the model. For example, when focusing on prices we should take into account the effect of mandatory biofuels policies which divert resources away from one purpose to another (Fischer, 2009). Another point of awareness is the effect of the CAP reform in which direct payments were decoupled from output. Brady et al. (2009) state that the decoupling has negative effects on the landscape under particular circumstances and that it results in higher land-rental prices, especially in those regions where market prices are sufficient to continue to use land for production. This points at the importance of the composition of price support since it can affect production and therefore also price.
4
agricultural economics. Also it includes a discussion of how the OECD estimates agricultural support. The third section deals with the model, the fourth with a description of the data and the fifth with the method used in the research. In section six the main result are discussed, followed by a conclusion.
2.
Literature Review
2.1 Academic literature
The following review of literature in the field of agricultural economics first deals with general equilibrium models and follows with a discussion of studies with a less broad perspective. The issue of decoupling is also touched upon. I will indicate how these studies differ from my own research in this thesis.
We start with Huan-Nimi et al. (2009) which uses the GTAP (Global Trade Analysis Project) multi-region and multi-sector computable general equilibrium model (CGE) to examine the effect of two tariff reduction formulas, one proposed by the EU and the other by the US, where the latter is the largest, on EU production and import and export. The authors find that EU export will decrease dramatically whereas imports will rise. The impact of tariffs, export subsidies and CAP reforms on several crops are decomposed. This suggests that lowering agricultural support would affect production in an impressive way. The study by Huan-Nimi et al. would lead us to expect that a change in price support and general support, as studied in this thesis, are negatively related to the change in production. However, one has to be aware of the differences. My thesis uses a partial equilibrium model, whereas Huan-Nimi et al. uses a general equilibrium model and considers, for example tariffs.
5
Making use of the same GTAP database Burfisher, Robinson, and Thierfelder (2003) define a multi-country computable general equilibrium model (CGE) where they explicitly model agricultural programs. A clear distinction is made between tariffs, export subsidies and domestic support, where the latter two vary with market prices. The agricultural markets of three developed countries, the US, EU and Japan, are considered here. The authors emphasize the importance of assumptions about elasticities in the factor market and the trade market and their effect on production, prices and, specifically, on the sensitivity of supply. They state that the complexity of CGE models is one of its disadvantages. Changing assumptions, therefore, changes outcomes. A counterintuitive result might indicate that elasticities are wrongly interpreted.
From the GTAP studies with a broad viewpoint we move to a study with a view that is more narrow. In a paper by Chavas and Kwansoo (2006) the US butter market is investigated for price dynamics and price volatility in response to a price support program. Like me Chavas and Kwansoo use a partial equilibrium model. However, they use a dynamic Tobit model under time varying volatility. The authors find that an increase (decrease) in price support raises (lowers) expected price and lowers (raises) price volatility through a censoring mechanism in the short run. These effects diminish towards the longer run. However, when the price support is set below market prices it can have a significant and positive effect on expected prices in the long-run. Although the paper focuses on prices in only one market, the result raises the expectation that the explanatory variable for price support in my research would relate to production in a positive way.
The impact of reforms on the cereals and oilseeds (CO) market in the Netherlands is assessed by Lansink and Peerlings (1996). Using a simulation model, the study finds that output of these goods (CO) decline while that of other crops somewhat increases. The reform entails reducing price support and abolishing deficiency payments while introducing a set-aside premium. The results suggest that agricultural support specific to one commodity can influence production decisions of another commodity also. Although the study deals with changes in agricultural support policy we have to be aware that the focus of this study is micro-economical whereas our research deals with macro-economic data.
6
find that part of the price-support subsidy, depending on the target price, is reallocated to consumers through lower prices, due to a moral hazard problem on the supplier-side. This is especially true for generous subsidies. A price-support subsidy also tends to increase production quantities. Lump sum transfers made at the start of the production period address the issue of moral hazard but tend to cause a decrease in production compared to a price-support subsidy. An agricultural price-support measure that does not depend on output or other conditions would, therefore, be expected to have a smaller effect on production quantities. One has to take into account that this study focuses on producer behavior in relation to agricultural policy, other effects have been not been considered. These effects, of which some are considered in this thesis, e.g. weather, could alter production outcomes.
Morley and Morgan (2008) write about the impact of financial support in the European Union on the export of agricultural products of one EU country to another. Productivity differences between countries are not the only reason for export to take place but also agricultural support, which differs in volume although the policy itself is similar. While this study focuses on the CAP member countries, the result might also be applicable to other countries. One country’s financial support affects production decisions of producers which by exporting disturb also foreign producers’ production decisions.
The degree of decoupling of an agricultural support measure is of interest to our research since it deals with two measures of agricultural support: price support and general support. These support measures differ in their relation with production. Hennessy (1998) investigates the impact of decoupling the explicit link between producer support and production decisions with a key role for uncertainty. This is done by comparing (de)coupled policies, followed by an empirical analysis, and a farm simulation. The author states that there are three effects of income support programs which influence the decision making process of the producer. The coupling effect (producer decisions influence the size of support), the wealth effect (higher average income) and the insurance effect (income stabilization) of which the latter two tend to maintain the relation between support and production, even after being officially decoupled. This result implies that lump sum transfers also influence production quantities, although to a smaller extent.
7
wealth and insurance effects, and dynamic effects. These are evaluated according to the degree of decoupling, where more decoupled measures have a smaller direct impact on production and trade compared with the benchmark. The research is based on the evaluation of several micro-economic studies which use the Policy Evaluation Model (PEM), a partial equilibrium model with the aim of evaluating the impact of policy support measures as given in the Policy Support Estimate (PSE) database of the OECD. Like the OECD study, this thesis makes use of the PSE database and a partial equilibrium model but takes a macro-economic perspective. Another OECD paper (OECD, 2005b) distinguishes three issues to take into consideration when evaluating the impact on production: area based payments, policy expectations, and different levels of support. Our research considers area based payments when segregating data based on conditions for payment, which is described in section 6.2.3. Policy expectations are not explicitly addressed. However, it is assumed that producers know beforehand about changes in agricultural support and are therefore able to adjust at the same time these changes occur.
The positive impact of area based payments on production is confirmed in a consecutive paper (OECD, 2005c). This paper primarily focuses on crop insurance subsidies in the Spanish market, which are likely to affect cereal production. Yield insurance is said to have a non-significant impact.
Based on the literature available in the area of agricultural support, one can conclude that agricultural support, whether depending on output or not, does seem to influence production quantities. Also, different measures of support tend to have varying effects. Besides the impact a support measure for a specific commodity has on the output level of the commodity itself, it also tends to influence production quantities of other commodities. Reviewing the available literature shows that this thesis is unique in its specification. Although partial equilibrium models have been tested, as far as I am aware none have been making use of a system of equations for five different commodity groups.
2.2 Estimates of Support
8
General support is measured by the Total Support Estimate of which the payments depending on output are subtracted. The size of these payments depend on the output volume of certain commodities, it is price support per ton multiplied by production quantities. The variable price support is price support per ton of commodity. It is explained in section 4.2 how these two variables are calculated. We start with a description of the Total Support Estimate and its components.
The variables which are used to explain variation in production are based on the support measures as calculated by the OECD.3 Total support, which is given by the Total Support Estimate (TSE), consists of the Producer Support Estimate (PSE), General Services Support Estimate (GSSE) and transfers to consumers from taxpayers (TCT), which is part of the Consumer Support Estimate (CSE). An overview of these support measures can be found in Table 1 which has been taken from a paper by Legg (2003). The Producer Support Estimate, the General Services Support Estimate and the Consumer Support Estimate are discussed separately.
The PSE is described as “the annual monetary value of gross transfers from consumers and taxpayers to agricultural producers, measured at the farm gate level, arising from policy measures that support agriculture, regardless of their nature, objectives or impacts on farm production or income”.4 Before 1999 the Producer Support Estimate was called the Producer Subsidy Equivalence as it refers to the payment necessary to compensate a loss of income due to the removal of a certain policy. It assigns a monetary value to the impact of various policy support measures. Part of the PSE is calculated according to the difference between domestic producer prices and border prices at farm gate level multiplied by production, which is called Market Price Support (MPS). Policy measures which are included in the PSE are: MPS, payments based on output, input use, input constraints, area planted/animal numbers, historical entitlements, overall farming income and others (items A to H).
The General Services Support Estimate is “the annual monetary value of gross transfers to general services provided to agricultural producers collectively, arising from policies that support agriculture regardless of their nature, objectives and impacts on farm production, income, or consumption. The GSSE does not include any transfers to individual
3 OECD, Stat extracts, http://stats.oecd.org/Index.aspx?DataSetCode=CSE_2010# (accessed July 10, 2012) 4 OECD, Introduction to the producer support estimate and related indicators of agricultural support.
9
producers”.5 The General Services Support Estimate encompasses payments for research and development (R&D), agricultural schools, inspection services, infrastructure, marketing and promotion, and public stockholding. Important to note is that this type of support is not given to producers individually.
The Consumer Support Estimate is “the annual monetary value of gross transfers from (to) consumers of agricultural commodities, measured at the farm gate level, arising from policy measures that support agriculture, regardless of their nature, objectives or impacts on
Table 1: Overview of OECD policy measures (Legg, 2003)
I. Producer Support Estimate (PSE) [Total of A to H] A. Market Price Support (net of production levies and excess feed costs)
B. Payments based on output
C. Payments based on area planted/animal numbers
D. Payments based on historical entitlements
E. Payments based on input use
F. Payments based on input constraints
G. Payments based on overall farming income
H. Miscellaneous payments
II. General Services Support Estimate (GSSE) [Total of I to O]
I. Research and development
J. Agricultural schools
K. Inspection services
L. Infrastructure
M. Marketing and promotion
N. Public stockholding
O. Miscellaneous
III. Consumer Support Estimate (CSE) [Total of P to S]
P. Transfers to producers from consumers
Q. Other transfers from consumers
R. Transfers to consumers from taxpayers
S. Excess Feed Cost
IV. Total Support Estimate (TSE) [I + II + III R]
T. Transfers from consumers
U. Transfers from taxpayers
V. Budget revenues
Note: The policy measures above include those that generate both positive and negative transfers. In
particular, the CSE is usually negative, indicating that consumers are taxed as a result of market price support to commodity production. From the CSE only R is included in the TSE because the TSE is net of budget revenues (Q) and P and S are already included in the PSE.
10
consumption of farm products”.5 The CSE includes transfers to producers from consumers, other transfers from consumers, transfers to consumers from taxpayers (TCT), and excess feed costs. Of the Consumer Support Estimate only the transfers to consumers from taxpayers (TCT) are included in the Total Support Estimate along with the Producer Support Estimate and the General Services Support Estimate. Transfers to producers from consumers and excess feed costs are already included in the PSE. Other transfers from consumers should not be included in the TSE. The CSE is negative most of the time which indicates the taxation of consumers (Legg, 2003).
3.
Model
Below the model used in this research is discussed. First, the economic model as well as the hypotheses will be given and explained. Second, the econometric model is elaborated on. 3.1 Economic model
The model used in this paper is a partial equilibrium model, since it only addresses the market for agricultural products. The basic model is:
ΔQit = β2 · Δ Pit + β3 · Δ Sit + β4 · Δ Zit (1)
The subscript i indicates a particular country and the subscript t represent a particular year. The betas stand for the coefficients of the variables. The dependent variable ΔQit is the change
in agricultural production for several commodities and commodity groups. The independent variables of interest are given by Δ Pit and Δ Sit which represent the change in two support
measures that together indicate changes in total agricultural support policy. The first measure is the change in price support per ton of a commodity or group of commodities. The other measure is the total value of support independent of output, also referred to as general support. The lag of the percentage change in GDP per capita, the change in the index of energy prices, the change in the occurrence of extreme weather events, the change in total population, and a lag of the change in production are the main control variables as given by the vector Δ Zit. The aim is to estimate the relation between agricultural policy and
agricultural production. The hypothesis tested is that agricultural support measures positively affect the production of a particular commodity (group). This hypothesis can be viewed in a
5 OECD, Introduction to the producer support estimate and related indicators of agricultural support.
11
broad sense, in which the relation between agricultural support measures and agricultural production in general is considered, and in a narrow sense, in which this relation is specified to a commodity (group).
The signs of the coefficients of the variable for the change in price support and general agricultural support are expected to be positive. An increase in support will probably raise production quantities. This especially applies to price support, which directly relates to output. General support is expected to influence production through a number of indirect linkages. For example, support can be subject to input conditions or the size of planted area. Also, general support might prevent producers from exiting the industry or encourage them to enter by providing an economic profit.
The percentage change in GDP per capita is also believed to have a positive relation with the change in production. As income increases people will demand more agricultural products. Producers will react to this change by increasing output. However, a rise in production might also affect GDP per capita, through the increased income of producers. Therefore, to avoid an endogeneity problem, the first lag of GDP per capita is included. A negative relation is expected between the change in the index of energy prices and the change in production levels.6 Energy is an important input in the agricultural production process and is associated with costs. When costs increase, potential profits decrease and the attractiveness of producing becomes lower. The coefficient of the change in occurrence of extreme weather events is expected to be negative. A negative relation means that as more extreme weather events occur this will tend to decrease agricultural production due to the destruction of crops. The change in total population is also expected to have a positive relation with the change in production. A rise in total population will increase demand for agricultural products and its production. The lag of a change in production can either be positively or negatively related to the current change in production. It is included to capture the change in production which is the proceed of earlier adjustments.
The two measures for agricultural support, which are captured by Δ P and Δ S, are expected to be positively related to the change in production. However they might differ in the size of impact. Those elements which are based on some criteria concerning the producer are expected to be of greater interest than those with characteristics of a lump sum transfer, as
6
OECD, OECD-FAO Agricultural Outlook 2012-2021.
12
was also stated in Phillips et al. (2009) and Hennessy (1998). Price support per ton can be seen as being founded more on producer criteria than the measure for general support. Therefore, it is expected that price support per ton has a larger positive effect on the change in production quantities than general support. The expectations of the variables of interest can be summarized by the following hypotheses:
H1: β2 > 0 (2)
H2: β3 > 0 (3)
H3: β2 > β3 (4)
Where Hypothesis 1 states that the coefficient of the change in price support per ton is larger than zero. The coefficient of the change in general support is also presented to be larger than zero by H2. Hypothesis 3 indicates that the change in price support per ton will have a greater impact on the change in quantity than does the change in general support.
3.2 Econometric model
The econometric model is given below, compared to the economic model above it has been extended with an intercept given by β1t and an error term εit.
ΔQit = β1t + β2 · Δ Pit + β3 · Δ Sit + β4 · Δ Zit + εit (5)
Equation (5) is the basic equation; it is used to specify the system of equations:
ΔQ(eggs)it = β1t + β2 · Δ P(eggs)it + β3 · Δ Sit + β4 · Δ GDP(capita)it-1 + β5 · Δ Enit +
β6 · Δ Weatherit + β7 · Δ Q(eggs)it-1 +β8 · Δ Populationit + εit (6)
Δ Q(milk)it = β9t + β10 · Δ P(milk)it + β11 · Δ Sit + β12 · Δ GDP(capita)it-1 + β13 · Δ Enit
+ β14 · Δ Weatherit + β15 · Δ Q(milk)it-1 + β16 · Δ Populationit + εit (7)
Δ Q(meat)it = β17t + β18 · Δ P(meat)it + β19 · Δ Sit + β20 · Δ GDP(capita)it-1 + β21 · Δ
Enit + β22· Δ Weatherit + β23 · Δ Q(meat)it-1 + β24 · Δ Populationit + εit (8)
Δ Q(cereals)it = β25t + β26 · Δ P(cereals)it + β27 · Δ Sit + β28 · Δ GDP(capita)it-1 + β29 ·
Δ Enit + β30 ·Δ Weatherit + β31 · Δ Q(cereals)it-1 + β32 · Δ Populationit + εit (9)
Δ Q(other)it = β33t + β34 · Δ P(other)it + β35 · Δ Sit + β36 · Δ GDP(capita)it-1 + β37 · Δ
Enit + β38 · Δ Weatherit + β39 · Δ Q(other)it-1 + β40 · Δ Populationit + εit (10)
13
the percentage change in GDP per capita, Δ En denotes the change in the index of world energy prices, Δ Weather is the change in the total of extreme weather events in a particular country, and Δ Population represents the change in total population. We will now present a more thorough explanation of the variables and data sources.
4.
Data
In this section the data used in this research is discussed. Data is gathered for all OECD member countries, where the EU is treated as a single country. We have data for fourteen countries over the period 1986-2010. This period was chosen so as to match the data available on agricultural support estimates, given by the OECD. Among the countries considered is the European Union in the current size of 27 member countries. Other countries are: Australia, Canada, Chile, Iceland, Israel, Japan, The Republic of Korea, Mexico, New Zealand, Norway, Switzerland, Turkey and the United States. Data on support measures for Chile is not available for the years 1986-1989 and for Israel no data is given for the period 1986-1994.
4.1 Dependent variable
The dependent variable used in this regression is the production quantity of a commodity per country per year. Although the OECD gives production levels for commodities which receive market price support (MPS commodities), it does not specify production levels for non-MPS commodities and it does not provide aggregates. Therefore, this data is used only for computing the price support per ton of commodity output; this will be discussed in the next subsection. The Food and Agricultural Organization (FAO) provides more detailed data on production quantities of various commodities, specific goods as well as aggregates.7 Therefore, these variables are used as dependent variables. The commodities considered in the system of equations are Eggs, Milk, indigenous Meat, Cereals, and Other. Where Other represents the commodities not covered by the four preceding categories but on which the OECD does provide data. These five commodity groups were chosen so that data on quantities and support for each group is available for all countries and for most years. An overview of the commodities included in each group can be found in Table 8 in Appendix A.
7
14
For Eggs both the production level and an estimate of price support is readily available. The same goes for Milk. However, the commodity group Meat had to be constructed from several commodities. For example, beef could not be considered as a separate group since data on price support was not present for all countries in the sample. A similar argument applies to the group Cereals. According to the FAO the commodity group Eggs include the production (total weight) of eggs (also for hatching) by domestic birds excluding waste on farms. Milk covers the net production (including feed to livestock) of milk by animal species referred to by the Supply Utilization Accounts. Cereals, which are reported in tons, include crops which are harvested for dry grain; cereals used for feed and silage is excluded. Meat refers to indigenous meat, which includes animal slaughtered within national borders and originating from that particular country. Quantities are given in tons of dressed carcass weight.8 The production quantity of Other commodities is calculated by adding the production levels of the commodities belonging to this group.
The FAO also provides an agricultural production index, which gives the sum of price weighted quantities, where international commodity prices (1999-2001 average) are used. The gross production index shows production quantities without deduction for feed and seed. The production index is used in a general regression, which we start with in the section on results.
4.2 Independent variables
The independent variables price support per ton (P) and general support (S) were calculated from data given by the OECD. The Producer Support Estimate (PSE), which was discussed in subsection 2.2, is used to calculate both variables. Part of the support measures covered by the Producer Support Estimate depend on output, which causes an endogeneity problem if not accounted for in the data. The endogeneity problem arises because this type of support does not only affect the level of production but the production quantity also affects the amount of subsidies paid. For example, when the EU gives €10 per ton of common wheat and the production quantity is 70,000 tons it will spend €700,000. Therefore the payments which were given to specific commodities depending on output were subtracted from the Producer Support Estimate. The PSE (minus output-depending support), the General Services Support Estimate and the transfers to consumers from taxpayers (TCT) together form the variable for general support (S) in the basic regression given by TSE.
15
Since the total subsidies based on output cannot be used directly in the regression due to the endogeneity problem, a new variable is constructed to be able to quantify changes in this type of support other than in output. This is the variable ‘support per ton of commodity output’ (P). Data of the OECD on subsidies depending on output contains the level of production in thousand tons and the value of total support based on output. Dividing the value of total support based on commodity outputs for a particular commodity by the level of production of that same commodity gives a rough estimate of the subsidy given per ton output. This is how, for example, the price support per ton of Eggs is computed. However, data is not available on every commodity for every country. Therefore, some estimates of price support are aggregated to form a group of commodities, for example Cereals. The data on support is given in millions (or billions) of local currency, except for the price support per ton. In appendix C one can find a more detailed description of the calculations that have been made.
4.3 Control variables
One of the control variables is the annual index of world energy prices given by the World Bank in nominal US dollar terms with 2005=100. This index is constructed from the world prices of coal, crude oil and natural gas, where the shares in the index are respectively 4.7, 84.6, and 10.8. The inclusion of the index is meant to catch movements in production due to the price effect of energy. Energy is an important input in agricultural production and can therefore be expected to have some influence on production decisions.9
To address any changes in demand due to increases in income, the lag of the annual GDP per capita growth in percentage per country, also given by the World Bank, is added.10
Since the weather, in particular more extreme weather events such as lack of precipitation, is of great influence on agriculture and the production quantity, it is sensible to include a control variable of this kind. Statistical indicators concerning the weather in a particular country in a particular year are hardly available. Therefore data on extreme weather events, which are assumed to influence agriculture, is taken from the Emergency Event
9 OECD, OECD-FAO Agricultural Outlook 2012-2021.
http://www.oecd.org/site/oecd-faoagriculturaloutlook/summary2012edition-oecd-faoagriculturaloutlook.htm (accessed August 18, 2012)
10 World Bank, “World development indicators (WDI) & Global development finance (GDF)” Databank.
16
Database, called EMDAT.11 The international disaster database provides data on several types of disasters, the ones used in this research are drought, flood, storm and wildfire. A particular event is labeled a disaster if it applies to at least one of the following criteria: ten or more people reported killed, hundred or more people reported affected, declaration of a state of emergency, call for international assistance.12 For the European Union as a whole no data on disaster events was available, therefore disaster events per EU country were obtained and then aggregated. A disadvantage of this approach is that one disaster event which affects multiple countries can result in several country level disasters. So, when the disaster events are aggregated for the EU some disasters will be double-counted compared to, for example the USA. Another disadvantage is that not all disasters included will have a significant impact on agriculture and can also differ in size; also not all events which do affect agriculture are included.
An overview of data sources can be found in Table 9 in Appendix B.
4.4 Omitted variables
Unfortunately, some relevant control variables could not be included because of lacking data. One variable which could influence production of agricultural goods is the tariff rate on imports applied by countries as well as non-tariff barriers to trade. However, there is a lack of availability (only at high costs) of data on this subject especially when considered over time. Besides, the Price Support Estimate already incorporates part of the effect of tariffs in the estimate of Market Price Support (Legg, 2005). Also, policy measures concerning the production and usage of bio-fuels could have an impact on production decisions in the agricultural sector. An example of a crop produced for bio-ethanol in the United States is corn (Yacobucci and Schnepf, 2007). Impacts will probably differ between policies aimed at first generation biofuels and second, or even third, generation biofuels, where the latter technologies tend to rely less on food stocks.
4.5 Correlation
The correlation matrices for the evaluated variables can be found in Table 10 and Table 11 in Appendix D. The variables on quantities of several commodity groups are highly correlated
11
EM-DAT, the International disaster database, “Disaster list” Centre for research on the Epidemiology of Distasters (CRED). http://www.emdat.be/advanced-search (accessed May 18, 2012)
12
17
ranging from 0.86 to 0.98. The highest correlation between price support measures is between Eggs and Meat (0.89) and the lowest between Other and Cereals (0.00). The variables of price support of all groups, except Cereals, are highly correlated with each other. The estimates of support the TSE, the PSE, the CSE, and the GSSE are also highly correlated, although negatively with the CSE. The lowest correlation exists between transfers to consumers from taxpayers (TCT) and the other measures of support. When the first difference of these variables is taken, they are no longer highly correlated. The highest correlation is between PSE and GSSE (0.50). The control variables do not show signs of correlation.
The correlation matrices for the dependent and independent variables can be found in Table 12 till Table 16 in Appendix D. Correlation between dependent and independent variables in a specific equation tends to be high when evaluating the variables in levels. When the first differences are analyzed, correlation decreases. One of the highest correlations exists between the change in the index of energy prices and the percentage change in GDP per capita, which is around 0.31 for all five equations. For the equation Eggs, correlation between the change in general support (using the TSE) and the change in total population is 0.47. Correlation between the change in population and the lag of the change in production is also 0.47 for the equation for Meat.
When two explanatory variables are highly correlated this may reduce the explanatory value of both variables. Fortunately, we can conclude from the above analysis that correlation between the variables in first difference in not high. Therefore, these variables can be estimated jointly without reducing the reliability of results.
5.
Method
5.1 First differences
18
combination. It can be interpreted as the existence of a long run relation between the variables.13 It is tested for by the Pedroni Residual Cointegration Test, which does not give reason to reject the null hypothesis of no cointegration. The test has been conducted for each equation in the system and has been assessed on the panel ADF statistic for the within-dimension, which is most relevant for panel data. Because the variables are not stationary and there is no cointegration the first differences are analyzed.
5.2 Seemingly Unrelated Regression
The SUR (seemingly unrelated regression) analysis is used to estimate the above model; it is a linear regression model with multiple regression equations which have different dependent variables and some different independent variables. The SUR model gives more efficient estimates than an OLS model in a multi equation regression when errors are correlated. It estimates the equations jointly while accounting for correlated errors. Because the equations considered here are highly associated and measured over the same time period and cross sections, the error terms are also expected to be correlated. When the errors would be truly uncorrelated SUR is similar to OLS. This is also true when the independent variables are the same for each equation in the system. The correlation matrix of the residuals for the basic estimation can be found in Table 17 (Appendix E). In general correlation between the residuals is not high. The residuals of Cereals and Other are estimated to be correlated 0.44, this is more than twice as high as the other correlations.
5.3 Alternative variables
Figure 1: Composition of Total Support Estimate (without support depending on output)
Level 1 Level 2 Level 3 Element 1 Element 2
PSE Element 3
TSE GSSE Element 4
TCT Element 5 Element 6
13
19
While the Total Support Estimate, with support depending on output subtracted from it, is used as the variable for general support in the basic regression one could also choose to make use of its components. As is shown in Figure 1 the Total Support Estimate (without support depending on output) consists of the Producer Support Estimate (PSE), the General Services Support Estimate (GSSE), and transfers to consumers from taxpayers (TCT). Instead of using the Total Support Estimate as a single variable for general support, we could include these three components in an alternative regression. Another regression, which goes even into more detail, includes the six elements which together form the Producer Support Estimate. These elements will be discussed in section 6.2.3. This leaves us with three types of regressions, each using variables for general support at a more detailed level. This is indicated in the figure as level 1, level 2, and level 3.
Using more detailed data may add value, since each element contains more specific information than the aggregate and therefore might give different results. It could be that a certain element does influence production decisions while the PSE as a whole does not. Due to, for example, specific information that is lost in the aggregate. The results are explained in section six.
5.4 Population
The countries which are included in the dataset are not of similar size in terms of land or in terms of population. To assess any differences in the estimation between large and small countries an interaction term with total population is added for each independent variable. This because there is no specific weighting method available in the statistical package used in this research.
5.5 Autocorrelation & heteroskedasticity
When there is correlation within variables over time one speaks of autocorrelation. The correlogram of the variables in levels shows signs of autocorrelation. However, when the variables are estimated in first difference the issue of autocorrelation is addressed and the correlogram has improved. Therefore, no further modifications need to be made.
20
6.
Results
6.1 General regression
Table 2: General Regression
Dependent variable: agricultural production in first differences
General Regression Variables, all in first difference 1 2 3 Intercept 0.710 0.733 0.943 (1.012) (1.042) (1.287) Price support -0.000 -0.000 -0.000 (-1.671) * (-1.624) (-1.537)
TSE -2.99E-05 -3.85E-05
(-0.943) (-1.164) PSE 3.49E-05 (0.4545) GSSE -6.59E-05 (-1.317) TCT -1.72E-05 (-0.106) GDP(capita)t-1 0.177 0.177 0.145 (2.399) ** (2.386) ** (1.789) * Energy 0.012 0.011 0.016 (1.334) (1.202) (1.562) Weather -0.008 -0.007 -0.034 (-0.257) (-0.232) (-0.947) Lag of prod. -0.371 -0.372 -0.379 (-6.780) *** (-6.773) *** (-6.819) *** Population 2.46E-07 1.72E-07 9.57E-07
(0.229) (0.160) (0.428)
Price Support t-1 -4.67E-05
(-0.566) TSE t-1 -3.04E-05 (-0.860) Energy t-1 0.010 (0.855) Weather t-1 -0.054 (-1.512) Population t-1 -9.64E-07 (-0.439) Observations 311 311 309 R2 0.205 0.208 0.217
Numbers in parentheses represent t-statistics.
21
Before considering the results of the SUR analysis I would like to start with a more general regression. The dependent variable in this case is an index of agricultural production provided by the FAO. The independent variables are price support per ton, which is an ‘average’ for all commodities, and the general support measure which is independent of output. The control variables are the same as in the basic model. All variables are in first difference. Cross section fixed effects are added and are proven to improve the estimation by a redundant fixed effects test. Three regressions are run; the first is according to the basic model. The second regression makes use of more detailed data for support, the components of the Total Support Estimate. Regression three includes lags of all independent variables.
As can be seen in Table 2, the coefficient of price support per ton is only significantly different from zero in the first regression at the ten percent level. The change in the general support measure (TSE) is not significant in any of the regressions. This is also true for the components of TSE, which are given by PSE, GSSE, and TCT. The coefficient of the change in GDP per capita (lag) is positive and significant at the five percent level for the first two regressions and at the ten percent level for the last. Also, the coefficient for the lag of variation in production receives a minus sign, which is significant at the one percent level for all regressions. The third regression, which includes lags of all independent variables, does not give significant results for the main explanatory variables.
This estimation shows counterintuitive results concerning the sign of the change in price support. Besides, the predictive power of the estimation is low. Whether the system of equations, which makes use of more detailed information, will improve the estimation will be discussed in the next subsections.
6.2 System of equations
6.2.1 Basic regression
22
significantly different from zero for Meat and Other. Its sign is counterintuitive since the occurrence of more extreme events would be expected to decrease production. Two other control variables, which have significant coefficients in this regression, are the lag of the change in production and the change in total population, denoted by Population. The coefficient of the lag of production is positive and significant for Milk and negative and significant for Meat, Cereals and Other. The significance level is one percent, except for the equation for Meat. Population is significantly different from zero at the one percent level for all equations excluding the equation for the commodity Milk. The coefficient is generally small and positive, striking is that it is negative for Cereals.
Table 3: Regression analysis, basic
Dependent variable: agricultural production in first differences
System of equations: SUR
Eggs Milk Meat Cereals Other
Variables, all in first
difference 1 2 3 4 5
Intercept -3125.890 -690358.5 -27465.73 -914623.6 -141.923 (-0.725) (-1.019) (-1.245) (-0.784) (-0.523) Price Support -0.002 -27.071 -0.358 -36.503 -5.01E-05
(-0.004) (-0.220) (-0.285) (-0.076) (-0.007) TSE -0.007 -17.546 -2.612 48.413 0.021 (-0.006) (-0.263) (-1.219) (0.450) (0.776) GDP(capita)t-1 1805.861 356080.7 7038.870 69487.39 -18.193 (1.699) * (2.189) ** (1.330) (0.251) (-0.279) Energy -8.591 -24357.61 248.969 35767.55 -1.368 (-0.060) (-1.091) (0.345) (0.959) (-0.153) Weather -743.245 -84893.72 6611.561 -163095.8 89.649 (-1.589) (-1.134) (2.674) *** (-1.443) (3.168) *** Lag of prod. -0.088 2.569 -0.099 -0.538 -0.326 (-1.526) (3.821) *** (-1.774) * (-12.031) *** (-6.743) *** Population 0.022 -0.697 0.193 -0.538 0.001 (5.351) *** (-1.208) (9.202) *** (2.794) *** (4.099) *** Observations 291 311 311 288 309 R2 0.122 0.064 0.251 0.326 0.157
Numbers in parentheses represent t-statistics.
*** 10 % significance level, ** 5% significance level, * 1% significance level
23
support, given by TSE, is significant in the general regression it no longer is in the system of equations.
6.2.2 Multiple explanatory variables
Rather than using the Total Support Estimate (TSE) as a proxy for general support one could also choose to include all components of the TSE separately and estimate them at the same time. The composition of the TSE was discussed in section 2.2. Like the Total Support Estimate, also the Producer Support Estimate (PSE), the General Services Support Estimate (GSSE), and the transfers from taxpayers to consumers (TCT) do not depend on output. The results of this regression can be found in Table 4. Like in the basic estimation (Table 3) none
Table 4: Regression analysis, multiple explanatory variables
Dependent variable: agricultural production in first differences
System of equations: SUR
Eggs Milk Meat Cereals Other
Variables, all in
first difference 1 2 3 4 5
Intercept -3748.138 -691579.0 -30334.53 -898571.5 -117.583 (-0.871) (-1.012) (-1.367) (-0.766) (-0.430) Price support 0.036 -30.731 -0.296 -8.060 -9.51E-06
(0.078) (-0.249) (-0.236) (-0.017) (-0.001) PSE 2.761 20.026 0.063 84.829 -0.022 (1.453) (0.120) (0.012) (0.314) (-0.326) GSSE -2.260 -42.619 -3.539 18.659 0.042 (-0.901) (-0.371) (-0.953) (0.100) (0.910) TCT -1.004 25.999 -13.484 178.830 0.086 (-0.399) (0.068) (-1.092) (0.283) (0.558) GDP(capita)t-1 1801.088 354542.7 7398.458 65403.32 -20.560 (1.691) * (2.171) ** (1.396) (0.236) (-0.315) Energy -45.718 -25106.10 187.594 35172.02 -0.478 (-0.316) (-1.114) (0.258) (0.935) (-0.053) Weather -697.131 -85041.17 6673.331 -167608.8 88.744 (-1.496) (-1.135) (2.703) *** (-1.477) (3.138) *** Lag of prod. -0.076 2.595 -0.100 -0.538 -0.329 (-1.302) (3.848) *** (-1.794) * (-11.997) *** (-6.809) *** Population 0.022 -0.715 0.196 3.574 0.001 (5.273) *** (-1.221) (9.234) *** (2.750) *** (3.979) *** Observations 291 311 311 288 309 R2 0.132 0.065 0.253 0.326 0.159
Numbers in parentheses represent t-statistics.
24
of the agricultural support measures has a coefficient that is significantly different from zero. The control variables are also similar to the basic regression concerning their sign and significance. However, the sign of Population has altered for the equation Cereals. Overall, this regression does not show results that are much different from the basic regression.
6.2.3 Composition of the PSE
The Producer Support Estimate can be split depending on the conditions for payment. The components are:
support based on commodity output; support based on input use (PSE(input));
payments based on current area planted/animal numbers/receipts/income (A/An/R/I), production required (PSE(PaP));
payments based on non-current A/An/R/I, production required (PSE(PnaP)); payments based on non-current A/An/R/I, production not required (PSE(PnaNP)); payments based on non-commodity criteria (PSE(none));
miscellaneous payments (PSE(miss)).
The italicized term between brackets gives the abbreviation of the variable and is used in Table 5, which displays the results. Data on these components is given by the OECD and is used to estimate a variation of the basic regression as was also discussed in section 5.3. Instead of using the Producer Support Estimate as specified before, six of the above mentioned elements are used. The first component, support based on commodity output, is excluded, since this would cause an endogeneity problem. The General Services Support Estimate and the transfers to consumers from taxpayers (TCT) are also added because, together with the Producer Support Estimate, they add up to the Total Support Estimate. This is in line with the regression using multiple explanatory variables in section 6.2.2. In total we now have eight variables representing general support, rather than one. Data on the composition of the Price Support Estimate is in proportions, thus in percentages of the PSE. Therefore, to obtain its monetary equivalent, the data is multiplied by the value of the PSE.14
By using elements of the Producer Support Estimate, rather than the PSE itself, the information in this Estimate is segregated. The segregation of data enables us to observe the effect of each variable separately and detect any differences with the more aggregated
25
Table 5: Regression analysis, composition of the PSE
Dependent variable: agricultural production in first differences
System of equations: SUR
Eggs Milk Meat Cereals Other
Variables, all in
first difference 1 2 3 4 5
Intercept -3731.749 -651152.5 -38129.65 -975230.5 -100.718 (-0.888) (-0.935) (-1.738) * (-0.821) (-0.361) Price support -0.002 -13.397 -0.167 -50.165 1.85E-05
(-0.006) (-0.095) (-0.132) (-0.097) (0.003) PSE(input) 0.003 -0.108 0.029 -0.279 -0.000 (0.097) (-0.056) (0.499) (-0.093) (-0.446) PSE(PaP) 0.094 2.663 0.139 8.809 0.001 (3.450) *** (0.670) (1.360) (1.397) (0.540) PSE(PnaP) -0.021 5.523 0.190 -4.692 -0.003 (-0.170) (0.272) (0.287) (-0.133) (-0.395) PSE(PnaNP) -0.016 -1.155 -0.264 0.050 -2.25E-05 (-0.559) (-0.259) (-1.882) * (0.007) (-0.013) PSE(none) 0.417 57.430 -1.891 19.539 0.005 (1.916) * (1.668) * (-1.743) * (0.342) (0.340) PSE(miss) 0.448 7.220 -2.118 47.336 -0.003 (2.922) *** (0.281) (-2.613) *** (1.145) (-0.329) GSSE -2.985 -27.955 -3.758 107.196 0.051 (-1.196) (-0.221) (-0.950) (0.526) (1.017) TCT -0.825 1.875 -18.275 248.060 0.107 (-0.328) (0.005) (-1.432) (0.370) (0.655) GDP(capita)t-1 1308.998 356426.0 7453.790 54325.64 -26.424 (1.267) (2.130) ** (1.414) (0.192) (-0.394) Energy 0.789 -31112.45 360.581 29443.79 -1.516 (0.006) (-1.354) (0.500) (0.768) (-0.164) Weather -365.653 -80297.59 5896.640 -224381.5 71.643 (-0.801) (-1.042) (2.414) ** (-1.772) * (2.300) ** Lag of prod. 0.004 2.639 -0.097 -0.528 -0.322 (0.067) (3.804) *** (-1.739) * (-11.168) *** (-6.056) *** Population 0.023 -0.808 0.214 2.298 0.001 (5.364) *** (-1.302) (9.804) *** (2.289) ** (3.720) *** Observations 291 306 306 283 305 R2 0.202 0.076 0.303 0.340 0.163
Numbers in parentheses represent t-statistics.
*** 10 % significance level, ** 5% significance level, * 1% significance level
variable. As discussed below, this regression shows that some coefficients are significantly different from zero. This indicates that in the aggregate some information is lost.
26
have significant coefficients. The only equation receiving a minus sign for the variable PSE(none) is Meat, which is significant at the ten percent level. The same variable also has significant coefficients for Milk and Eggs at the ten percent significance level but these are positive. PSE(miss) receives a negative sign for Meat and a positive one for Eggs both are significant at the one percent level. The GSSE and the TCT both do not have coefficients which are significantly different from zero.
Compared to the basic regression, GDP(capita)t-1 for Eggs is no longer significant. The
coefficients of Weather for Meat and Other are now significant at the five percent level rather
Table 6: Regression analysis, lags
Dependent variable: agricultural production in first differences
System of equations: SUR
Eggs Milk Meat Cereals Other
Variables, all in first difference 1 2 3 4 5 Intercept -4475.822 -543729.3 -24896.78 -747716.6 -135.707 (-1.044) (-0.823) (-1.109) (-0.629) (-0.507) Price Support 0.034 -27.334 -0.421 -27.953 -0.000 (0.062) (-0.193) (-0.333) (-0.057) (-0.035) TSE -0.235 -63.099 -1.713 32.147 0.013 (-0.198) (-0.882) (-0.762) (0.283) (0.467) GDP(capita) t-1 2213.374 185159.5 10158.56 -71958.80 -47.834 (1.941) * (1.116) (1.790) * (-0.245) (-0.706) Energy 14.622 -8522.838 -254.982 53556.97 2.272 (-0.091) (-0.358) (-0.316) (1.286) (0.233) Weather -1097.099 -210681.6 6249.932 -240541.2 24.484 (-2.102) ** (-2.572) ** (2.255) ** (-2.105) ** (0.778) Lag of production -0.129 3.030 -0.095 -0.516 -0.291 (-2.227) ** (4.538) *** (-1.692) * (-11.251) *** (-6.002) *** Population -0.058 20.416 0.143 12.935 0.004 (-1.849) * (4.170) *** (0.866) (1.545) (2.078) ** Price Support t-1 -0.085 -30.736 -0.214 -50.340 0.001 (-0.153) (-0.200) (-0.173) (-1.204) (0.102) TSEt-1 -0.067 92.629 -1.783 -3.572 0.021 (-0.053) (1.335) (-0.777) (-0.031) (0.748) Energy t-1 -28.016 60249.44 -1399.208 53239.13 12.117 (-0.158) (2.288) ** (-1.569) (1.160) (1.127) Weather t-1 -1012.282 -204928.2 -865.302 58804.76 -121.880 (-1.939) * (-2.487) ** (-0.314) (0.472) (-3.666) *** Pop t-1 0082 -21.413 0.051 -8.916 -0.003 (2.580) *** (-4.343) *** (0.305) (-1.072) (-1.618) Observations 290 309 309 286 309 R2 0.154 0.154 0.258 0.332 0.207
Numbers in parentheses represent t-statistics.
27
than the one percent level. For Cereals this variable’s coefficient is estimated to be smaller than zero at the ten percent significance level. Besides, the change in total population is less significant (5%) for Cereals. The R-squared has increased for all equations, thus increasing its predictive power.
6.2.4 Including lags
Another variant of the basic regression is one in which lags of all explanatory variables are added. For this regression we therefore have an intercept, Price Support, the TSE, Energy, Weather, the first lag of Price Support, the TSE, GDP(capita), Energy, and Weather and the lag of production. The predictive power of this model has increased for all equations as can be seen in Table 6.
The coefficients of both price support and general support (TSE) are not significantly different from zero, this also applies to the lags of the two variables.
6.2.5 Population interaction terms
In this estimation interaction terms between the population total and basic explanatory variables are added. Using an interaction term helps to identify whether the effect of one variable is altered by another. One has to be aware that, when an interaction term is significantly different from zero the main effects, which are given by the original variables, lose their meaning (B.F. Braumoeller, 2004). Therefore, the focus is on the significance and sign of the interaction terms. The results are given in Table 7.
28
We can conclude from this regression that the size of the population matters in terms of the magnitude of the effect of a change in either price support or general support on the change in production.
Table 7: Regression Analysis, population interaction terms.
Dependent variable: agricultural production in first differences
System of equations: SUR
Eggs Milk Meat Cereals Other
Variables, all in first
difference 1 2 3 4 5 Intercept 4812.486 2117044 -8678.439 311715.9 195.890 (1.100) (4.232) *** (-0.370) (0.264) (0.728) Population 1.93E-05 -0.048 0.001 -0.005 -8.82E-07 (0.577) (-12.624) *** (6.041) *** (-0.631) (-0.428) Price Support 0.204 97.552 0.105 1099.295 -0.001 (0.443 (0.556) (0.081) (0.381) (-0.129)
Ps*Population -7.06E-07 -5.99E-06 -4.13E-07 -7.83E-05 3.09E-09 (-2.577) ** (-0.482) (-1.157) (-0.427) (0.303)
TSE -2.551 -18.657 2.181 -58.062 -0.033
(-1.130) (-0.343) (0.894) (-0.484) (-1.157) TSE*Population 1.47E-08 3.01E-06 -1.18E-07 2.74E-06 1.62E-09
(1.517) (4.847) *** (-4.206) *** (1.985) ** (4.853) *** GDP(capita)t-1 887.580 -899314.6 3920.796 -129582.5 -70.059
(0.681) (-6.264) *** (0.591) (-0.376) (-0.909) GDP(capita)t-1*Pop. 2.41E-07 0.020 -1.37E-05 0.004 1.37E-06
(0.021) (15.710) *** (0.229) (1.454) (1.943) *
Energy 28.556 34841.64 -453.9425 3125.165 4.260
(0.166) (1.815) * (-0.518) (0.070) (0.411) Energy*Population -3.64E-07 -0.001 1.08E-05 0.000
-1.49E-07
(-0.340) (-6.070) *** (1.936) * (0.603) (-2.194) ** Weather -603.570 -43145.94 -9873.371 -924587.5 -57.922
(-0.386) (-0.246) (-1.233) (-2.336) ** (-0.607) Weather*Population -6.28E-07 -7.62E-05 3.57E-05 0.001 3.24E-07
(-0.174) (-0.189) (1.929) * (1.630) (1.478)
Lag of production 0.489 -1.531 0.577 -0.776 -0.370
(4.556) *** (-1.336) (4.216) *** (-5.549) *** (-2.284) ** Lag of prod.*Pop. -1.47E-09 8.82E-09 -1.61E-09 8.94E-10 2.33E-10
(-4.719) *** (2.934) *** (-4.234) *** (1.987) ** (0.572)
Observations 291 311 311 288 309
R2 0.143 0.525 0.243 0.347 0.218
Numbers in parentheses represent t-statistics.
29 6.3 Summarizing the results
The first explanatory variable Price Support, which measures the change in price support per ton of a commodity (group), has a coefficient that is not significantly different from zero in any of the equations or regressions; this is also true for its lag. This leads us to reject Hypothesis 1, which states that the coefficient of a change in price support is expected to be larger than zero. Nevertheless, the coefficient of a change in price support multiplied by the population variable is negative and significant at the five percent level for Eggs. This indicates that the effect of a change in price support gets smaller as countries get larger in terms of population.
The coefficient of the second explanatory variable, the change in general support (TSE) is also not significant in any of the regressions. Also, when general support is separated and given by PSE, GSSE, and TCT the coefficients of these variables are not significantly different from zero. When a further segregation is made for the variable for general support, some variables’ coefficients are significantly different from zero. These variables are elements of the Producer Support Estimate, as explained in section 6.2.3. The fact that the segregation of data in the regression shows results which are more significant for some equations demonstrates that more detailed information might give better insights. In more aggregate data information is blurred. However, it is not sufficient to confirm Hypothesis 2, which states the expectation that the coefficient of a change in general support is larger than zero. The interaction term of TSE with the population variable shows relevant results for Milk, Meat, Cereals and Other. The coefficients are small and positive for all except Meat. These results state that for most commodity groups an increase in general support has a bigger effect for larger countries.
Besides the rejection of the first two hypotheses, Hypothesis 3 could also not be proven. This contradicts the outcome of Philips et al. (2009) which affirms that a price support subsidy has a larger impact on production than a lump sum transfer.
30
7.
Summary and Conclusion
Contrary to the claims by for example Philips et al. (2009) and the OECD (OECD, 2005a) that agricultural policy influences production decisions of agricultural producers, this research does not find firm evidence for such a link. The above results do not provide enough evidence to conclude that there exists a positive or negative relation between policy support in the agricultural sector and production levels. The estimation of five equations for the commodity groups Eggs, Milk, Meat, Cereals and Other does not show a positive and significant relation between the OECDs estimates of support and the index of production for any equation. Estimating the five equations in a Seemingly Unrelated Regression gives insights in the differing relations that might exist between the variables, depending on the kind of commodity (group). Having more detailed data enlarges the potential for a regression to provide useful information. It could be that the relation between agricultural support and production as measured in this research is distorted by variations in other commodity markets as was pointed out by Keeney and Hertel (2005) and Lansink and Peerlings (1996). Where the latter also showed that a change in agricultural support in the cereals and oilseeds market does affect production. Morley and Morgan( 2008) illustrate the influence of financial support on the export of agricultural products, which might influence supply in foreign markets.
31
8.
References
Bale, M. D., and E. Lutz (1981), Price Distortions in Agriculture and Their Effects: An International Comparison, American Journal Of Agricultural Economics, Vol. 63, No. 1, pp 8-22.
Burfisher, M., S. Robinson, and K. Thierfelder (2003), Modeling OECD Agricultural Programs in a Global Context, Paper presented to the Empirical Trade Conference on Strengthening Analytical Capabilities to Support Trade Negotiations, Washington, D.C.
Brady, M., K. Kellermann, C. Sahrbacher, and L. Jelinek (2009), Impacts of Decoupled Agricultural Support on Farm Structure, Biodiversity and Landscape Mosaic: Some EU Results, Journal Of Agricultural Economics, Vol. 60, No. 3, pp 563-585.
Braumoeller, B.F. (2004), Hypothesis Testing and Multiplicative Interaction Terms, International Organization, Vol. 58, No. 4, pp 807-820.
Chavas, J., and K. Kim (2006), An econometric analysis of the effects of market liberalization on price dynamics and price volatility, Empirical Economics, Vol. 31, No. 1, pp 65–82.
Fischer, G. (2009), World Food and Agriculture to 2030/50: How do climate change and bioenergy alter the long-term outlook for food, agriculture and resource availability? How to feed the World in 2050, Proceedings of a technical meeting of experts, Rome, Italy, 24-26 June 2009 pp 1-49.
Hennessy, D. A. (1998), The production effects of agricultural income support policies under uncertainty, American Journal Of Agricultural Economics, Vol. 80, No. 1, pp 46-57.
32
Huan-Niemi, E., L. Kerkela, H. Lehtonen, and J. Niemi (2009), Implications of Trade Liberalization and Domestic Reforms on EU Agricultural Markets, International Food And Agribusiness Management Review, Vol. 12, No. 4, pp 29-60.
Keeney, R., and T. Hertel (2005), GTAP-AGR: A Framework for Assessing the Implications of Multilateral Changes in Agricultural Policies, GTAP Technical Papers, Paper 25.
Lansink, A., and J. Peerlings (1996), Modelling the New EU Cereals and Oilseeds Regime in the Netherlands, European Review Of Agricultural Economics, Vol. 23, No. 2, pp 161-178.
Legg, W. (2003), Presidential Address Agricultural Subsidies: Measurement and Use in Policy Evaluation. Journal of Agricultural Economics, 54(2), 175-2010.
Morley, B., & Morgan, W. (2008). Causality between Exports, Productivity and Financial Support in European Union Agriculture, Regional Studies, Vol. 42, No. 2, pp 189-198.
OECD (2005a), Decoupling: Policy Implications, OECD Papers, Vol. 12, No. 22, pp 1-27.
OECD (2005b), Decoupling: Illustrating some Open Questions on the Production Impact of Different Policy Instruments, OECD Papers, Vol. 5, No. 11, pp 1-41.
OECD (2005c), The Impact of Crop Insurance Subsidies on Land Allocation and Production in Spain, OECD Papers, Vol. 5, No. 11, pp 1-31.
Phillips, O.R., A.M. Nagler, D.J. Menhaus, and C.T. Bastian (2010), Experimental Work on Subsidies, Moral Hazard, and Market Power in Agricultural Markets, Contemporary Economic Policy, Vol. 28, No. 4, pp 488-501.
33
Appendices
APPENDIX A: COMMODITIES
Table 8: Commodity Groups
Commodity Code Commodity Code
Cereals Other
Barley BA Refined Sugar RS Common Wheat CW Rapeseed RP Durum Wheat DW Ssunflower SF
Maize MA Soybeans SB
Oats OA Wool WL
Other Grains OG Cotton CT
Rice RI Non MPS commodities XE Sorghum SO Beans BN Wheat WT Potatoes PO Apples AP Eggs Grapes GR Eggs EG Tomatoes TM Wine WI Meat Flower FL
Beef and Veal BF Avocados AV
Pigmeat PK Bananas BS
Poultry Meat PT Grapefruit GP
Sheep Meat SH Orange OR
Pepper PB
Milk Peanuts PN
