How did the abolishment of the milk quotas affect the technical efficiency of European dairy farms

(1)

Master thesis

How did the abolishment of the milk quotas affect

the technical efficiency of European dairy farms

Isabelle Elise Dorhout 10528938 15th of July 2018

University of Amsterdam Faculty of Economics & Business Thesis coordinator: Timo Klein and second corrector: Prof. dr. M.P. Schinkel

(2)

Statement of originality

This statement is written by UvA student Isabelle Elise Dorhout. I take the responsibility of the content in this document. I declare the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references are used in creating it. The Economics and Business faculty is only responsible for the supervision of completion of the work, not for the contents_.

Abstract

The European milk quotas have been applicable for a long time. Only in 2015 the quotas have been abolished and it is interesting to investigate how the quotas have influenced the sector. Quotas can have ambiguous effects on technical efficiency. Technical efficiency is the maximum output a farm can produce given its inputs. This research investigates the technical efficiency of the European dairy sector at the time of the abolishment of the milk quotas from 2004 until 2015. This research applies two well-tried methods to measure technical efficiency: the stochastic production frontier and the data envelopment analysis. The stochastic frontier approach requires assumptions about the distribution of the error term and can measure technical efficiency with a significance interval. Where data envelopment analysis is a non-parametric method which does not make assumptions on the functional form but does not give any significance. The data used are averages of in- and outputs of European dairy farms. The results show no significant influence of the abolishment of the quotas on the average technical efficiency of the European milk production. However, using averages of countries is not ideal; farm-based panel data would have given more accurate results. Further research is needed to investigate the heterogeneous effects of the abolishment of the milk quotas in the different European countries.

Keywords: technical efficiency, European milk quotas, stochastic production frontier, DEA.

(3)

1.Introduction

In April 2015 , the European Union abolished the quota system for the European cow milk1 market. The milk quotas (a quota is a maximum amount per farmer for the production of raw milk) were set on 2 April 1984 to stop the butter mountains and milk lakes from growing . 2 The milk quotas have been protecting the farmer from the effects of overproduction, like low prices and dumping at the world market as a consequence. The quotas have been successful at stabilizing the European milk price and production . However, the European 3 quota system has disturbed the milk market: the European raw milk price was above the world price in the quota system decennia. A growing world population causing a growing4 milk demand turned the quotas to be considered as too distortionary . To smoothen the 5 effects on the dairy market the abolishment of the quotas was implemented with a soft landing, meaning the quotas were adjusted bit by bit instead of abolishing the quotas all at once. These adjustments started in 2004. There were many projections on how the abolishment of the milk quotas would influence the milk production of the individual member countries. A lot of assessments were done on how the abolishment of the quotas would influence the market. The soft landing was chosen, which allowed farmers to anticipate and adapt to the “shock” of the quota abolishment. Therefore, the impact of the abolishment was stretched over a longer period of time from 2004 until 2015.

Time to adjust was needed, since farmers could not change their production overnight and the quota system influenced the market for a long time. The milk quotas were introduced in Europe 30 years ago and have disturbed the free market (Falvey, 1975) and could therefore have influenced the technical efficiency of the dairy production. Technical efficiency is almost similar to economic efficiency (Farrell, 1957). Technical efficiency captures the gap between the maximum output a firm can produce given the inputs and its actual production (Xueqin Zhu et al., 2012). However, the output of the milk market was partly restricted, because of the quotas. The influence of the quotas on technical efficiency6 is not researched although technical efficiency in dairy farming is. Thereby a lot of assessments are done on how the abolishment of the quotas would influence the dairy market. The milk quotas are abolished among other reasons to make competition in the

1_{https://ec.europa.eu/agriculture/milk-quota-end_en} 2_{Corron, N., Xue-Zhong He & Frank Westerhoff, 2007}

3_{European commission}_{http://ec.europa.eu/avservices/video/player.cfm?sitelang=en&ref=I101081} End of milk quotas: Increased Market competitiveness

4_{Graph 1 in the literature review.}

5_{European commission http://europa.eu/rapid/press-release_MEMO-15-4697_en.htm} 6_{Depending on how binding and tradable te quotas are.}

(5)

dairy market fair and to improve efficiency. It can be questioned whether this goal of enhancing the technical efficiency is achieved by the abolishment of the quotas in the European milk market.

One could argue that technical efficiency could be improved by having more economies of scale without milk quotas. The opportunity of having economies of scale differs across European countries as stressed by member states, all differing with respect to farm structure and having possibilities to benefit from the abolishment of the quotas.

On the other hand, because of the milk quotas the European milk prices were above world prices and European farmers could have higher margins than in the rest of the world. These margins could cause more investments. Investments improve technical efficiency. It is harder to invest without quotas and lower prices; European milk might be too costly and put farmers into financial risk. The cheapest milk producing countries are in South America . 7 However, Europe has, with 24% of the world milk production, the second largest share of8 milk production in the world. The production could potentially shift to more cost-efficient parts of the world when they improve their technical efficiency, copying the European comparative advantage of technical efficiency in producing milk.

It is interesting to see how the abolishment of the quota system has influenced the technical efficiency of the European milk market. If the technical efficiency does not change with the abolishment of the quotas; this would be an indication that the milk quotas did not disturb the milk market that much.

This thesis focuses on the research question: Does the abolishment of the milk quotas enhance technical efficiency in the European milk market?

This thesis focuses on Europe and it intends to give some insights on technical efficiency in the European milk sector. To know and understand the technical efficiency of Europe one must calculate it first in order to compare it to other parts of the world in order to see the differences and potentials of different countries.

To answer the research question, I applied two methods of measuring technical efficiency: the stochastic frontier approach (SFA) and the data envelopment approach (DEA). The SFA makes assumptions regarding the error term, where the DEA does not. The SFA however can therefore give a confidence interval where the DEA does not. The DEA gives rather efficiency scores per country per year. I applied these methods to test four hypotheses.

7_{IFCN, 2015, Milk prices and production costs worldwide By Torsten Hemme on}

https://www.verantwoordeveehouderij.nl/upload_mm/8/a/0/69580e65-361e-46d0-ab0b-39e97a9a0425 _Article%20Milk%20prices%20and%20production%20costs%20world%20wide.pdf

8_{IDF, International Dairy}

(6)

1. Technical efficiency is enhanced over time, due to the abolishment of the quotas, the soft landing. This hypothesis is not confirmed by the data. A possible explanation could be that the quota system had less effect on the many small producing countries and there is no adjustment for the share of production. However, the quotas may be more binding and restricting for big milk producers than for small producers, so the effect may be underestimated.

2. Technical efficiency is converging between European countries. Convergence follows the first hypothesis. Convergence is tested, because of the differences in technical efficiency across European countries are more likely to disappear when countries can adjust their output. This hypothesis is rejected by the data. It may take more time to see this effect. Moreover, variance might not be appropriate for looking at convergence.

3. The outcome differences between different models and methods are small. This hypothesis is rejected by the data as described in literature. This is an indication that the SFA might have made wrong assumptions regarding the functional form of the production function.

4. There are increasing returns to scale. The hypotheses are put in place to test whether my results are in line with previous research or contradicting and furthermore to see how the abolishment of the quotas influenced the technical efficiency of the European milk market. The returns to scale might be a potential source of heterogeneity which can be the focus of further research.There are significantly increasing returns to scale in this sample. The SFA and DEA assume constant returns to scale and this can explain the results from the other hypotheses.

This thesis is structured in the following way. First, the relevant literature regarding economic theory about quota and the technical concept of efficiency will be discussed as well as the empirics. In the third section, the data will be reviewed. In the fourth section the methodology will be explained; how to measure technical efficiency. Thereafter, the results are shown. These results will be discussed, and critical notes are made in the fifth and sixth section respectively and in the last seventh section conclusions are made. At the end of this thesis, you will find the references and appendix with relevant figures including a special section on the intensifying trend within Europe.

(7)

2. Literature review

Extensive a priori research which considers the heterogeneous effects of the abolishment of the milk quotas on different countries is done. Although this research considers the abolishment ex post, the hypotheses are based on previous literature and therefore these forecasts are reviewed. Thereby is a lot of research done on the technical efficiency of milk farms. Relevant literature on both subjects of technical efficiency and forecasts are reviewed in this section.

The first part of this literature review is about the general economic theory regarding quotas. Following, is a short review on the assessments done on the abolishment of (the milk quotas) and how this will influence the European milk market. Reviewing quota theory is followed by a part of looking at technical efficiency and its literature. This part on technical efficiency is divided into two sections: one is theoretical, technical efficiency in general and its concept. The second part will be more applied and reviewing all the relevant empirical research done on technical efficiency of milk farms in the European Union.

2.1 Quotas

Quotas are the central subject in this thesis therefore relevant literature on general quota theory and applied research to the European milk market will be reviewed.

2.1.1 General Quota theory

The economic theory regarding quotas in general is important to understand the motivation for milk quotas specifically. As has been mentioned, binding quotas effectively disturb the free market. The degree of the distortion caused by the quotas depends on the of elasticities of demand and supply (Falvey, 1975). Quotas are restrictions on supply and therefore influence the market prices . Since the European milk market is one of the major producers9 on the world market, higher European prices resulting from the quota serves to influence the world price (Kuga et al., 2010).

Graph 1 below shows the relationship between the abolishment of the milk quotas and the world and European prices. The prices with the European milk quotas in place, before 2004, are unconforming. When the soft landing starts in 2004 there is a direct and gradual change in the European milk prices. Subsequently, from 2007 onwards is a noticeable effect on the world market prices. The milk prices without the quotas are more

(8)

volatile, but more corresponding to the world market prices (European Commission, 2012). Note the natural seasonal fluctuation in milk prices, because cows give more milk in the lighter times of the year than in the darker days (Ramandan et al., 2014).

Graph 1: the European and world milk prices in times of a soft landing 10

With the quota system, producers were restricted in their production on country-level and could not maximize their profit by adjusting their prices or production. The most convenient way to still maximize profits is to minimize costs, so that only the most cost-efficient farms within a country would survive (Johansson, 2005). In this way, the competition between farms within a country was increased.

10_{graph from:}

http://ec.europa.eu/avservices/video/player.cfm?ref=I101081&videolang=EN&devurl=http://ec.europa. eu/avservices/video/player/config.cfm

(9)

With the abolishment of the quotas in 2015, the market opened up to compete; not only within countries but also between countries. The competitiveness shifts from the focus within countries to a focus between countries. However, it should be noted that quotas were tradable. Nevertheless, they still function as a restriction (OECD, 2006), but milk production is hard to adjust as the supply is inflexible. Therefore, the soft landing adjusted production only gradually and the early announcement of the abolishment of the quotas made it easier for farmers to adjust their production according to their expectations. As such, most farmers expanded their production to anticipate and fulfill the growing world demand. 11

As mentioned before, farmers are not capable of adapting their production capacity on short notice. This means that adjustments to abolishment take place in the long run. However, the adjustments are announced well in time and because of the anticipation of the abolishment, we would expect to see some adjustments in the short-run, which is the focus of this study.

The impact of the quota system varies from country to country. For instance, the values of the quotas are different across countries (Bouamara-Mechemache et al. 2008). This is due to different economies of scale, the degree of the tradability of the quotas and national politics (Bouamra-Mechemache et al., 2008). The EU-15 countries had a free tradable quota system and other countries had restrictions. Trade restrictions of the quota system could explain a lower production.

Another factor stressed in literature causing differences across European countries is the costs of transport play an important role; the further away the milk production takes place from the milk plants or trading platforms, the more expensive the milk becomes. Transport costs form a significant element of the price of milk (Nubern, et al. 1997) and might explain some of the other differences in the valuation of the quota system in different European countries.

The direct quota levels per country are not included in the models. I used time to analyze the effects on technical efficiency, because the quota levels differ per country per year and depend on many different factors. Milk quotas are tradable in some countries, in some extent and depend on the amount of protein and fat mixture of the milk and are not restricting for every country. This makes things too complicated and is therefore out of scope of this thesis. That is why I use time as a type of indirect measure through which the quota influence the European milk production.

11_{Matthew Johnson-dairy analyst at the Rabobank in 2016 at}

https://research.rabobank.com/far/en/sectors/dairy/perplexing-persistence-of-EU-milk-production-grow th.html

(10)

2.1.2 Abolishing the quota system in the European milk market

Many simulations and assessments are done on the abolishment of the milk quotas a priori, because the abolishment of the milk quotas is early announced and the abolishment is done step by step (soft landing). When the soft landing was implemented, the effects of the abolishment were investigated. An increase of milk production was forecasted for Denmark, Ireland, Italy, Luxembourg, the Netherlands and Spain, while the production was predicted to stagnate in Germany, Greece, Portugal and Sweden (M. Lips and P. Rieder, 2004). The EU-wide forecast for raw milk production was an output increase of 3% and a price decline of 22%. This forecast turned out to be accurate. However, the effects of the reduced growth of world and Chinese demand and the Russian embargo in 2014 where unexpected shocks and the losses in demand were divided across Europe, where some countries made losses. For example France produced 2,25% less milk in 2017 and Ireland for example did not12 make any losses. Still, the European milk production increased indeed as forecasted: a 3% increase in total. European production and the differences across Europe are large. Efficiency might explain these differences. However, little research is done on the ex-ante assessment of the abolishment of the milk quota. This thesis will contribute to farming policy evaluations and to general economic theory to see whether implementation of quota has the same, but reverse effects as abolishment.

As mentioned in the previous section, the change in technical efficiency could indicate how distortionary the quotas have been for the milk sector. The degree of distortionary depends on the price elasticity of the demand and supply. The more elastic demand and supply are, the more demand and supply would react on a price change which would mean that a fixed number of a quotas would be distortionary. It has been proven that supply and demand of milk for most European countries are rather inelastic, between 0.2 and 0.3 (Réquillart, 2008). The inelastic behavior would imply that even small shocks could have big price fluctuations. In the US there is a future and an option market for commodities like butter and milk. These future and option markets can potentially smooth out these shocks . However, this is not in the scope of this thesis and for further research to13 investigate. The next subsection will go deeper into the theoretical concept of technical efficiency and why this is a relevant feature of the abolishment of the European milk quotas.

12_{See appendix picture 1 for the division of production losses and gains across Europe.} 13_{http://www.cmegroup.com/trading/agricultural/dairy.html}

(11)

2.2 Technical efficiency

Technical efficiency is the dependent variable of this research. In order to explain the methods to measure technical efficiency in the next section (research method) a general idea of what technical efficiency is and relevant literature applied (to the milk sector) are discussed.

2.2.1 The concept of technical efficiency

Heady (1946) was the first to discuss the concept of ‘Technical Efficiency’ (TE). TE measures the extent to which producers produce optimally. Assuming that producers are rational and that they optimize their output, the TE measure looks at whether the producers produce the maximum output given their input and the technology they use. The departures from the maximum production are partly attributed to technical inefficiency. How the errors in the production function are attributed to technical inefficiency depends on the method used. The production frontier is formed through the observations of the sample and gives the maximal output at a certain input. The observations below this production frontier are considered as inefficient. The technical efficiency is measured as the ratio of the maximum amount of output it could produce given its inputs and divided by the realized output. This is shown in the graph below.

Graph 2: Graphical presentation of the production frontier and technical efficiency_.14

Since 1946 this method has developed a lot. How to define the production function is Important. Translog, Cobb-Douglas, maximum likelihood, parametric and non-parametric functions are used with time-varying and time-invariant features and most important the

(12)

distribution of the error term: half-normal, truncated normal, gamma or exponential. Accordingly, technical efficiency is measured in different ways. Summarizing, there are many ways to estimate technical efficiency. In the scope of this thesis, I will use the methods that are most convenient to use. These might not be the best methods. I will give a short description of the methods and I will discuss the pros and cons of each method.

The stochastic production frontier method is a deterministic model and makes assumptions on the distribution of the error terms and the functional form (Aigner, Lovell, & Schmidt, 1977). The other method I use is the DEA, data envelopment analysis, which is a non-parametric model and does not make assumptions on the functional form, However, this method does not give the significance of the results, but rather the relative yearly efficiency scores per country.

The method to measure technical efficiency is not only used for milk production, but is used for other sectors as well, for example in research on banks and airlines (Favero & Papi, 2006 and Aly et al., 1990). It is a widely used method and often used in the farming and milk sector. In the next subsection, the empirical results of measuring technical efficiency of the European milk sector will be analyzed.

2.2.2. Technical efficiency in the European milk production

As a recent study suggests, further research needs to be done on the role of technical efficiency in the economic performance of the European milk market (Madu et al., 2017). In this research the margins to improve are rather small. Nevertheless, this result must be further verified. This subsection highlights the studies done on the technical efficiency of the milk production in Europe.

A meta-regression analysis on technical efficiency for all kinds of farming in countries all over the world shows that the technical efficiency in Europe is the highest (Bravo-Ureta et al., 2006). Furthermore, meta-analysis shows that the way of measuring TE, time of measuring and the sample size can all make a difference. On average, the way of how the TE is modelled makes only a small difference. For instance, non-parametric deterministic models give higher estimates than the stochastic frontier models. The higher results are due to the assumptions on the distributional form of the stochastic frontier models can be inappropriate.

Farm structures are different across Europe since some countries are more land abundant than others and different environmental rules need to be taken into account. Not only structures and environmental rules explain the differences across Europe’s milk production. At the time of quotas some countries had tradable quotas in different degrees.

(13)

The degree of tradability of the quotas influenced the market (price). For example, the efficiency losses due to distortions in Dutch milk quota trade are researched and the conclusion is that small farms benefit more from the quota trade than large farms and that all farms gain profits when quotas are unconstrained tradable (Boots et al. 1996).

Besides the degree of quota tradability; efficiency is attributed to many factors. For instance, a conclusion for Portugal is that technical efficiency appears positively correlated to farm size, but it is independent to the degree of specialization (Hallam and Machado, 1995). Growing farm sizes is a trend within Europe for decades . It is recognized that the milk 15 quotas influenced the structural change in the milk sector (Huettel and Jongeneel, 2010). How the quotas influenced the structural change in the milk secor is a change in the herd size of the Netherlands and Germany in the quota period, the number of farms that stopped producing milk is lower, but the intensifying increased. Enlarging farms is encouraged by 16 the quota abolishment (Groeneveld et al., 2016). This effect is measured and turned out to be of a stronger impact in the Netherlands than in Western Germany (Huettel and Jongeneel, 2010). Having more and more intensive farms is a trend that the whole of Europe acknowledges (European Commission, 2000). Technical efficiency is needed to intensify farming.

Will the whole of Europe further intensify or will this process of intensification end? And will this process take place predominantly in just a few countries or equally throughout the whole of Europe? There are European countries which have more space to intensify than others. Malta for example is a small country and has an intensive milk producing sector. In 2016 the average farm consisted of 60 cows at 4 hectares (and a stock density of 20 per 17 hectare). Slovakia might have some space to intensify, since the average Slovakian farm has 190 cows at 880 hectares (a stock density of 0,5 per hectare). This is below the European mean density of 30 cows at 30 hectares (and has a stock density of 1.77 per hectare) . It may be true that other more cost-efficient countries will intensify more and end18 up being more efficient than the countries lagging behind. On the other hand, countries with the lowest efficiency may catch up. However, since intensive dairy farming is not good for

15_{This is observed within the sample see the appendix, the intensifying trend.}

16_{Note that intensifying of the dairy sector can be split in a few features. A higher milk yield (more milk} production a cow), less farms with more cows and less land per cow. To see these trends in Europe, look at the graphs in the Appendix.

17_{Stock density= Density of the average number of cows (except calves for fattening) per hectare of} forage utilized agricultural area.

18_{To see the share of production of every country of the total European milk production see the} appendix picture 4.

And to see the shares changing see image 1 and 2 in the appendix and graphs in the appendix under intensifying also for the density and intensifying of the sector within the European Union.

(14)

the environment (Willeke-Wetstein, 1997), in many countries environmental rules limit the intensifying of dairy farming after the abolishment of the quotas.

Intensive farming entails technical efficiency; for example, the Netherlands has relatively efficient and intensive farming (Reinard et al, 2002) around 0.89. The environmental efficiency however, is lower: 0,441. Conclusions from previous research is that intensive farms are more environmentally and technically efficient than extensive farms. However, intensifying the milk production might still get restricted by environmental rules and therewith technical efficiency. Higher technical efficiency lowers the costs, since farms produce closer to their full potential and less wastage.

Still, the efficiency and structures of the milk farms differ across European countries. For example, technical efficiency of Portuguese farms was rather low and was around 60 to 70 percent in the nineties (Hallam and Machando, 1996). This may explain why the Netherlands is producing more milk than Portugal. The Netherlands have a comparative advantage over Portugal due to the fact that the Netherlands has the most technical efficient production of the two countries. This comparative advantage might dissipate with time, since later on the technical efficiency of Portuguese milk farming is estimated a lot higher at around 0.99 from 2004 until 2012 (Madau et al., 2017). However, the estimates in this research seem a lot higher than in other researches. The catching-up effect in technical efficiency within milk producing is relevant for this research and policy.

Interestingly, the source of inefficiency is more or less the same across Europe. For example a higher share of family labour increased the technical efficiency in Dutch and Portuguese farms whereas hired labour decreases technical efficiency in both countries (Xueqin Zhu et al., 2012). This result is in line with other research (Hallam and Machado, 1996). Likewise, subsidies are commonly found to be negatively associated with farm technical efficiency in meta-analysis of empirical results (Minviel and Latruffe, 2016).

The consistency of measuring technical efficiency can be questioned. In the meta-regression analysis for the technical efficiency of farming from Bravo-Ureta et al. (2006) there is consistency in the Dutch milk farming with a technical efficiency of around 80 percent, measured by Asmild in 2004 and by Reinhard in 2000. Only once with a different sample the results are somewhat different, because they include other kinds of farming as well. However, in other research the efficiency of Dutch farms between 2004-2012 was 100% under constant returns to scale as under variable returns to scale and leads to the conclusion that the Netherlands mostly has constant returns to scale regarding milk farming (Madau et al., 2017). The technical efficiency in this research done in different European countries is higher than in the other researches. Furthermore, they conclude that the

(15)

countries with increasing returns to scale are more inefficient, implying that these farms, with increasing returns to scale, to improve on technical efficiency should expand.

3. Data

The data used in this research are data on European farms following the FADN accounting system obtained by the public database from the FADN by the European Commission .19 Every European country will be taken into account, except for Greece, Croatia and Cyprus, since on these countries almost no data are present in the database. In total there are 25 countries and 134 variables, of which 7 will be used. The country-level averages are obtained by taking samples of milk farms. The samples could differ a bit yearly, since not all farms are sampled every year (some are at a maximum of 5 years in a row sampled). Most research is done with panel data on farms, but individual data is hard to obtain, therefore aggregated data will be used. This dataset is suboptimal, as the methods of measuring technical efficiency are based on outliers. Thereby, there is less data, which makes the results of this research less precize. Instead of having a precize result with this dataset there is a more general result. All countries are taken into account, the results are more general to the whole EU instead of just one country or region. On the other hand, averages may be more reliable than individual farms, since there are always a few extreme farms or measurement errors in the sample and they can influence the results more than the averages which are based on a larger sample. If all farms have the same factor costs and the same technical production function and were all allocatively efficient then relative average cost data would be sufficient to measure relative technical efficiency (Timmer, 1971). These assumptions are not likely to hold for farms across European countries. However, these assumptions are more likely to hold for the farms within European countries, since input prices and structures of farms within countries are more homogeneous than across European countries. Moreover, because the production frontier is based on the extreme observations the frontier is highly subject to errors in the data (Timmer, 1971). Errors will be less of a problem with averages since errors are more likely to influence outliers than averages based on a large sample.

The FADN panel data on farms are not flawless, since the sample might not be representative of the sector in every country (Madau, 2017). Mostly the extremely small or large farms are not sampled, so this makes actually the previously mentioned problem less

19_via

http://ec.europa.eu/agriculture/rica/database/report_en.cfm?dwh=SO&CFID=262122&CFTOKEN=978 f949c078e8d9b-76609E82-BFA7-9B04-33C1C37D4942DCFA

(16)

of an issue. However, removing outliers gives bias too. Still, most research use FADN data since there is not that much choice of other existing datasets.

The data that is used for this research are averages of European farms obtained from the years 2004 until 2016. This is only one year after the abolishment. The abolishment is a soft landing starting from 2004 onwards with adjustments in the quota levels; it is still possible to analyze the effects of the abolishments of the quotas. Variables included in this dataset are the so-called input-output variables so the revenue from the milk and the factors needed to produce the milk. With these variables there will be made a Cobb-Douglas production function. With this function the maximum of producing can be estimated. The ratio of the realized, observed production divided by the maximum production is the technical efficiency. Maximum production is the maximum production possible with the same amount of input. Some input is measured as costs, but as recognized by Samuelson (1938) and Shephard (1953), there is a dual relationship between costs and the production level. However, it is important to note that I use a production function and not a cost function, so preferably I use no costs.

Output outcome is typically the amount of produced milk kilos by the country ( _{i) in} time (_t), yearly. Taking yearly averages smooths out the natural cycles of the milk yield. The outcome variable of output is the milk yield, amount of kilos milk given per cow, times the number of dairy cows. The input variables consist typically of: labour hours, feeding costs, machinery and buildings, agricultural area or land, energy and other direct inputs which are the inputs that I use for my model. Since the model I will use is a production function and not a cost function, although related, the input and output variables are given in their own units like kilos of milk output, hours of work and if it is not possible to give them in another unit they will be given in costs. Other direct inputs include water, accountant costs and insurance

. The sample is an unbalanced panel dataset. 20

Notable, not included information in the data are the milk prices . Milk prices started 21 to be more volatile since the soft landing was put into place . Furthermore, the production22 went up with the step by step abolishment of the quotas. The farmers started to anticipate the shock of the abolishment of the quotas by investing more in scale enlargements and intensifying the milk production to fulfill the growing demand to milk. The world demand for milk was not as much growing as expected after the abolishment, because the thrust of the

20_{http://ec.europa.eu/agriculture/rica/definitions_en.cfm}

21_{Milk prices are not needed for measuring technical efficiency and thereby depend on fat, protein etc.} 22_{Graph 1 from the literature review}

(17)

Chinese is lessen(ed) and Russia boycotted. Switzerland abolished the quota already in23 2009 and the farmers did not benefit from it . 24

The data confirms the intensifying of the milk sector; there is almost for every country a yearly increase in the number of cows and the milk yield. To see the intensifying of the milk production on average in the dataset there are some descriptive graphs in the appendix. The trend of intensifying due to policy changes and as a reaction to stay competitive confirmed by the literature (Keane et al., 1997 and Hemme et al., 2004).

The next section will explain how to use this data to measure technical efficiency.

Table 1: Summary statistics of the data

Variabele Mean SD Minimum Maximum

Output 374581.3 316843.9 11747.95 1582845

Feed 87448.78 107179.4 2389 586703

Cows 54.61 43.99 3.60 219.73

Machinery and buildings 12550.24 13611.68 171 67265

Land 101.05 182.13 3.72 1041.92

Energy 14226.24 21447.9 451 147167

Other direct inputs 10736.64 14975.88 48 114715

23_{See picture 3 in the Appendix to see the Chinese infant formula demand growing until 2015,} because of the scandal with the Chinese infant formula in 2008. This has influenced the European milk market. https://www.theguardian.com/world/2008/sep/18/china

(18)

4. Research method

As previously discussed technical efficiency is the ratio of the observed output to the maximum researchable output, given the inputs. The maximum amount of achievable output will be calculated by the production function based on the observed data on input and output (Seiford and Thrall, 1990). For this research two methods are used and compared in order to calculate technical efficiency: the Stochastic Frontier Analysis (SFA) and the Data Envelopment Analysis (DEA). The stochastic frontier analysis using stochastic production frontier models to measure inefficiency. The other method used is a Data Envelopment Analysis is a non-parametric method to measure the technical efficiency and makes no assumptions regarding the functional form. This section is structured as following: first the stochastic production frontier will be explained and thereafter the DEA will be explained. After the methods are explained the hypotheses that will be tested are motivated.

4.1 Stochastic Frontier Analysis (SFA)

The SFA estimates a stochastic production frontier in which the error term partially captures the technical inefficiency. The method allows for estimating both time-varying and time-invariant technical efficiency. Both time-varying and time-invariant models are estimated and compared in order to see whether technical efficiency is dependent on time.

A stochastic production frontier is the most classic way to measure technical efficiency. As mentioned there is a lot of variation on how to split and distribute the error term. Most papers use a time-invariant technical efficiency ratio since it is easier to make assumptions in order to split the error term. However, because this paper investigates the efficiency through time there has to be a time-varying technical efficiency calculated. To calculate the time-varying technical efficiency, I will use another kind of production function, a log-likelihood function. The production functions are estimated by a Cobb-Douglas production functions which assumes constant returns to scale (Romer, 2012).

4.1.1. The time-invariant SFA

The time-invariant production function is a simple Cobb-Douglas production function given by 25

(1)

Y

_it

= β

₀

+ ∑

k

x

ε

j=1

β

_{j ijt}

+

_it

(19)

All the parameters are in natural logarithms.

Where Y_itis the output measured in kilos of milk at an average farm per country _iat time _t which variates per year.β , β ₀ _j and s are to be estimated. β ₀is a constant. x_jitis input _j associated with country _{i at year t} . The first element here is normalized to 1. β_jis an estimated estimated coefficient for each out of _k=7 inputs. The error term is defined as following ε _itis the total error of country _i at time _t. This error consists of

(2)

ε

_it

=

v

_it

− u

_i

is a random variable and is assumed to be independent, random and normally distributed

v_it

(0, ) and is independent from . a random, independent and non-negative variable

N σ2_v u_i u_i

defined by truncation (at zero) of the normal distribution N+( ,μ σ2_u). It is assumed that are random variables independently distributed of the input variables in the and u

v_it _i

production function. Stata has a preprogrammed command to estimate this model, since splitting the error term is challenging. The command I used is xtfrontier. This command is used to estimate the stochastic production frontier (time-varying and time-invariant) and these estimates are used to calculate the technical efficiency. The farm effect is a generalization of a half-normal distribution with a mean of zero (Stevenson, 1980). The farm effects can be seen as the inefficiency and are the u_ipart in the formula. However, this measure makes the technical efficiency time-invariant. This means that farms do not adjust their input value according to the discovery of their own technical inefficiency. The time-invariant model can be used per year, so you measure a different yearly technical efficiency, but the sample gets smaller by doing so. The time-invariant technical efficiency is defined by Battese (1988) by

(3)

T E

_i

=

E(y | u ,x ,t=1,2,...)*it i it

E(y | u =0, x ,t=1,2,...)* it i it

Where y* _{is the measured value of production for country i in year t}. The numerator states

it

the measured output and is divided by the denominator; the maximum attainable output (with is zero; assuming that technical inefficiency is zero), with the same amount of input. The

u _i

technical efficiency or TE is always a value between 0 and 1. If the TE is 0.8 this means the firm produces on average 80% of the total possible, maximum production with the same input values. The part that this model assumes that country-effects are constant through time, is a strong assumption especially during the abolishment of the quotas. The time-invariant model is originally from Battese and Coelli (1988) and features non-negative country-effects.

(20)

4.1.3 Time-varying SFA

Finally, to account for time effects a log-likelihood model will be estimated with the difference in u_it of26

(4)

u

_it

= e

xp(− (t

η − T

_i

))u

_i

Where T_iis the last period in the country _i, _{η is the decay parameter,} v_itand u_i are the same as in the time-invariant model and have the same properties except that u_itis now dependent on time.

When _{η > 0, technical efficiency increases with time; when η < 0, there is a negative} time trend with technical efficiency. Because _{t =} T_i in the last period, the last period for country _{i contains the base level of inefficiency for this country. When η = 0, the time-varying} decay model reduces to the time-invariant model. If the estimate of η is close to zero, then the other estimates are not too far from those of the time-invariant model. Time effects are in this case not of influence on the data in the estimated model. This model time-varying model is different from the time-invariant model in the sense that this time-varying model is based on the assumption that inputs are separable from one and another, but not from time (Fan, 1991). In the short notation 27

xp f(x , , )

xp(v )

xp(−

)

y

_it

= e

_jit

_{t β e}*

_it

_{ e}*

u

_it

xp f(x , , )

xp(v )

Y

_max

= e

_jit

_{t β e}*

_it

T E

_it

=

yit

Ymax

=

exp f(x ,t,β)exp(v )_jit _it

exp f(x ,t,β)exp(v )exp(−u )_jit _it _it

With the time-varying model the technical efficiency is dependent on time with some time trend. The assumption for stochastic models is that there is a common trend in technical efficiency of all producers to split the error term in white noise and technical inefficiency (Greene, 2007). A common trend in technical efficiency is a reasonable assumption for this sample with the averages of Europe, because of the European common market. Since the production is assumed to be positive, so is the efficiency and thus is the distribution of the efficiency not normal, rather truncated or half-normal to exclude the negative values of the distribution. The negative effects in the error term are due to random shocks. To avoid making assumptions about the distribution of the error term we can estimate the production function with a non-parametric model.

The issue of how the error term is split is questioned in many papers. There is no way of knowing if the SFA captures persistent or firm-heterogeneity inefficiency or both. Especially, with the time-varying variant it is hard to know which inefficiency is persistent and

26_{exp stands for exponential.} 27_{As suggested by Bezat, A. (2011)}

(21)

which transient. Put it differently we do not know how to split the country-effects (heterogeneity) from noise (Badunenko, Kumbhakar, 2016). Unfortunately, this is not testable. However, one can compare the efficiency scores with another non-parametric method, the DEA, to have an indication of whether the function specification may be incorrect.

4.2 Data envelopment analysis (DEA)

Data envelopment analysis, as discussed, is a non-parametric method to model a production frontier. Data envelopment analysis is a linear programming method for assessing the efficiency and productivity of the so-called decision-making units (DMUs). In this research European countries function as DMUs, which might be illogical since countries do not choose the amount of input or output. Farms choose the amount of output and input, not a country on average. Although this is the same type of data limitation as I discussed before. There are no assumptions made over the farm-specific error term or the functional form as with SFA.

DEA compares all observations per year and ranks them. With this technique the production efficiency of a DMU is measured by which amount the output can be increased to increase efficiency, given the observed amount of input; this is the same as with the SFA. The difference is in the technique; with DEA the optimally weighted ratio of output to the input of the DMU is used to measure the production efficiency.

DEA measures efficiency of DMUs using linear programming techniques to envelop observed input–output vectors (Boussofiane, Dyson, and Thanassoulis 1991). DEA can be used for multiple inputs and outputs. Technical efficiency is with DEA measured in proportional changes in inputs or outputs. A DEA model can be used as an input-oriented model, which minimizes inputs with the restriction of the at least given output levels (note: this is suggested what quotas do). And used for this research is an output-oriented model, which maximizes outputs without requiring more or less of any observed input levels (more similar to SFA than input-oriented variant). The DEA method does not have a specified distribution as already mentioned, because the efficiency scores are all relative, so the scores do not include a confidence interval.

The DEA is a linear programming technique used to find the set of coefficients (w's and v's) that will give the highest possible efficiency ratio of output to inputs for the countries being evaluated. For the following formulas I am using the following notation:

(22)

j = country number _j.

0 = the country being evaluated.

= amount of output _r produced by country _j

Y_rj

= amount of input _iused by country _j

X_ij

i = number of inputs used by the countries

r = 1 the number of output generated by the countries = coefficient or weight assigned by DEA to output_r

ur

= coefficient or weight assigned by DEA to input _i

v_i Maximize θ₀ = x ∑m i=1vi i0 y ∑s r=1ur r0

Restricted by country 1 until j by

,

x ∑m i=1 v_{i ij} y ∑s r=1ur rj

≤ 1

until 0 U₁ Us ≥ And V₁until Vm ≥ 0

The first restriction states that the technical efficiency cannot be higher than 1 for any country. The second restriction states that the inputs and output should be valued. Multiple optimal weights are very likely to be present. When we solve the DEA model, weights generated by computer runs can be different from each other. This is the reason why many countries are considered as fully efficient. This formula can also be linearized with the denominator taken as 1 (which creates a restriction). This linearization maximizes the numerator for the unit being evaluated, trying to assign it the highest possible efficiency. This is formulated by Maximize θ₀=

∑

y

s r=1

u

_{r r0} Subject to

∑

x

=1 m i=1

v

_{i i0} And

∑

y

s r=1

u

_{r rj}

≤

∑

m

x

i=1

v

_{i ij} j=1 until _n

y

∑

s r=1

u

r rj

−

∑

x

m i=1

v

i ij

,

≤ 0

, > 0

u

r

v

i

(23)

In this lineair form there is a duality. In most researches the efficiency is calculated in an input-oriented approach and the efficiency is minimized. This is actually the same efficiency as calculated in this primal form. Stata has a preprogrammed command for this method that I used, but DEA can also be done by excel.

In earlier research both methods are used and compared. The firm-specific effects (because of the half-normal distribution of the SFA) model make comparisons from different studies less meaningful (Stevenson, 1980). Nevertheless several papers are attributed to find the difference between the different approaches by using the same panel datasets. One of the most relevant studies done is to the technical, allocative and economic efficiency in Swedish milk farms (Johansson, 2005). This research uses an unbalanced panel dataset of Swedish farms and concludes the DEA may be more appropriate to measure efficiency, since it does not assume a parametric form. Technical efficiency was found to be 0.77 in this research (Johansson, 2005). The relation between size and efficiency scores were found to be a significant positive relationship. The trend of intensifying milk farms puts more importance on efficiency. There are fewer and larger farms and a waste of resources can have a big influence on the production, because there are bigger debts and more employees than before (Johansson, 2005).

A noticeable feature of the DEA is that when adding more inputs or outputs (you can use a DEA for multiple outputs, another difference with the SFA) the efficiency scores will never go down (Coelli et al., 2002). A larger amount of output and input compared to the total number of observations will always cause higher efficiency scores. Data envelopment analysis indices should not be used for comparison between different studies, however some meta-studies (Bravo-Ureta et al., 2006 and Minviel and Latruffe, 2016) show the different technical efficiencies in various researches with diverse research methods and it shows that on average the DEA causes higher efficiencies. Johansson finds a difference between the SFA and DEA, but these differences are not significant. Due to the non-pre-specified functional form for the DEA, when the DEA efficiency scores are higher than the SFA, this is an indication that the SFA assumes an inappropriate functional form (Johansson, 2005). Besides the amount of input and output, the DEA can be extremely sensitive to variable selection and data errors (Kalirajan and Shand, 1999). Moreover, in small samples the DEA method is sensitive to the difference between the numbers of farms and the sum of inputs and outputs (Seiford, 1996). Because of this limitation; for this thesis I used the DEA to calculate the efficiency for each country by year. The sample is small, only 25 countries, so this may be a problem. Due to this sensitivity, it may be so that countries may seem efficient although actually they are not. DEA efficiency scores are more likely to

(24)

overestimate efficiency due to a small sample size. In this research it is hard to tell if the different outcomes between SFA and DEA are due to a small sample bias of the DEA or inappropriate assumptions of the SFA regarding the functional form.

Both methods are alike in the sense that they both calculate efficiency by looking at farms and their potential through the sample, so they both measure their relative efficiency. Using DEA connects the points with the lowest inputs for any given output, connecting those points to form the efficiency frontier. Again a country below this line is considered inefficient.

4.3 Hypotheses

As already introduced in the research question I tested four hypotheses. The hypotheses are put in place to test whether my results were in line with previous research or contradicting. Furthermore, to see how the abolishment of the quotas did influence the technical efficiency of the European milk market. This research method is used to test the following hypotheses:

4.3.1 Hypothesis 1: TE is increased with time due to the abolishment of the quotas,

the soft landing.

One could expect the farmers will expand their production after the abolishment of the quotas to fulfill the growing world demand. Since the abolishment of the quotas were step by step, well announced and rational behavior of farmers is assumed to optimize production and therefore enhance their technical efficiency by expanding the production. Before the abolishment producers were restricted on output they could not optimally use their inputs. Thereby the inputs might be altered because of the restricted output, because of the quota. The effect of the quota can be seen as the disturbance of the quota is less and less of influence on the production, because of the soft landing. Technical efficiency may change more before the total abolishment of the milk quotas because of the quota are loosened over time and the anticipation than after 2015. Most of the effect will take place if the quota is totally abolished (after 2015). However, the sample does only include 2015 and 2016, which might not be a time span long enough to capture this effect, however because it is anticipated, it might be.

This hypothesis implies that TE is increasing over time. And therefore we expect the time-varying model of the SA will show a positive eta, indicating an increasing TE over time. Using a DEA one can see the trend of the efficiency scores.

(25)

4.3.2 Hypothesis 2: TE is converging between European countries.

It is expected that the differences in technical efficiency across European countries are becoming smaller, because of the intensifying trend in the milk sector. Since the market for milk is not restricted in production anymore; one would expect more competition and therefore a “catching-up effect” from the less technical efficient European countries.

This hypothesis implies that TE differences between countries are vanishing.

This hypothesis is rather hard to test since the SFA does not show a yearly or country-specific efficiency, so it is hard to compare TE between countries over time. This is due to a relatively small sample of 25 countries. In order to still look at the results of this hypothesis we have to look at the DEA, because the DEA gives efficiency scores for every country in the sample per year. However, because the DEA is only from relative input-output relations, there is no distribution behind these values and I cannot test if these differences are significantly disappearing, but we can look at the variance of the DEA efficiency scores. If the variance of the DEA estimated TE is on average decreasing over time this would mean that the on average the countries are more alike.

4.3.3 Hypothesis 3: The differences between outcomes for TE in different models

(DEA, time-varying and time-invariant SFA) are small.

This is a more technical hypothesis. Although there are many functional forms and variations on and for the production frontier and therefore for measuring the technical efficiency via the SFA (especially in the assumptions regarding the error term); the functional form (DEA or SFA) should not matter that much if the SFA assumes the right distribution (Johansson, 2005). This issue is discussed a lot in the literature, because as already discussed the assumptions being made for the production function are rigorous. Still, the meta-regression analysis and review show that the differences between the TE measured in different studies around the same time and countries differ not much. Because this thesis is using averages instead of individual farm panel data there are no outliers and therefore fewer differences between the models. However, one can argue the opposite is true as well since it is a small sample of just 25 countries, which makes efficient averaging over the data itself hard.

This hypothesis implies the outcomes regarding the measured TE by different models are not that large.

Again, due to the different distributions behind the technical efficiency scores it is hard to compare these methods and outcomes. There is no a significant difference in technical efficiency measurable, because the DEA gives TE scores for every year and country it is a different outcome than the SFA which gives just one estimate for the TE.

(26)

However, one can look at the average of the DEA and see if it is in the confidence interval of the SFA TE. If it is not on average DEA gives a different estimate for TE than the SFA (outside the 95% confidence interval of the SFA).

4.3.4 Hypothesis 4: There are increasing returns to scale.

Models I use (DEA and some part of the SFA) are assuming constant returns to scale, however this assumption is in line with the existing literature. Besides, it is in the first place more convenient to have a production restriction by a quota for a market that has increasing returns to scale and therefore the incentive to overproduce than for a market that has no increasing returns to scale. But does this assumption hold in the sample? It might be that the milk quotas were implemented more for the large milk producers and certain countries than others and thereby on average do not feature increasing returns to scale. If this hypothesis is true, the intensifying trend within the sector is more likely to continue or even increase. This can cause negative externalities for the environment and the market would likely to be restricted again by environmental regulation rather than direct producing quotas. Environmental regulation might have the same effect as the quotas. The difference in effects of environmental regulation instead of quotas depend on the size and preferences of the regulating institute. The environmental regulation can be on European Union level or member country-level.

(27)

5.Results

In this section the results of this research will be presented. The way this section is structured is by showing the results per hypothesis.

5.1 Hypothesis 1: TE is increased with time

First of all, we look at this hypothesis; we are looking at the TE estimates of the SFA and the DEA to look if the results verify. To verify this hypothesis by the SFA there has to be a significant difference between the time-varying and the time-invariant production frontier. If this difference is significant the time-invariant approach with the time-variant approach is that time has a significant effect on the average European technical efficiency.

5.1.1 Results from the SFA

Stochastic production Frontier according to Battese and Coelli (1988) time-invariant

ln(Totallabourinput) 0.04 (0.03) ln(feed) 0.02 (0.01) ln(otherdirectinputs) -0.03 (0.01)** ln(land) -0.11 (0.04)** ln(cows) 0.93 (0.03)** ln(machineryandbuildings) 0.16 (0.02)** ln(energy) 0.07 (0.02)** constant 7.61 N 332 * p<0.05; ** p<0.01

All the inputs are significant for the output except total labour and feed is only significant for 10%, while the other inputs are significant for 5% level or more.

(28)

Mean SD Z-value P-value 95% confidence interval mu -0.25 0.71 -0.35 0.72 -1.64 1.14 lnsigma2 -2.00 1.16 -1.73 0.08 -4.27 0.27 lgtgamma 4.24 1.18 3.60 0.00 1.93 6.55 sigma2 0.13 0.16 0.01 1.30 gamma 0.9857931 0.0165111 0.8731428 0.9985725 sigma_u2 0.13 0.16 -0.17 0.44 sigma_v2 0.00 0.00 0.00 0.00

Gamma is printed in bold, which stands for the technical efficiency. Again u_it is truncated normal with mean mu and variance sigma_u2. Sigma_u2 is the technical inefficiency component and sigma_v2 is the idiosyncratic error component with mean zero and normally distributed. Both of these sigma terms are hetroskedastic with the variance expressed as a function of the covariates defined. Since these parameters are not normally distributed and therefore there is no _Z-value or T-test possible. Another interesting thing are components of _{u and v}, because through the distributional assumptions of these are important in defining the shape of the ε distribution. If these are very similar the profile of the log-likelihood becomes in the flat part and hence here are no solutions found, because they produce nontrivial numerical maximization problems. To check if this is the case one can look at the signal to noise ratio or sigma _{u to sigma v which is in this case undefined which} might indicate that this is indeed the case. This might cause inconsistent results, however as my inputs and model overall is significant and there are not a lot of iterations occurred this problem should not be harmful for the results.

Stochastic production Frontier according to Battese and Coelli (1992) time-varying

Frontier ln(Totallabourinput) 0.02 (0.04) ln(feed) 0.02 (0.01) ln(otherdirectinputs) -0.03 (0.01)** ln(land) -0.11 (0.04)** ln(cows) 0.92 (0.03)** ln(machineryandbuildings) 0.16 (0.02)** ln(energy) 0.08 (0.02)** constant 7.512 (0.13)** N 332

(29)

Mean SD Z-value P-value 95% confidence interval mu -0.27 0.75 -0.36 0.72 -1.73 1.20 eta -0.01 0.00 -1.66 0.10 -0.01 0.00 lnsigma2 -1.99 1.19 -1.68 0.09 -4.32 0.34 lgtgamma 4.26 1.21 3.52 0.00 1.89 6.63 sigma2 0.14 0.16 0.01 1.40 gamma 0.9860428 0.0166346 0.8686076 0.9986772 sigma_u2 0.13 0.16 -0.18 0.45 sigma_v2 0.00 0.00 0.00 0.00

Although we cannot test the differences between the efficiency between the time-varying and the time-invariant efficiency because of the non-normal distribution, we can consider the confidence interval. The time-invariant te is within the confidence interval of the time-varying technical efficiency and we can therefore conclude that time does not make a significant difference in the technical efficiency of the average European farm. Again, input results are kind of the same in the frontier as the time-invariant version.

Furthermore, the time-varying variant gives extra results of eta, which is the time trend. This time trend is negative, but it is not significant different from zero, so we cannot conclude that there is a decreasing technical efficiency through the years of this sample. Eta is defined as following

, where is again truncated normal (at zero) and iid∼ N +(µ, ).

xp(− (t

))u

u

_it

= e

η − T

_i _i u_it σ2

u

When η > 0, the degree of inefficiency decreases over time; When η < 0, the degree of inefficiency increases over time.

Because _{t =}

T

_i in the last period, the last period for country _i contains the base level of inefficiency for that country. If η > 0, the level of inefficiency decays toward the base level. If η < 0, the level of inefficiency increases to the base level.

Eta is zero in the time-invariant model.

The estimate of η is close to zero, and the other estimates are not too far from those of the time-invariant model. This indicates that time is not an important factor in the technical efficiency of the European milk sector in times of the abolishment of the milk quota, which seems counter-intuitive. These results may be explained by the fact that the quotas were not constraining for every member country. Hence on average without adjusting for the production share of each country, time and the quotas have no distortionary effect on the production. These results might be driven by heterogeneous effects of different countries.

(30)

5.1.2 DEA results

As already mentioned at the method of the DEA does not give an interval of the results, because there are no assumptions made about the distribution of the frontier the significance of the results cannot been shown and the results cannot be tested. DEA is performed in this research with constant returns to scale assumption. This might not be appropriate, which will be tested in hypothesis 4.

The DEA allows to attribute an efficiency score for every country every year. This will be used for an indication of hypothesis 2, but for this hypothesis it can give a sense of the direction the average European technical efficiency is going. The graph below shows the average European efficiency according to the DEA through time. The line of efficiency scores is downwards sloping rather than upward sloping indicating an on average decreasing technical efficiency through time. It should be noted however that this line is an unweighted average of all European countries (including Greece, Cyprus and Croatia which are left out of the rest of the analysis) and is unbalanced.

Graph 3: the average European DEA efficiency scores 2004-2016

Taking weighted averages (the percentage share of total production per country in the European Union) without Greece, Cyprus and Croatia, because of the imbalance, the result may be different. However, the data needed to correct for the production share are not all available since this data are confidential.

(31)

However as mentioned in the methodology the DEA method might not be the most reliable method in this case since it is sensitive for which inputs are used and again because of the use of averages. Thereby these DEA efficiency scores are calculated for every year with a comparison of just 25 countries, which might be not a big enough sample to test for which is the case for the SFA.

Hypothesis 1 is not confirmed. There is no significant result for an increasing trend in technical efficiency across Europe on average in the time span of 2004-2016.

How did the abolishment of the milk quotas affect the technical efficiency of European dairy farms

Master thesis