1
The short-term economic impact of the Brexit
vote on the regional inequality in the UK
Master Thesis
University of Groningen
Faculty of Economics and Business
MSc International Economics and Business
ABSTRACT
Even before the Brexit has actually occurred, the UK has experienced a slow-down in its economic growth. Several studies have calculated this slow-down on national level, but no study yet has aimed to calculate the effects on regional level. The regional level has relevancy, as the regional economic disparities between the UK regions are significant. This thesis examines the effects of the Brexit vote on the 41 UK regions on NUTS-2 scale for the period 2016-2017. The synthetic control method has been used in which the 41 UK regions are compared with their “synthetic” UK region. A “synthetic” UK region consists of the weighted average of non-UK regions, which best resemble the economic characteristics of the UK region pre-Brexit vote (and are not affected by the Brexit vote).
Keywords: economic growth, real GDP per capita, Brexit vote, synthetic control method
Author: Sarina van Doorn Student number: S3480429
Mail address: s.o.van.doorn@student.rug.nl Supervisor: prof. Dr. B. Los
2
Table of Contents
1. Introduction ... 3
2. Literature Review ... 5
2.1 The creation of uncertainties and downward expectations by the Brexit Vote ... 5
2.2 The effects of the uncertainties and downward expectations on firm and consumer investments 6 2.3 Studies using the synthetic control method ... 7
2.4 Determinants of regional economic growth ... 9
2.5 Differences in economic characteristics of UK regions ... 11
2.6 Summary... 14
3. Methodology ... 15
3.1 Synthetic control method ... 15
3.2 Inference studies ... 17
3.3 Sensitivity analysis ... 17
4. Data ... 19
4.1 Regional data ... 19
4.2 The outcome variable: real GDP per capita ... 20
4.3 The predictor variables ... 21
5. Results ... 23
5.1 The “synthetic” UK regions... 23
3
1. Introduction
The Brexit vote, which took place on the 23rd of June 2016, has shaken up the economic landscape in the United Kingdom (UK) by creating uncertainties and downward expectations among firms and consumers about the future. These downward expectations and uncertainties have slowed-down the economic growth, as growing numbers of firms pause their spending plans, raising the risk of job losses in the future (Partington, 2019). Furthermore, several firms located in the UK have decided to reallocate parts of their business, staff, assets, or legal entities from the UK to the EU in preparation of the Brexit (Wright et al, 2019).
As firms are holding back, the economy has become increasingly reliant on consumers carrying on spending (Partington, 2019). However, the downward expectations and uncertainties created by the Brexit vote among the consumers have caused a decrease in consumption spending (Born et al, 2018). The decrease in consumption spending is one of the main causes of the output losses experienced by the UK after the Brexit vote. By the end of 2017, the output loss was already 1.8 percent of the GDP (close to 25 billion pound) (Born et al, 2018).
The output losses calculated by Born et al (2018) are for the overall economy of the UK. However, these losses will likely not be representative for the regions within the UK as they vary in the dependency they have on the EU market. It is expected that the regions with the highest dependency experience a larger GDP loss than regions with a lower dependency on the EU market. As it is expected that the Brexit vote will cause a larger increase in uncertainties and downward expectations among firms and consumers in the regions with the highest dependency. Within the UK, the rural areas often have a higher dependency on the EU market than urban areas. Currently, the rural areas are already lagging behind economically compared to the urban areas in the UK due to a higher productivity in the urban areas. The expectation is that the Brexit will further induce the inequality in the UK. Therefore, the focus of this thesis is on the short-term economic impact of the Brexit vote on the regional inequality in the UK. Thus, the central research question in this paper is: What has been the short-term economic
impact of the Brexit vote on the regional inequality in the UK?
4 similar scenarios, they calculate different GDP losses for the UK. Overall, the results of the forecasting studies are inconclusive.
As the aim of this thesis is to identify the economic impact of the Brexit vote for the regions within the UK, the data analysis in this thesis is on regional level. For the data analysis, the dataset consists of all the regions of the 28 member states of the European Union. This data is acquired through the Eurostat regional database and the OECD regional database and is used for the data analysis. The data analysis in this thesis consists of the synthetic control method. For the synthetic control method, the data of the UK regions is compared with a weighted average of non-UK regions, which compose the “synthetic” UK regions. In other words, the regions of all European countries besides the UK can be part of the “synthetic” UK regions. Which of the non-UK regions are included in the “synthetic” UK regions and what weight they get in the “synthetic” UK regions is based on determinants of economic growth and economic output called predictor variables. In this thesis, two different approaches of the synthetic control method are included which have different predictor variables. As in the first approach, the predictor variables included the average of the determinants of economic growth and the economic output: investment ratio, sectoral composition, population density, standards of living, and human capital. While in the second approach of the synthetic control method, the predictor variable included is the yearly pre-Brexit vote economic output of a region: the real GDP per capita. For both approaches of the synthetic control method, the differences between the outcome variable (the real GDP per capita) of the actual UK regions and “synthetic” UK region after the Brexit vote will show the causal effects of the Brexit vote.
5
2. Literature Review
The Brexit vote has created uncertainties and downwards expectations about what the future brings for the UK and the regions within. However, the question remains on what the forces are behind these uncertainties and downwards expectations of the Brexit vote and what the consequences are? In other words, why there is an impact of the Brexit vote and what is the impact of the Brexit vote on the real gross domestic product (GDP) per capita of the UK. To answer these questions the literature review will be structured as follows. Section 1 focuses on why there are uncertainties and downwards expectations. Followed by section 2 which focuses on the economic impact of these uncertainties and downwards expectations. Section 3 contains the method that can be used to calculate the total impact of the Brexit vote. Section 4 contains the determinants of regional economic growth. Then the last section contains the regional disparity within the UK.
2.1 The creation of uncertainties and downward expectations by the Brexit Vote
Before the Brexit referendum, several forecasts were made on how the UK economy would react on the Brexit vote. First, the treasury predicted that the UK would fall into a recession immediately after the Brexit vote (HM Treasury and Osborne, 2016). Second of all, the Bank of England predicted that the UK would keep growing, but that there was a risk of a recession (Tetlow and Stojanovic, 2018). In reality however, the UK did not fall into a recession after the Brexit vote, but the country has experienced a slow-down in economic growth because of the uncertainties and downward expectations created among firms and consumers.
The uncertainties and downwards expectations are first caused by the notion that after the Brexit, the barriers between the EU and UK can only increase. As these barriers increase, it is likely that the trading costs between the two parties will also increase. That the barriers can only increase is caused by the fact that in the last decades, the UK has benefitted from no tariff barriers and customs procedures within the borders of the EU. The EU thereby goes further than any standard free-trade agreement as it gives all its member states access to an internal single market (CBI, 2019). Therefore, a standard trade-agreement between the two parties can never decrease barriers, but only increases it.
In addition, it is still unknown what kind of agreement there will be between the two parties. This indistinctness about what kind of deal there will occur between the two parties induces the uncertainty among the firms and consumer further. As the deal is not certain yet, there is no clarity about how large this increase in barriers will be and which regions/sectors will be affected the most. Moreover, there is still a possibility of a no-deal scenario between the EU and UK, which includes that there is no trade agreement at all between the two parties. Without a trade agreement, the trade rules between the EU and UK are going to be based on the World Trade Organization (WTO) rules. This scenario almost became reality last March. As officially, the Brexit should have occurred the 29th of March 2019. However, due to a lack of agreement
6
2.2 The effects of the uncertainties and downward expectations on firm and consumer investments
The slow-down in economic growth is largely caused by the effects that the uncertainties and downwards expectations of the Brexit vote have had on both the inward and outward investments of firms. According to Barrero et al (2017), the Brexit vote has generated long-term uncertainties about the future of the UK’s trading relationship which resulted in low rates of investments and Research and Development (R&D). For the UK, the inward FDI has seen a substantial decrease of around 11 percent since the Brexit vote, which is close to a loss of 3.5 billion dollar (Breinlich et al, 2019). This is caused by that a growing numbers of firms have paused their spending plans, raising the risk of job losses in the future (Partington, 2019) or even cancelled their investments.
According to Bénassy-Quéré et al (2001), inward FDI to a country decreases when the competitiveness of an alternative country rises. The expectation is that as the UK leaves the EU, the competitiveness of the country compared to other countries will decrease and therefore it becomes less attractive for firms to invest in the UK. An example of a firm that cancelled their investments into the UK after the Brexit vote is Nissan. Nissan announced that the plans to build a new sport utility vehicles at the plants in the UK (Sunderland, Northeast England) have been cancelled. The investments into the plants would have created 740 new jobs (Burroughs, 2019).
Furthermore, for the UK the outward FDI has experienced an increase. As several firms located in the UK and firms with headquarters in the UK have decided to reallocate parts of their business, staff, assets, or legal entities from the UK to the EU in preparation of the Brexit (Wright et al, 2019). Especially, as the extent of these higher barriers and restrictions are unknown this induces further uncertainty and downwards expectations about the increase in trading costs for firms located in the UK. This induces firms to reallocate from the UK to the EU. According to Breinlich et al (2019) by the third quarter of 2018, there were 181 Greenfield and M&A transactions (value around 8.3 billion dollar) from the UK to the EU that would not have occurred in the absence of the Brexit. The increase in outward FDI can be seen as a lost investment for the UK. As the assumption can be made that in the absence of the Brexit, these investments would have occurred in the UK (inward FDI) instead of the European countries (outward FDI).
7 plans which raises the risk of job losses in the future (Partington, 2019). The uncertainty among consumer about future income is further induced by the notion that in the total UK economy more than 2.5 million jobs are exposed to the trade effects of Brexit, which is around 8.2 percent of the total employment (Los et al, 2017a).
Furthermore, another potential cause for these downwards expectation is the effect the Brexit vote has had on the current living standards of the households. Directly after the Brexit vote the pound experienced a large decline in value because of an increase in uncertainty (Breinlich et al, 2017). The lower exchange rate raised the costs of importing consumption goods and intermediate inputs. In addition, the depreciation in the exchange rate has caused a rise in the inflation of the UK since the Brexit vote. As for the UK, the Consumer Price Index inflation increased from 0.4 percent in June 2016 to 3 percent in September 2017 (Breinlich et al, 2017). For this increase in inflation to have no effect on the living standards of household, the increase in inflation should be compensated by an income rise. However, the rate of nominal wages has not experienced a change in growth after the Brexit Vote. Therefore, the real wages experienced a sharp decline after the Brexit vote from 1.7 percent in June 2016 to -0.3 percent in August 2017 (Breinlich et al, 2017). Thus, the increase in inflation has caused a decrease in living standards of households.
2.3 Studies using the synthetic control method
To calculate the effects the holding back of firms and consumers has had on the economic output of the UK regions, the synthetic control method will be used instead of a simple comparison. A simple comparison could include a comparison between the economic output of the pre-Brexit vote period and the post-Brexit vote period for UK regions. Furthermore, a simple comparison could include a comparison between the UK regions and non-UK regions. To make the explanation of the models more understandable, the case of Tees Valley and Durham is used to explain the model.
8 period, the differences between Tees Valley and Durham and the “synthetic” Tees Valley and Durham shows the causal effects of the Brexit vote instead of just a variation in economic output between periods.
Furthermore, a simple comparison could be made between two regions. A simple comparison in this thesis then would for example include Tees Valley and Durham, which could be compared to the Dutch region Utrecht. However, to study the causal effects of the Brexit vote, the two regions should have similar economic growth characteristics. So that the only difference between Tees Valley and Durham and Utrecht would be the causal effects of the Brexit vote. However, it rarely happens that two regions are an exact match for the economic growth characteristics. Therefore, the calculated causal effect of the Brexit vote, could merely be a reflection of pre-Brexit vote differences in economic growth characteristics. Furthermore, even when Utrecht would be a good match for Tees Valley and Durham, with the synthetic control method the match could be improved.
As the synthetic control method is a method, which makes a weighted average of European regions which together best, resemble the economic determinants of a UK region such as Tees Valley and Durham in the pre-Brexit vote period. These European regions of the weighted average only includes regions, which did not experience the Brexit vote. The weighted average allows that for example for Tees Valley and Durham, the comparison region does not only have to include Utrecht. As the comparison region can include several other European regions which together best resemble the economic growth characteristic of Tees Valley and Durham and were not affected by the Brexit vote.
Currently, there are several studies which have successfully applied the synthetic control method (Abadie et al, 2003; Abadie et al, 2010; Abadie et al, 2014; Campos et al, 2014; Born et al, 2017; Born et al, 2018; Springford, 2018). For the UK, the synthetic control method has been applied by Born et al (2017, 2018) and Springford (2018) to calculate the causal effects of the Brexit vote on the national level. The causal effects calculated by Born et al (2018) show that since the Brexit vote the output loss was by the end of 2017 already 1.8 percent of the GDP which is close to 25 billion pounds.
By making conclusions based on the synthetic control method one should take into account assumptions that are made. First, the method assumes that only the actual unit is affected by the treatment (Brexit vote) and that there are no spill-over effects to the comparison units. In the case of the Brexit vote, this assumption includes that the Brexit vote has only had an impact on the regional economies of the UK and not on the regional economies of the other EU regions. This however, is a weak assumption as the Brexit vote created uncertainties and downward expectations for EU countries, thus the regions, about the future trading relationship with the UK.
Second, it is assumed that there is no variation in shocks in the post-treatment period. In our case the assumption is that after 23rd of June 2016, there was no variation in shocks between
9 economic shocks would indeed have affected both. Therefore, this is a valid assumption in our case. Furthermore, it is assumed that the intervention is affecting the treated unit from the moment of the treatment and not pre-treatment. In the case of the Brexit vote, this assumption
would include that the effects on the UK of the Brexit vote started from 23rd of June 2016
onwards (Born et al, 2017). This is in our case a valid assumption as the expectation for the Brexit referendum was that a majority would vote to stay in the EU and thus the shock of a Brexit came with the results of the referendum.
2.4 Determinants of regional economic growth
Within this thesis, the synthetic control method is used to calculate the causal effects of the Brexit vote. To calculate the causal effects of the Brexit vote on the regional economic growth, this method composes “synthetic” UK regions, which consist of several EU regions which are not affected by the Brexit vote. Which non-UK regions are included is based on several predictor variables which consist of either the economic determinants of regional economic growth and the economic output (first approach) or the yearly economic output of the pre-Brexit vote period (second approach). In the first approach, the economic determinants of regional economic growth are based on the study of Abadie et al (2003, 2014). Furthermore, in the first approach the economic output is based on the study of Abadie et al (2014). Moreover, the second approach in which the economic output is the predictor variable is also based on the study of Abadie et al (2003). In addition, the predictor variables included decide which non-UK regions are included in the “synthetic” non-UK regions. As based on the predictor variables the weighted average of regions is chosen which resembles the economic characteristic of the UK regions such as Tees Valley and Durham.
According to the traditional economic growth model of Solow (1956), economic growth can be explained through the total factor productivity which depends on physical and human capital. In other words, economic growth displays the increase or decrease in the output level of a region. A measure of the output level is the real GDP per capita. The real GDP per capita represent the value of all goods and services, which are produced per inhabitant of a region during a certain time period. As the real GDP per capita is a measure of economic growth, it is chosen as the outcome variable of this thesis. In addition, the real GDP per capita is included as predictor variable in the first synthetic control method. As the real GDP per capita measures the economic output per person, it is an indicator of how productive a region is. How higher the economic output per person, how higher the productivity is in a region.
10 The first determinant of economic growth, and thus is included as predictor variable is population density. Population density is an indicator for whether a region is part of an urban area or rural area. As indicated by Glaeser (2011) “urban density provides the clearest path from poverty to prosperity”. As urban areas such as London experience a clustering of firms and thereby people in the same location, which can create several external scale economies. In other words, a greater density thus greater agglomeration generates positive externalities, which are behind the increased productivity in cities (Fujita et al., 1999; Duranton and Puga, 2001; Glaeser, 2011). First, the clustering of people and firms in a location, increases the knowledge spillovers. As in cluster economies, there are a larger number of firms and consumer, the accumulation of knowledge is larger which enabling further learning between workers which in turn spurs innovations (Duranton and Puga, 2003). Second of all, both forward and backward linkages between companies, suppliers, and buyers make interaction between economic actors more efficient. Last of all, in cluster economies there is a pooled labour market, which allows for an easier matching between firms and workers (Frick & Rodriguez-Pose, 2018).
The second determinant of economic growth in a region is investment. Investment can have a direct effect on the productivity level of production through physical capital (Gal and Egeland, 2018). An investment into the physical capital can consist of an investment in equipment and machinery or the public infrastructure (Gal and Egeland, 2018). Furthermore, the third determinant of economic growth in a region is human capital. According to Pelinescu (2015), an increase in skilled human capital will increases the competitive advantages by means of innovation and diffusion technology. Human capital can be defined as the set of knowledge, skills, competencies, and abilities embodied in individuals which can be acquired through education, training, medical care, and migration (Benos, 2014; Schultz, 1961;Becker, 1964). In addition, human capital can consist of skilled human capital, but also of unskilled human capital. In this thesis, the unskilled human capital is defined as individuals (25-64 year) which have an educational attainment level of less than primary, primary or lower secondary education. While on the other hand, skilled human capital is defined as individuals (25-64 year) which have a tertiary education. When the level of human capital increases, the level of the labour productivity also increases (Pelinescu, 2015). In turn, the increase in labour productivity increases the output level and thereby increases the economic growth.
11 industry has become geographically more dispersed and is nowadays often off-shored to third world countries. The transition to services within the developed countries is still ongoing. Therefore, for some regions such as within the UK the industry sector is still responsible for a large part of the economic output. Therefore, industry is still included as a predictor variable in this thesis.
Furthermore, for services a differentiation is made based on public services and market services. As the public services can induce regional economic growth through several features: provision of legal and social framework, defence police services, judiciary, enforcement of property rights, correction for the inadequacies of an unrestrained marketplace, development of the economic infrastructure, regulation of externalities and transfer payments for maintaining social harmony and improving the productivity of the labour force (Grossman, 1988). Therefore, public services is a separate predictor variable from the market services. Thus, overall the predictor variables will include both the market- and public services.
2.5 Differences in economic characteristics of UK regions
The results of Born et al (2017, 2018) and Springford (2018) on national level are not representative for the regions within the UK as the regional disparity between UK regions is significant. This regional disparity between UK regions has several key features. First of all, within the UK there are long-standing, deep-seated inequalities in prosperity and employment between regions, characterised by the dominant position of the south-east of England (centred on London) over the rest of the UK regions (Bachtler, 2004). More generally, there is a ‘dualism’ in the economic development of the UK.
12
Figure 1: Real GDP per capita for UK regions (NUTS-2 level) in 2015 (£)
The large variations in the real GDP per capita is partly caused by the variation of human capital within UK regions. Currently, the younger workers and particularly the group in their twenties with high qualifications and skills tend to move towards higher productivity and more dynamic regions in search of employment opportunities (Faggian et al, 2006, 2007; Abreu et al, 2015). In other words, the population between the ages of 18 till 34 move largely to the urban areas including London (OECD, 2011, 2013). As a result, the UK displays two separate labour markets which largely co-exist with each other, but display very few interactions between them. There are urban areas, which have a large human capital stock caused by a labour market of highly skilled individuals such as in London. While on the other hand, there are the more geographically peripheral regions of the UK which are lacking far behind with the human capital stock. Both regions co-exist, however there is a lack of interacting between the London economy and the geographically peripheral regions of the UK (McCann, 2016). Moreover, within countries such as the UK, a transition is going on from industrial sectors (including manufacturing) to more information-intensive services sectors. These sectors benefit more largely of agglomeration in urban areas compared to the agriculture sector or the industry sector. Thereby, the information-intensive services sectors increase the productivity largely in cities compared to rural areas and thereby increasing the regional disparity.
13 2015 is included in the post-Brexit vote period. However, this period is not influenced by the Brexit vote as the Brexit vote occurred in 2016.
In figure 2 it is shown that after the Brexit vote occurred in 2016, the GPD per capita growth rates for the UK regions experienced changes. The most striking change occurred in Cumbria as before the Brexit vote it had one of the higher real GDP per capita growth rates of respectively 3.2 percent. While after the Brexit vote the real GDP per capita growth rate of Cumbria experienced one of the largest declines of respectively -1.7 percent which gives a total decline of 4.9 percent.
Figure 2: Annual average real GDP per capita growth (%)
The differences in the real GDP per capita growth rates between the pre- and post-Brexit vote period can be caused by the variation in uncertainties and downward expectations for UK regions. The UK regions with the highest shares of local GDP exposed to the Brexit experience likely a higher increase in uncertainty than regions with a lower share of local GDP exposed to the Brexit. According to Chen et al (2018), the UK regions with the highest local GPD exposed to the Brexit include Cumbria (16.3%), East Yorkshire and Northern Lincolnshire (15.8%), and Gloucestershire, Wiltshire, and North Somerset (15.6%). Furthermore, the regions with the highest shares of local labour income exposed to the Brexit include Cumbria (16.8), East Yorkshire and Northern Lincolnshire (15.1%), and Leicestershire, Rutland and Northamptonshire (14.8%) and Gloucestershire, Wiltshire and North Somerset (14.7). Therefore, following the shares of local GDP and local labour income exposed to the Brexit, the first hypothesis of this thesis is: the regions with the highest exposure to the Brexit are the
14 The between differences in local GDP exposed to the Brexit is caused by the variation in the sector composition of regions and by the geographical location of regions. According to McCann (2016) regions that specialize in manufacturing, agriculture, and extraction industries, which are more located in rural areas in the UK, tend to be more dependent on the EU market. In specific, the manufacturing industries have a high dependency on the EU market. This is caused by that the majority of the UK exports and imports to the EU consist of intermediate goods and services (Ijtsma et al, 2018). On average 9.3 percent of UK firm’s intermediate inputs in 2014 came from the EU, however for the sector manufacturing this percentage was significantly higher at 16.3 percent (Ijtsma et al, 2018).
The regions with the highest shares of manufacturing include Cumbria (24.1%), East Yorkshire and Northern Lincolnshire (23.9%), and Cheshire (22.8) and these regions are thus expected to be largely affected by the Brexit vote. This is in line with the results of Chen et al (2018) as the regions including Cumbria and East Yorkshire and Northern Lincolnshire have the highest local GDP exposed to the Brexit. While on the other hand, services and in specific financial services in which urban areas such as London are specialized in have a lower dependency on the EU market (McCann, 2016). Therefore, as the rural areas, which are already lagging behind compared to the urban areas, are expected to be larger affected by the Brexit vote, the expectation is that the inequality between regions will increase. Following this, the second hypothesis of this thesis is: the Brexit vote has accelerated the inequality between UK regions.
2.6 Summary
15
3. Methodology
3.1 Synthetic control method
Following Abadie et al (2003), this thesis uses the synthetic control method. The synthetic control method is used to find the best “synthetic” region which is comparable to the actual UK region and did not experience the intervention which in this case is the Brexit vote. In other words, the synthetic control method uses economic characteristics to find the weighted average of regions which together best resemble the economic characteristics of the UK region. After the “synthetic” UK region has been created, the actual UK region and the “synthetic” UK region are compared in the post-Brexit vote period such that the causal effects of the Brexit vote can be calculated. The causal effects can be calculated as the actual UK region is affected by the Brexit vote, while on the other hand the “synthetic” UK region is not affected by the Brexit vote. This procedure is repeated for all UK regions at NUTS-2 level.
The dataset thus consist of all the regions of the 27 EU countries, which can potentially be included into the “synthetic” UK region and then consist of one UK region such as Tees Valley and Durham. Then for the second analysis, the dataset again consist of all regions of the 27 EU countries and one UK region such as Cumbria. In total, there are 41 data analysis done, as on NUTS-2 level the UK consist of 41 regions. To make the explanation of the model more understandable, the case of Cumbria is used to explain the model.
Suppose we observe 𝐽𝐽 + 1 units (e.g. regions) over T > 1 period (year 2010-2017). J includes all the comparison regions outside the UK. While the plus one indicates the UK region such as Cumbria (𝑗𝑗 = 1). This UK region is the only region that is affected by the Brexit vote at period T0 +1 and T0 +2 (years 2016 and 2017). While the other non-UK regions, which are not affected
by the Brexit, vote thus the remaining regions: 𝑗𝑗 = 2 till 𝑗𝑗 = 𝐽𝐽 + 1 are all potential comparison regions and compose the “donor pool”. This “donor pool” consist of regions, which potentially are included in the “synthetic” UK region. The non-UK regions that are included in the “donor pool” are thought to be driven by the same structural process as Cumbria but they did not experience the Brexit vote shock.
Let W = (𝑤𝑤1, ..., 𝑤𝑤𝑗𝑗) denote a ( 𝐽𝐽 × 1) vector of non-negative weights which sum to one. The
scalar wj (j = 1, … . , J) represents the weight of the non-UK region j in the “synthetic” UK
region. Each different value of W will lead to a different synthetic control region. The weighted average allows that for example for Cumbria (𝑗𝑗 = 1) the comparison does not only have to include one non-UK region (𝑗𝑗 = 2). But can include several other non-UK regions (𝑗𝑗 = 2 till 𝑗𝑗 = 𝐽𝐽 + 1) which together best resemble the economic growth characteristic of Cumbria. But how do we decide which regions best resemble the economic growth characteristics of Cumbria?
Let 𝑋𝑋1 denote a ( 𝐾𝐾 × 1) vector which contains the predictor variables of the pre-Brexit vote
16 the values of these same predictor variables for the potential non-UK comparison regions (J) in the “donor pool”. In this thesis, two different approaches of the synthetic control method are used.
First, the predictor variables consist of the economic determinants of economic growth and the economic output which include investment ratio, sectoral composition, population density, real GDP per capita, and human capital indicator. While the second approach includes the predictor variable that consist of the actual economic output thus the yearly real GDP per capita (pre-Brexit vote period). For both methods the differences between the predictor variables pre-(pre-Brexit vote of the UK region and the potential non-UK comparison regions is given by the vector X1−
X0W. The aim is to minimize the size of the differences thus weight the non-UK comparison
regions such that the resulting values of X0W closely resemble X1. The W that is chosen to
minimize the difference is the vector W*. In other words, W* includes several non-UK comparison regions which together best resemble for example Cumbria in the pre-Brexit vote period. However, with the first approach of the synthetic control method, the predictor variables will have different importance on the outcome variable: the real GDP per capita. Therefore, let
V denote as the diagonal matrix with nonnegative components. The values of the diagonal
elements of V reflect the relative importance of the different variables. The values are decided by the code of the synthetic control method, which in this thesis is based on the code used by Abadie et al (2010).
Let YjtN denote as the value of the variable of interest (real GDP per capita) that would have
been observed if the UK region j is not affected by the the Brexit vote at period t. In this thesis period t is from 2010 till 2017. And YjtA if the UK region j is affected by the Brexit vote. The
data set is balanced which means that all non-UK and UK regions are observed at the same time periods. t =1,...,. T. Let T0 denote the number of pre-intervention periods, in this case 2010 till
2015.
The assumption is made that the intervention thus the Brexit vote has no effect on the outcome before the implementation period such that YjtN= YjtA is for 2010 till 2015 ( 𝑡𝑡 ∈ {1, . . . , 𝑇𝑇0}),
and all non-UK regions ( 𝑗𝑗 ∈ {2, … , 𝐽𝐽 + 1}). In equation 1 be the effect of the Brexit vote for region j at time t:
𝛼𝛼𝑗𝑗𝑗𝑗 = YjtA− YjtN (1)
If region j is exposed to the Brexit vote in 𝑡𝑡 = 𝑇𝑇0, … . . . , 𝑇𝑇, therefore:
YjtA= YjtN+ 𝛼𝛼𝑗𝑗𝑗𝑗 (2)
Let Djt be an indicator, which takes the value of 1 when region j is exposed to the Brexit vote
at time t, and value zero if region j is not exposed to the Brexit vote. The observed outcome for region j at time t is
YjtA= Y
17 As mentioned earlier only the first UK region (𝑗𝑗 = 1) is exposed to the Brexit vote and only after period 𝑇𝑇0 thus after 2015, therefore:
Djt= �1 if 𝑗𝑗 = 1 𝑎𝑎𝑎𝑎𝑎𝑎 𝑡𝑡 > 𝑇𝑇0 otherwise 0 (4)
The aim is to estimate the effects of the Brexit vote for 𝑡𝑡 > 𝑇𝑇0 thus calculating the causal effects
as shown in equation 5
𝛼𝛼𝑗𝑗= Y1tA− Y1tN= Y1t − Y1tN (5)
3.2 Inference studies
To ensure that the results of the synthetic control method show the causal effects of the Brexit vote on the UK regions a region placebo study will be done. This indicates that for several regions, which have a large weight in the “synthetic” UK regions and should not be affected by the Brexit vote, the causal effect of the Brexit vote is calculated. If the divergence in the post-Brexit vote period for the real GDP per capita for the UK regions is caused by the post-Brexit vote, then the divergence of the real GDP per capita of the region-specific synthetic controls in the post-Brexit vote period should be considerably smaller than in the case of the UK regions. To examine this the data of the post-Brexit vote is compared relative to the pre-Brexit vote fit. In other words, the post-and-pre-Brexit-vote ratio of the root mean squared error (RMSE) is used to compare the pre-Brexit vote fit with the post-Brexit vote fit. Before the post-and-pre-Brexit-vote ratio can be calculated the RMSE of the pre-Brexit post-and-pre-Brexit-vote period and the post-Brexit post-and-pre-Brexit-vote period have to be calculated. The equations for this are:
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑝𝑝𝑝𝑝𝑝𝑝= �∑ ( 𝑎𝑎𝑎𝑎𝑡𝑡𝑎𝑎𝑎𝑎𝑎𝑎 𝐺𝐺𝐺𝐺𝐺𝐺 𝑝𝑝𝑝𝑝𝑝𝑝 𝑎𝑎𝑎𝑎𝑝𝑝𝑐𝑐𝑡𝑡𝑎𝑎𝑖𝑖− 𝑠𝑠𝑦𝑦𝑎𝑎𝑡𝑡ℎ𝑝𝑝𝑡𝑡𝑐𝑐𝑎𝑎 𝐺𝐺𝐺𝐺𝐺𝐺 𝑝𝑝𝑝𝑝𝑝𝑝 𝑎𝑎𝑎𝑎𝑝𝑝𝑐𝑐𝑡𝑡𝑎𝑎𝑖𝑖) 𝑛𝑛 𝑖𝑖=1 2 𝑎𝑎𝑎𝑎𝑛𝑛𝑛𝑛𝑝𝑝𝑝𝑝 𝑜𝑜𝑜𝑜 𝑜𝑜𝑛𝑛𝑠𝑠𝑝𝑝𝑝𝑝𝑜𝑜𝑎𝑎𝑡𝑡𝑐𝑐𝑜𝑜𝑎𝑎𝑠𝑠 𝑐𝑐𝑎𝑎 𝑡𝑡ℎ𝑝𝑝 𝑝𝑝𝑝𝑝𝑝𝑝 − 𝐵𝐵𝑝𝑝𝑝𝑝𝐵𝐵𝑐𝑐𝑡𝑡 𝑜𝑜𝑜𝑜𝑡𝑡𝑝𝑝 𝑝𝑝𝑝𝑝𝑝𝑝𝑐𝑐𝑜𝑜𝑎𝑎 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑝𝑝𝑝𝑝𝑝𝑝𝑗𝑗 = �∑ ( 𝑎𝑎𝑎𝑎𝑡𝑡𝑎𝑎𝑎𝑎𝑎𝑎 𝐺𝐺𝐺𝐺𝐺𝐺 𝑝𝑝𝑝𝑝𝑝𝑝 𝑎𝑎𝑎𝑎𝑝𝑝𝑐𝑐𝑡𝑡𝑎𝑎𝑖𝑖− 𝑠𝑠𝑦𝑦𝑎𝑎𝑡𝑡ℎ𝑝𝑝𝑡𝑡𝑐𝑐𝑎𝑎 𝐺𝐺𝐺𝐺𝐺𝐺 𝑝𝑝𝑝𝑝𝑝𝑝 𝑎𝑎𝑎𝑎𝑝𝑝𝑐𝑐𝑡𝑡𝑎𝑎𝑖𝑖) 𝑛𝑛 𝑖𝑖=1 2 𝑎𝑎𝑎𝑎𝑛𝑛𝑛𝑛𝑝𝑝𝑝𝑝 𝑜𝑜𝑜𝑜 𝑜𝑜𝑛𝑛𝑠𝑠𝑝𝑝𝑝𝑝𝑜𝑜𝑎𝑎𝑡𝑡𝑐𝑐𝑜𝑜𝑎𝑎𝑠𝑠 𝑐𝑐𝑎𝑎 𝑡𝑡ℎ𝑝𝑝 𝑝𝑝𝑜𝑜𝑠𝑠𝑡𝑡 − 𝐵𝐵𝑝𝑝𝑝𝑝𝐵𝐵𝑐𝑐𝑡𝑡 𝑜𝑜𝑜𝑜𝑡𝑡𝑝𝑝 𝑝𝑝𝑝𝑝𝑝𝑝𝑐𝑐𝑜𝑜𝑎𝑎
As both RMSE for the pre-Brexit vote and the post-Brexit vote have been calculated, the ratio can be calculated by dividing the post-Brexit vote RMSE by the pre-Brexit vote RMSE.
3.3 Sensitivity analysis
19
4. Data
4.1 Regional data
As mentioned earlier the units used in this thesis are regions at the NUTS-2 level. The NUTS is a common classification used for regions and provides an uniform breakdown of territorial units. As it subdivides the economic territory of the member states into territorial units which allows for a comparison between regions of different member states. How larger the NUTS number, how smaller the population threshold. In our case, a larger NUTS level will give a more specific and narrow representation of the effects of the Brexit vote on the populations of different regions. Therefore, the regions at NUTS-2 level are used instead of NUTS-1 level. However, going more narrow and specific on NUTS-3 level was not an option for this thesis as the data for human capital (the population aged 25-64 by educational attainment level, sex), and the investment ratio (gross fixed capital formation for all NACE activities) is only available on NUTS-2 level.
Furthermore, the regions included on NUTS-2 level for this study are all regions from the 28 European member states. As it is expected that the other 27 European member states (besides the UK) have the most comparable economic characteristics compared to the UK regions. The data of the UK regions and most of the other European Union regions where gathered through Eurostat on NUTS-2 level. However, the data of France, the Netherlands, and Poland was missing in the Eurostat regional database. Therefore, those data has been gathered from the OECD regional database. The OECD regional database uses another classification system than the Eurostat regional database. For the Netherlands, the NUTS-2 level regions are comparable to the TL3 regions of the OECD regional database. While for Poland, the NUTS-2 level regions are comparable to the TL2 level regions of the OECD regional database. Furthermore, for France the TL2 regions of the OECD regional database are used which are comparable to the NUTS-1 regions. As the data comparable to the NUTS-2 classification was not available for France.
The UK regions and the other European regions have been observed for the period 2010 till 2017. The start of the period is 2010, as by then the effects of the financial crisis, which occurred in 2008, have disappeared. This thesis does not want to include the financial crisis effects, because regions responded different to the crisis. The pre-Brexit vote period would then not only show the long-term economic characteristic of a region, but also the effects of the financial crisis. Furthermore, the end of the period is 2017, as for the year 2017 the most recent data is available for the UK. Ideally, the year 2018 would have been included as this would give a longer period through which the effects of the Brexit vote could have been observed. However, this data is not available and therefore it could not be included in the dataset.
20 variables should show the economic characteristics of a region and thus based on these variables comparable regions for the “synthetic” regions could be found. However, a period of 6 years is a short observation period and thereby it does not account for shocks. These shocks effect the predictor variables, which will thus show the shocks instead of the long-term economic determinants which impact the economic growth of a region. If a longer pre-Brexit period is included, it is likely that the results will vary from this thesis, as the economic determinants will show variation. In other words, with a longer pre-Brexit observation period, the data will correct for short-term shocks and thereby show the long-term economic characteristics of a region more precise.
Furthermore, in the first- and second approach of the synthetic control method, the financial crisis shocks could affect the economic output of the regions. Which in that case would show the variation in effects of the financial crisis instead of the economic output that would normally occur in those regions. In other words, the regions, which are included in the “synthetic” UK regions, would be based on which regions are effected equally by the financial crisis instead of regions which normally would experience similar economic output.
The second period that needs emphasizing is the period 2016-2017 which gives 2 years of post-intervention data. The post-post-intervention of 2 years is relatively short for an observation period. Especially, realizing that only the year 2017 was completely affected by the Brexit vote. As for 2016, the first 6 months till 23rd of June 2016, the Brexit vote had not yet occurred. As the
expectation was that the Brexit vote would result in a majority voting for remain. The effects of the Brexit vote likely started from 23rd of June 2016 onwards. However, as the data of 2017
is the most recent available data, the post-Brexit observation period cannot be extended.
4.2 The outcome variable: real GDP per capita
The outcome variable for this thesis is the real GPD per capita in PPS. To calculate the dataset of the real GDP per capita several steps had to be taken. First, the data of the Gross domestic product (GDP) at current market prices by NUTS 2 regions is gathered through Eurostat regional database. The unit of this variable is purchasing power standard (PPS, EU27 from 2019), per inhabitant. First, the data has already accounted for the population and is thus the GDP per inhabitant also called the GDP per capita. Furthermore, the unit is in purchasing power standard (PPS) which means that the GDP per capita (in national currency) is divided by the purchasing power parities (PPP). The PPS is an artificial currency unit, and one PPS can buy the same amounts of goods and services in each region. In addition, as the GDP per capita is divided by the PPP it accounts for the inflation or deflation thus price differences. By removing the price differences from the GDP per capita, the comparison over the years can focus specifically on the output losses.
21 database has accounted for the population and is thus the GDP per inhabitant also called the GDP per capita. However, the GDP per capita has not yet accounted for the price differences. To account for the price differences, the data of the purchasing power parities for the GDP with the unit purchasing power parities (EU27_2019=1) has been gathered from Eurostat regional database. To calculate the real GDP per capita in PPS for France, the Netherlands and Poland the GDP per capita has been divided by the PPP. As the GDP per capita has been divided by the PPP, the unit of this variable becomes the purchasing power standard (PPS). The equation for calculating the real GDP per capita is:
Real GDP per capita = Purchasing power paritiesNominal GDP per capita
4.3 The predictor variables
The predictor variables in this thesis consist of the determinants of economic growth and the economic output. Which of the economic determinants and economic output are included in this thesis as predictor variables is based on the study of Abadie et al (2003, 2014). There are two synthetic control method approaches, which use two different types of predictor variables. In the first approach, the predictor variables consist of the economic determinants of a region and the economic output in the pre-Brexit vote period. While on the other hand in the second approach, the predictor variable consist of the yearly economic output in the pre-Brexit vote. In addition, in the first approach the average of the pre-Bexit vote period (2010-2015) is taken for all predictor variables. Based on these averages the decision is made on which non-UK regions are included in the “synthetic” UK regions. In the second approach, the yearly economic output is taken into account instead of the average. For both methods, the weighted average of non-UK regions, which best resembles the non-UK regions based on the predictor variables will be the “synthetic” UK regions.
22 variables is gathered, the average of the pre-Brexit vote period is taken and included as predictor variables.
For the predictor variables investment ratio and sectoral composition several extra steps have to be taken compared to real GDP per capita, population density, and human capital. First, for the investment ratio, the data of the gross fixed capital formation for all NACE activities by NUTS 2 regions in million units of national currency has been gathered through Eurostat regional database. Furthermore, the Gross domestic product (GDP) at current market prices by NUTS 2 regions in national currency has been gathered from Eurostat regional database. The investment ratio has been calculated by dividing the gross fixed capital formation by the Gross Domestic Product (GDP). Second, for the shares of the sectors, the data of the gross value added (at basic prices) for all NACE activities, industry, market services, and public services have been gathered for the NUTS-2 regions from Eurostat regional database. The data of the industry, market services, and public services have been divided by the total NACE activities to calculate the shares of the sectors in a region. Also for the investment ratio, and shares of sectors the averages are taken of the pre-Brexit vote period and are included as predictor variables.
Table 2: Predictor variables and unit of measure
Variables Unit
Real GDP per capita National currency per inhabitant adjusted for
PPP
Population density Inhabitants per square kilometre
Investment ratio Percentage on previous period
Sectors
Industry (except construction) Percentages (%)
Market services1 Percentages (%)
Public services 2 Percentages (%)
Human capital
Levels 0-2 education3 Percentages (%)
Levels 5-8 education4 Percentages (%)
Sources: own calculations based on Eurostat regional database, World Bank, and the OECD regional regional database
1 Wholesale and retail trade; transport; accommodation and food service activities; information and communication
2 Public administration and defence; compulsory social security; education; human health and social work activities; arts,
entertainment and recreation, repair of household goods and other services
3 Less than primary, primary and lower secondary education (age 25-64) 4 Tertiary education (age 25-64)
23
5. Results
The main aim of this thesis is to find the short-term economic impact of the Brexit vote on the regions within the UK. To calculate the causal effects of the Brexit vote the synthetic control method has been used. The results of the synthetic control method are discussed in this section.
5.1 The “synthetic” UK regions
The 41 “synthetic” UK regions created in this thesis are the comparison regions of the 41 actual UK regions. The “synthetic” UK regions serve as the counterfactual regional economies that have not been affected by the Brexit vote. In this thesis the regions, which are included in the “synthetic” UK region, is based on several predictor variables. In this thesis, two different approaches of the synthetic control method are included. The difference between the two approaches is the predictor variables. In the first approach of the synthetic control method, the averages of the economic determinants and the economic output are the predictor variables. While on the other hand, in the second approach of the synthetic control method the yearly economic output is the predictor variable.
First, the results of the first approach of the synthetic control method in which the average of the economic determinants and the economic output in the pre-Brexit vote period are the predictor variables: real GDP per capita, population density, investment ratio, sectoral shares, and human capital. In figure 3, the real GDP per capita for Inner London – East and the “synthetic” Inner London – East is shown. The “synthetic” Inner London - East consist of the regions Budapest (HU) (W=0.28), Noord Holland (NL) (W=0.18), and Région de Bruxelles-Capitale / Brussels Hoofdstedelijk Gewest (BE) (W=0.14). The non-UK regions which are included into the “synthetic Inner London – East seem plausible. As Budapest, Noord-Holland and Brussels are urban areas like Inner London – East and therefore have a high population density, and skilled human capital. Both are important determinants of economic growth in a region. It is shown in figure 4 that the “synthetic” Inner London – East closely follows the same path as the actual Inner London – East in the pre-Brexit vote period. Futhermore, in figure 4 the real GDP per capita for Gloucestershire, Wiltshire and Bristol/Bath area and the “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area is shown. The “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area consist of the regions Latvija (W=0.31), Prov. Brabant Wallon (BE) (W=0.22), and Pohjois- ja Itä-Suomi (FI) (W=0.21). It is shown that also for Gloucestershire, Wiltshire and Bristol/Bath area the “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area follows the same path in the pre-Brexit vote period.
24 is an exception was also visible in figure 1 as there it was shown that Inner London – West had by far the highest real GDP per capita.
Figure 3: The real GDP per capita for Inner London – East and the “synthetic” Inner London – East with the first approach of the synthetic control method
Figure 4: The real GDP per capita for Gloucestershire, Wiltshire and Bristol/Bath area and the “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area with the first approach of the synthetic control method
2010 2011 2012 2013 2014 2015 2016 2017 year 4 4.5 5 5.5
real GDP per capita
104
Actual Inner London - East Synthetic Inner London - East
2010 2011 2012 2013 2014 2015 2016 2017 year 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
real GDP per capita
104
25
Figure 5: The real GDP per capita for Inner London – West and the “synthetic” Inner London – West with the first approach of the synthetic control method
As the “synthetic” UK regions, with the exception of the “synthetic” Inner London – West, largely follow the same path as the actual UK regions in the pre-Brexit vote period there can be looked at the post-Brexit vote period for these UK regions. In other words, Inner London – West is excluded as there was not a suitable “synthetic” Inner London – West and therefore for this region the causal effects of the Brexit vote cannot be calculated. In figure 6, the output losses and gains as percentage of the real GDP per capita for the UK regions (without Inner London – West) are shown for the post-Brexit vote years: 2016 and 2017. In other words, in figure 6, the difference between the lines of the “synthetic” region and the actual region in figure 3-6 for the years 2016 (blue bar) and 2017 (orange bar) are shown for the 40 UK regions. It is shown that North Eastern Scotland experiences the highest output loss of 21.2 percent of the real GDP per capita (9063 pound per capita) in the post-Brexit vote period followed by Hampshire and Isle of Wight with 11.8 percent (3590 pound per capita) and Lincolnshire with 10.6 percent (2147 pound per capita). This is not in the line with the expectation as North Eastern Scotland has one of the smallest local GDP exposed to the Brexit (Chen et al, 2018). Furthermore, the figure shows that the region West Midlands experienced the highest increase in output gain of 8.2 percent of the real GDP per capita. This could be caused by that firms in the West Midlands stockpile ahead of the Brexit.
2010 2011 2012 2013 2014 2015 2016 2017 year 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
real GDP per capita
105
26
Figure 6: The causal effects of the Brexit vote on the real GDP per capita with the first approach of the synthetic control method (%)
27
Figure 7: the correlation between the causal effects of the Brexit vote and the local GDP exposed to the Brexit for the first approach of the synthetic control method
The second hypothesis in this thesis was the question whether the Brexit vote has accelerated the inequality between the UK regions. As Inner London – West is left out of the calculations, it is difficult to examine the second hypothesis which ask the question whether the Brexit vote has accelerated the inequality between UK regions. As Inner London – West is by far the most productive region in the UK and therefore it has relevancy to know how large this regions is affected by the Brexit vote. However, without the region Inner London – West, it is shown in figure 8 that there is a correlation between the real GDP per capita in the year 2015 (pre-Brexit vote) and the output changes that are experienced after the Brexit vote. The scatter plot shows that the most productive regions experience the largest output loss as percentage of the real GDP per capita. This contradicts the second hypothesis of this thesis that the Brexit vote has accelerated the inequality between the UK regions.
28 Second of all, the results of the second approach of the synthetic control method. In this approach of the synthetic control method, the predictor variable consist of the yearly real GDP per capita (pre-Brexit vote period). In figure 9, the results are shown for Inner London – East. In comparison with the first approach of the synthetic control method, the results shown that in the pre-Brexit vote period the fit between the actual Inner London – East and the “synthetic” Inner London has improved. The regions included in the “synthetic” Inner London – east are Luxembourg (W=0.23), Région de Bruxelles-Capitale / Brussels Hoofdstedelijk Gewest (BE) (W=0.20), and Valle d’Aosta/Vallée d’Aoste (IT) (W=0.17). Both the regions Luxembourg and Région de Bruxelles-Capitale / Brussels Hoofdstedelijk Gewest (BE) are comparable regions to Inner London – East, however Valle d’Aosta/Vallée d’Aoste is not. As Valle d’Aosta/Vallée d’Aoste is a rural area which lies in the Western Alps. The fit in the first synthetic control method approach is more plausible as the regions included there are all urban areas.
Furthermore, in figure 10 the results are shown for Gloucestershire, Wiltshire and Bristol/Bath area. The same results are visible for this region as for Inner London – East. As with the second approach of the synthetic control method the fit of the “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area to the actual Gloucestershire, Wiltshire and Bristol/Bath area in the
pre-Brexit vote period also improved. The regions included in the “synthetic” Gloucestershire,
Wiltshire and Bristol/Bath area are Prov. Brabant Wallon (BE) (W=0.24), Groningen (NL)
(W=0.08), and Valle d’Aosta/Vallée d’Aoste (W=0.04). ). Furthermore, the results for Inner London – West are shown in figure 11. The “synthetic” Inner London – West consist of the region Luxembourg (W=1). There is thus no variation between the first- and second approach of the synthetic control method. This indicates that for Inner London – West there is still no suitable “synthetic” Inner London – West found.
Figure 9: The real GDP per capita for Inner London – East and the “synthetic” Inner London – East with the second approach of the synthetic control method
2010 2011 2012 2013 2014 2015 2016 2017 year 3.5 4 4.5 5 5.5
real GDP per capita
104
29
Figure 10: The real GDP per capita for Gloucestershire, Wiltshire and Bristol/Bath area and the “synthetic” Gloucestershire, Wiltshire and Bristol/Bath area with the second approach of the synthetic control method
Figure 11: The real GDP per capita for Inner London – West and the “synthetic” Inner London – West with the second approach of the synthetic control method
In figure 12, the output losses and gains as percentage of the real GDP per capita calculated with the second approach of the synthetic control method are shown for the post-Brexit vote years: 2016 and 2017. In other words, in figure 12 the difference between the lines of the “synthetic” region and the actual region in figure 9-11 for the years 2016 (blue bar) and 2017 (orange bar) are shown for 40 UK regions (not Inner London – West). It is shown that Cumbria experiences the largest output loss of 18.9 percent of the real GDP per capita (4947 pound per capita) ) in the post-Brexit vote period followed by Lincolnshire with an output loss of 12.0 percent of the real GDP per capita (2432 pound per capita) and North Eastern Scotland with an output loss of 11.3 percent of the real GDP per capita (4816 pound per capita). This is in line with the expectation as Cumbria is the region with the highest local GDP exposed to the Brexit (Chen et al, 2018). 2010 2011 2012 2013 2014 2015 2016 2017 year 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
real GDP per capita
104
Actual Gloucestershire, Wiltshire and Bristol/Bath area Synthetic Gloucestershire, Wiltshire and Bristol/Bath area
2010 2011 2012 2013 2014 2015 2016 2017 year 0.5 1 1.5 2 2.5
real GDP per capita
105
30 There is a difference in the results of the second approach of the synthetic control method compared to the results of the first approach of the synthetic control method. As in the first approach, the region North Eastern Scotland experienced a larger output loss of 21.2 percent of the real GDP per capita which is almost 10 percent more than with the second approach. While Cumbria experienced a smaller output loss of 6.4 percent of the real GDP per capita which is almost 13 percent less than with the second approach. In addition, the West Midland experienced the largest gain in the output as percentage of the real GDP per capita in the first approach of the synthetic control method. While in the second approach Southern Scotland experiences the highest gain in output of 5.8 percent of the real GDP per capita, followed by Cheshire with a gain in output of 4.2 percent of the real GDP per capita.
Figure 12: The causal effects of the Brexit vote on the real GDP per capita with the second approach of the synthetic control method (%)
31
Figure 13: The correlation between the causal effects of the Brexit vote and the local GDP exposed to the Brexit with the second approach of the synthetic control method
Figure 14: The correlation between the output changes in the post-Brexit vote period and the real GDP per capita in the pre-Brexit period with the second approach of the synthetic control method
32
Figure 15: The correlation between the real GDP per capita losses calculated by the first- and second approach of the synthetic control method
5.2 Inference studies
33
Figure 15: The results of the region placebo experiments for the first approach of the synthetic control method (post-and-pre Brexit vote ratio of the RMSE)
5.3 Sensitivity analysis
The synthetic control method assumes that there are no spill-over effects of the Brexit vote onto the non-UK regions. However, several non-UK regions experience a high share of their local GDP exposed to the Brexit. According to Chen et al (2018), of the non-UK regions, the regions of Ireland experience the highest percentage of the local GDP exposed to the Brexit followed by the regions in Germany and the Netherlands. If these regions are included in the “donor” pool, the causal effects calculated by the synthetic control method are biased downwards. As the “synthetic” UK regions would also have experienced a slow-down in economic growth due to the Brexit vote, and thus experience a lower real GDP per capita than that they would have without the Brexit vote. Therefore, the gap between the actual UK regions and the “synthetic” UK regions would be smaller than it would have been if the “synthetic” UK regions would not have been affected by the Brexit vote.
In addition, in the pre-Brexit vote period Croatia became a member of the European Union and therefore experienced a large economic shock. The economic determinants in the pre-Brexit vote period should show the long-term economic characteristics of a region. However, as Croatia became a member in the pre-Brexit vote period the results show the short-term shocks of the EU membership for the regions of Croatia instead of the long-term economic characteristics. This is caused by the fact that in this thesis the pre-Brexit vote period is relatively short (only 6 years).
34 whether the spill-over effects of the Brexit vote onto non-UK regions and the EU membership of Croatia has influenced the results, a sensitivity analysis is done.
In figure 16, it is shown that without Ireland, Germany, the Netherlands, and Croatia in the “donor” pool, the results change for the first approach of the synthetic control method. The expectation was that the output losses as percentage of GDP would have increased, as the “synthetic” UK regions do now not experience a slow-down in economic growth caused by the Brexit vote. For the UK regions including North Eastern Scotland, Inner London – East and Cornwall and Isles of Sicily the output losses as percentage of the GDP have indeed increased. However, for most regions the output losses as percentage of the GDP have decreased as shown for several regions including Cumbria, and Lancashire. This results is surprising.
The surprising result could be caused by the fact that the fit in the pre-Brexit vote period between the “synthetic” UK region and the actual UK region has become worse. This is for example the case for Inner London – East as shown in figure 17. With the first approach of the synthetic control method the ”synthetic” Inner London – East consisted of Budapest (HU) (W=0.8), Noord-Holland (NL) (0.18), and Région de Bruxelles Capitale / Brussels Hoofdstedelijk Gewest (0.14). While in the sensitivity analysis the “synthetic” Inner London – East consists of Région de Bruxelles Capitale / Brussels Hoofdstedelijk Gewest (1). Without the Netherlands in the “donor pool”, this study would not have been able to find a suitable Inner London – East.
35
Figure 17: The real GDP per capita for Inner London – East and the “synthetic” Inner London – East with the sensitivity analysis for the first approach of the synthetic control method.
In addition, in figure 18 the results are shown for the second approach of the synthetic control method. The expectation was that the output losses as percentage of GDP would have increased, as the “synthetic” UK regions would not have experience a slow-down in economic growth caused by the Brexit vote. The results in figure 18 show that indeed for most UK regions, the output loss as percentage of the real GDP per capita has increased which is in line with the expectation. However, the results partly contradict the results of the sensitivity analysis of the first approach of the synthetic control method. As there it is shown that for example for Lancashire the output loss as percentage of the real GDP per capita changes in a gain in real GDP per capita. Furthermore, it is shown that the changes in output losses and gains is smaller for the second approach of the synthetic control method than for the first approach of the synthetic control method. This indicates that the results of the first approach of the synthetic control method are more sensitive thus influenced by the inclusion of the regions of Ireland, Germany, the Netherlands, and Croatia in the “donor” pool.
2010 2011 2012 2013 2014 2015 2016 2017 year 3.5 4 4.5 5 5.5 6 6.5
real GDP per capita
104
36
Figure 18: A comparison between the output losses as percentage of the GDP of the second approach of the synthetic control method and the sensitivity analysis
In the sensitivity analysis of the first approach of the synthetic control method, the fit in the pre-Brexit vote period between the “synthetic” UK regions and the actual UK regions has become worse. However, this is not the case for the sensitivity analysis of the second approach of the synthetic control method as shown in figure 19. As in figure 19 the results of Lancashire is shown and it is shown that in the pre-Brexit vote period the fit is still perfect.
Figure 19: The real GDP per capita for Lancashire and the “synthetic” Lancashire with the sensitivity analysis for the second approach of the synthetic control method.
5.4 Discussion
With the use of the synthetic control method there are several assumptions made. First, the assumption is made that the Brexit vote only affects the UK regions and has thus no spill-over effect to the comparison units. However, if the Brexit would occur not only the UK regions are exposed but also the EU regions. Therefore, the uncertainties and downward expectations
2010 2011 2012 2013 2014 2015 2016 2017 year 1.5 2 2.5 3 3.5
real GDP per capita
104
37 created by the Brexit vote will not only have affected the UK regions but will have had spill-over effects to the EU regions. However, it is expected that the Brexit spill-spill-over effects are small. Thus, that the economic effects of the Brexit vote for the UK regions is larger than the economic effects for the rest of the EU regions. Overall, the results are thus biased downwards as the “synthetic” UK regions also have likely experienced a small economic slow-down due to the Brexit vote.
On the other hand, the Brexit vote has diverted investments from the UK to the EU. As mentioned earlier according to Breinlich et al (2019) there were 181 Greenfield and M&A transactions from the UK to the EU that would not have occurred in the absence of Brexit. The increase in outward FDI from the UK to the EU has likely caused an increase in the real GDP per capita in the regions of the EU. In other words, as the “synthetic” UK regions have likely experienced an increase in real GDP per capita due to extra investments from the UK, the results are biased up-wards.
Furthermore, the main criticism in using the synthetic control method is that as long as the synthetic control cannot reproduce exactly the characteristics of the UK regions pre-Brexit vote, the causal effects calculated can be merely a result of differences in growth predictors between the UK regions and the “synthetic” UK regions. With the first approach of the synthetic control method, the economic determinants and the economic output of regions are included as predictor variables. It is shown in figure 5 that for the region Inner London – West this study was unable to find a suitable “synthetic” Inner London – West. This is caused by the fact that Inner London – West is the most productive region in the EU and thus there is not a region that can reproduce the path of Inner London – West. Furthermore, in figure 3 and 4 the results of the regions Inner London – East and Gloucestershire, Wiltshire and Bristol/Bath area are shown. For both regions, the “synthetic” region closely resembles the actual regions in the pre-Brexit vote period, however there could still be improvement as the fit is not exact.
