1
The Impact of the Brexit Referendum
on British Bank Lending Behavior: A
Synthetic Control Method Analysis
Master Thesis
July 2018
MSc Economics: Monetary Policy & Banking
Toon van Beek, 10540709
2
Statement of Originality
This document is written by Student Toon van Beek who declares to take full responsibility
for the contents of this document.
I declare that 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 have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of
completion of the work, not for the contents.
3
Master Thesis
The Impact of the Brexit Referendum on British Bank
Lending Behavior
Toon van Beek
July 2018
Abstract
This paper analyzes the impact of the Brexit referendum on British lending behavior. Quarterly aggregate panel data from 25 European Union member countries over the period 2006Q1–2017Q4 are retrieved. The synthetic control method is used to evaluate how British bank lending behavior would have developed in the absence of Brexit. By the end of 2017, the Brexit referendum led to a negative loans-to-assets gap of 10.47%. Country placebo studies are used to show that the results are significant. The placebo studies indicate that the gap is caused by the referendum and not randomly. The same analyses are done for domestic loans and Eurozone loans to investigate whether banks behave differently towards domestic and foreign counterparties. A significant 8.64% loss is found for domestic lending. For Eurozone lending, a positive loans-to-assets gap is observed after Brexit. However, the 3.85% gap at the end of 2017 is insignificant.
4
Table of Contents
1. Introduction ... 5
2. Literature Review ... 7
2.1 Impact of the Brexit Referendum on Financial Market and Economy ... 7
2.2 Effect of Uncertainty on Bank Lending ... 9
2.3 Banks’ Lending Behavior ... 10
3. Data and Methodology ... 12
3.1 Data ... 12 3.2 Descriptive Statistics ... 12 3.3 Methodology ... 14 4. Results... 16 4.1 Empirical Results ... 16 4.1.1. Total Loans ... 16 4.1.2. Domestic Loans ... 19 4.1.3. Eurozone Loans ... 21 4.2 Placebo Studies ... 24 4.2.1. Total Loans ... 24 4.2.2. Domestic Loans ... 26 4.2.3. Eurozone Loans ... 27 5. Conclusion ... 30 Bibliography ... 32 Appendix ... 36
5
1. Introduction
June 23, 2016, was a historic day for the United Kingdom (UK) and for the European Union (EU). On this day, the citizens from UK could vote during a referendum if they wanted to stay or leave the EU. The expectation was that the British population would vote to remain in the EU. Even though large cities, such as London, Liverpool and Edinburgh, voted to remain, a slight majority (51.9%) of the voting population voted in favor of Brexit (i.e., Britain’s exit from the EU). On June 24, 2016, it became clear that the UK would become the first country to break away from the 28-nation bloc (Sampson, 2017). The markets responded immediately to this unexpected result. The exchange rate of the British pound fell to its lowest level since 1985 (Yan et al., 2016). The London stock market faced large losses in share prices. The FTSE250, which is made up of mostly mid-sized British companies, experienced a fall of 7%. Indexes in London, New York, Frankfurt and Paris also lost around three percent. The referendum created much economic uncertainty and the Economic Policy Uncertainty (EPU) Index reached its highest level of the 21st century (Yan et al., 2016). After the announcement of Brexit, many researchers began to investigate the effects of Brexit on, among other things, British consumption and investments (Dhingra et al., 2017), gross domestic product (Born et al., 2017; Sampson, 2017) and immigration (Portes and Forte, 2016). Among the potential effects of Brexit are its impacts on lending behavior, but the literature on this matter remains rather scarce. To address this gap in the literature, this paper investigates the impact of the Brexit referendum to British banks’ lending behavior. The lending behavior of banks can be described as the loans-to-assets (LTAs) ratio.
Similar to this study, Berg et al. (2016) have investigated the effect of the Brexit referendum on the syndicated loan market.1 Using the difference-in-difference methodology, they found that the
issuance of British syndicated lending dropped by 25%. Belke et al. (2018) and Krause et al. (2016) have shown that the Brexit vote negatively affected the stability of the financial market. Britain’s decision to leave the EU has harmed its economy as well. The consequences are an output loss (Born et al., 2017), resulting in decreasing British per capita income (Sampson, 2017) and higher costs of living (Breinlich et al., 2017). Crafts (2016) and Campos et al. (2014) show that the UK significantly benefited from joining the EU. Furthermore, the literature shows that macroeconomic uncertainty leads to decreasing lending behavior among banks (Baum et al., 2002, 2009; Talavera et al., 2012; Quagliariello, 2009; Ibrahima and Shah, 2012).
The banking sector is crucial to the British economy. The balance sheet total for the British banks was more than 500% of Britain’s gross domestic product (GDP) in 2013, the second-highest European percentage, after that of Irish banks. The service sector is the UK’s strongest sector, and it accounts for 79% of GDP. The service sector of the second-largest European economy is driven by banking and insurance companies (UK Trade & Investment, 2016). This explains the importance of the research question.2
1 A syndicated loan is offered by multiple lenders to one borrower. This is usually the case when the loan is too large for one lender or when a specialized lender with expertise in a specific asset class is required (Berg et al., 2016).
2 A political reason that illustrates the contribution of this paper is the fact that in many other EU countries, Euro-skeptical parties have become more and more popular. Those parties have also tried to force EU exits for
6 To address the research question, the Synthetic Control Method (SCM) is used. The SCM is a methodology that combines components of the difference-in-difference and matching methodology.3 The SCM is applied in papers that are published in highly ranked journals and by
famous researchers, such as Moritz Schularick, Alberto Abadie and Giovanni Peri. The methodology is applied to many fields of research, such as economics, politics, criminology and health policy.
For this paper, data is collected from the ECB’s Statistical Data Warehouse, the International Monetary Fund (IMF) database and the Eurostat European Commission database. The data include 25 of the 28 EU member countries. The control group consists of 24 countries because the UK is the treated country. With the data of the 24 control countries, a counterfactual UK is created. The counterfactual UK models how British lending would have behaved in the absence of Brexit. The gap that arises after the Brexit referendum between the actual UK and the counterfactual UK is the result of the Brexit vote. To check the significance of the results, country placebo studies and root mean squared prediction error (RMSPE) ratios are generated and analyzed. It is possible to divide the loans into two different categories: domestic loans and Eurozone loans. The same analysis is used to assess the two groups separately to check whether any differences arise between domestic and Eurozone lending behavior.
The results obtained with the synthetic control method show that the LTAs ratio of the UK decreased by 10.5% compared to its doppelganger. The synthetic UK is formed by Germany, the Czech Republic, Finland, Cyprus, Austria and Spain. The country placebo studies and the RMSPE ratios demonstrate that the results are significant. British domestic lending behavior follows the same path as total lending behavior: it decreased significantly. However, British banks reacted differently to Eurozone loans after Brexit. A positive gap is observed for the Eurozone LTAs ratio. However, this gap is insignificant.
The paper is structured as follows. Chapter 2 gives an overview of the existing literature about the effect of Brexit to the financial market and real economy, the impact of macroeconomic uncertainty on bank lending and the determinants that describe banks’ lending behavior. Chapter 3 describes the data collection process, descriptive statistics and the methodology. Chapter 4 presents and explains the result while the Chapter 5 gives the conclusions. The paper ends with the bibliography and appendix.
their respective countries. Their leaders often make dubious economic claims. Therefore, this study may furnish information useful in informing public debate.
3 As in the difference-in-difference methodology, a control group is created, but in the SCM, the weights are assigned systematically. A long pre-intervention period is required to create a synthetic treatment country that follows the treatment country as closely as possible prior to the treatment. The synthetic country is also sometimes called the counterfactual country or the country’s doppelganger (Abadie et al., 2003, 2010, 2015).
7
2. Literature Review
This section provides a literature review. It starts with an overview of the existing literature about the effects of the Brexit referendum on the financial markets and the real economy. The EU referendum has created much uncertainty, which is why a description of the effects of macroeconomic uncertainty on banks’ lending behavior follows afterwards. The chapter closes with an explanation of the determinants that affect banks’ lending behavior. These determinants are then used as covariates in the synthetic control method analysis.
2.1 Impact of the Brexit Referendum on Financial Market and Economy
As already described in the introduction, Berg et al. (2016) analyzed how the Brexit referendum affected the British syndicated loan market. A difference-in-difference methodology is used to conclude that the volume and number of British syndicated loans issuances dropped by 25% compared to comparable syndicated loans markets. This decrease is mostly caused by a drop in lending to domestic firms and by domestic banks. Only a limited and insignificant decrease in lending by international banks and firms in the British syndicated loan market has been observed. This suggests that Brexit primarily affects British banks and borrowers and that the consequences for international activities in the UK are limited. Given that data was available only for the three months following the referendum for this study, the results should be interpreted with caution. Schiereck et al. (2016) investigated whether the Brexit referendum had consequences for the banking sector similar to those of the fall of Lehman Brothers. Their analyses, mostly focused on the stock and credit default swap (CDS) markets, showed that the referendum is not another Lehman moment for the global banking sector. However, for the banks in the EU member countries, the short-term decrease of the banks’ share prices grew after the Brexit announcement, but the increase in the CDS spread was lower following the Brexit referendum than the Lehman bankruptcy. The spread’s increase was mostly concentrated with EU banks, showing little contagion to non-EU banks. What may explain this lack of contagiousness is that financial markets are currently more resilient in handling shocks within the financial system. These shocks create political uncertainty, which generally leads to changing bank behavior (for further, see Section 2.2). Belke et al. (2018) focused on the impact of Brexit on the international financial market. They expect that the Brexit referendum would have a considerable impact on the financial markets of other countries since the financial markets are highly interlinked. By using both panel and single-country Seemingly Unrelated Regressions (SUR) estimation methods an analysis was made of the impact on long-term interest rates, stock returns and credit default swaps. Belke et al. conclude that the political uncertainty will continue to create instability in financial markets, and it is likely that it will also harm the real economy in the UK and EU. The countries that will suffer the most from Brexit, outside the UK, are Greece, Ireland, Italy Portugal and Spain (known as the GIIPS countries). Krause et al. (2016) have researched the effect of Brexit for the stability of the financial market, obtaining similar findings: the Brexit announcement resulted in a decrease of stability in the financial market. The high degree of political uncertainty led to decreasing bank indices. The negative consequences harm not only the UK but also the EU. A deprecation of the currency of the UK, the Sterling, indicates that the currency loses its attractiveness. This devaluing can happen because of a decline in attractiveness of the British financial market and reduced demand for its currency.8 The existing literature concerning the general (macro)economic consequences of the Brexit announcement to the UK is considered here. Born et al. (2017) used the SCM to analyze the impact of the referendum on the macroeconomy. In Born et al.’s study, the synthetic UK is formed for a large part by Japan, Hungary, United States and Canada. Together they contribute for more than 80% of the doppelganger’s weight, although small contributions also come from for New Zealand, Ireland, Luxembourg, Italy and Norway. The results indicate that households and firms lower spending in a response to the Brexit vote. This decrease in spending leads to an output loss of more than one percent, which can be explained by the high macroeconomic uncertainty. For the uncertainty analysis, the EPU index is used. The EPU index is based on newspaper articles that contain the words “uncertain” or “certain,” “economy” or “economic,” and one or more policy-relevant term. Another reason for the output loss is the decline of expectations of long-term income. The significance is tested with placebo studies. The results of the tests suggest that the gap is caused by the EU referendum.
Sampson (2017) has concluded that it may not be beneficial for the British economy to leave the EU. Sampson estimates that it will cost the citizens from 1–10% of UK per capita income in the long term. This decrease is caused by new trade and immigration barriers. The UK is not the only country that will suffer from Brexit. Other EU countries will also suffer economically, although their expected losses are much smaller than the UK’s. Only Ireland faces similar losses. Crafts (2016) has examined the economic effects of the UK joining the European Community (EC) in 1973, finding that it benefitted the country’s economy: GDP per capita and productivity growth increased. The reason for this increase was the fact that product market competition increased due to the fall of markets barriers. This led to a decrease of the domestic firms’ market power and an increase in productivity improvements investments. Campos et al. (2014) have conducted similar research, but they focused more on quantitative analyses, using the SCM. Similar to Crafts, they conclude that the UK enjoyed substantial economic benefits from joining the EU. Ten years after joining the EC, the country’s GDP was 8.6% higher than it would have been without EC membership. One cannot assume that the opposite effect would have happened if the country had never joined the EC, but these analyses suggest that the UK has substantially benefited from being an EU member. Ramiah et al. (2016) have investigated the immediate effect of the Brexit referendum on multiple sectors of the economy with the event study methodology, as measured by abnormal returns. The banking, travel and leisure sectors are negatively affected; the banking sector is hit especially hard. Breinlich et al. (2017) have suggested that the British inflation increased by 1.7% within a year of the Brexit referendum. The increasing inflation is constantly shared across the income distribution, but not across districts. London is less affected, while Wales, Northern Ireland and Scotland are hit the hardest. The vote in favor of Brexit has therefore increased the costs of living for British households.
The existing literature shows that Britain’s integration with the EU in 1973 has since been very good for the British economy. Gross domestic product and productivity increased, due the lowering of market barriers. After the citizens decided to leave the EU, some economists researched the consequences. As a result of Brexit, bank lending is expected to decrease in UK, and financial market stability is expected to decrease in both the UK and the EU. An output loss will ensue because of the Brexit vote, resulting in lower GDP per capita. New trade and immigration barriers are the main reason for this gap. The banking sector will suffer the most, but the households will also feel the consequences through higher costs of living, resulting in lower spending.
9
2.2 Effect of Uncertainty on Bank Lending
Section 2.1 has already clarified that a high degree of political and macroeconomic uncertainty negatively effects output and financial market stability (Belke et al., 2018; Born et al., 2017). This section focuses on the effects of uncertainty on bank lending. The most used uncertainty index is the EPU index, which Born et al. (2017) also use in their paper. Figure 1 plots the EPU index. The data is collected from the website of Economic Policy Uncertainty. Clearly, after Brexit both Europe and the rest of the world reached their highest uncertainty scores in the 21st century. Thus, the time prior to and after the Brexit referendum are considered times of high (political and macroeconomic) uncertainty.
Figure 1: Economic Policy Uncertainty (EPU) index
Notes: This figure plots the EPU index for the EU and for the whole world in the 21st century. The data is collected from the website of the Economic Policy Uncertainty: http://www.policyuncertainty.com/brexit.html.
Baum et al. (2002) have considered how macroeconomic uncertainty affects bank lending. They use the loans-to-assets (LTAs) ratio as an indicator for bank lending and an increase in the variability of industrial production as an indicator of macroeconomic uncertainty. Annual and quarterly US-bank level data are used for the analysis. The results show that higher uncertainty leads to a lower LTAs ratio. The LTAs ratios for larger, more profitable, highly-ranked banks are more constant over time. Talavera et al. (2012) have studied the relationship between macroeconomic uncertainty and bank lending in the Ukraine for 2003Q2-2008Q2. Their findings show that increasing uncertainty is followed by decreasing lending ratios. One explanation could be that managers have higher risk aversion during times of uncertainty. However, the bank behavior of Ukrainian banks on uncertain changes is not uniform, and it depends on bank-specific characteristics. In contrast to paper Baum et al. (2002), Talavera et al. (2012) conclude that less-profitable banks are less likely to be affected by changes in the macroeconomic climate.
Dinc (2005) has examined at the lending behavior of government-owned banks during election years in emerging countries and India. Elections are a good indicator of times with high
10 political uncertainty. Dinc finds that lending increased significantly among government-owned banks, as compared to private banks. Baum et al. (2009) provide evidence that macroeconomic uncertainty affects the LTAs ratio. The results show that LTAs ratios change from 6-10% if macroeconomic uncertainty doubles. Quagliariello (2009) has researched how macroeconomic uncertainty plays a role in Italian banks’ lending decisions. Macroeconomic uncertainty also accounts for some sorts of bank-specific uncertainty. The results show that uncertainty negatively affects LTAs ratios. This finding is in line with previous findings. Ibrahima and Shah (2012) have conducted a similar analysis on the Malaysian economy. They report similar results; higher uncertainty leads to less lending.
Based on the literature, it can be assumed that macroeconomic uncertainty has a substantial influence on banks’ lending behavior. Macroeconomic uncertainty is negatively correlated with lending, since higher uncertainty leads to decreasing LTAs ratios. Therefore, it can be expected that Brexit will have a negative impact on British banks’ behavior, with decreasing LTAs ratios as a result.
2.3 Banks’ Lending Behavior
Section 2.2 has shown that macroeconomic uncertainty influences banks’ lending behavior. This variable is not added to the covariates since it is difficult and debatable how it should be measured. Gambacorta and Mistrulli (2004) have checked whether bank lending is affected by bank capital. The results indicate that the effect of capital on lending is bigger for less-capitalized banks. Bank capital also affects the way banks respond to GDP shocks. The less-capitalized banks are more procyclical than the well-capitalized banks, so they are more influenced by GDP shocks. Bank lending decreases by 20% for high-risk banks as a result of the implementation of a mandatory capital ratio of 8%. The capital ratios are part of a regulatory framework. This also indicates that bank capital has a significant influence on banks’ lending behavior. Real GDP growth and inflation are added to the regression as control variables, since they influence the loan demand. Xiong (2013) confirms this relationship in his discussion paper written at the behest of the Bank of Finland. Large banks are less likely to react to monetary policy changes by adjusting their behavior. Small banks and banks that are not well capitalized respond more to tightening monetary policy, while large banks are more sensitive to monetary policy easing. Real GDP growth and the Consumer Price Index (CPI) should also be considered as variables that affect banks’ lending behavior. They both affect credit demand, which lead to changes in bank behavior (Xiong, 2013). Olokoyo (2011) expresses banks’ lending behavior as commercial banks’ loans and advances (LOA). The Nigerian LOA is affected by the independent variables: interest rates, volume deposits, capital requirements, liquidity ratio, investment portfolio, GDP and foreign exchange. The paper rejects the hypothesis that there is no relation between the dependent variables and the explanatory independent variables. Berger and Bouwman (2009) demonstrate that the size of the banks, expressed as the total assets, is of interest in changes in bank behavior. For this reason, a logarithm of the total assets is added to the covariates.
Jiménez et al. (2012) find that the issuance of loans is affected by macroeconomic conditions such as the business cycle. As such, they use real GDP growth as a control variable. Jeanneau and Micu (2002) conclude that real GDP growth and real interest rate have a procyclical influence on international bank lending. This influence is caused mainly by European and Japanese banks, since they tend to behave procyclical, while American banks have a countercyclical lending pattern. Allen et al. (2012) conclude that mandatory capital and liquidity requirements lead to
11 changing behavior among banks. Chiuri et al. (2006) report that the introduction of capital requirements negatively impacts loan supply in emerging countries. This effect is even stronger if these requirements are implemented after a currency or financial crisis. However, the implementation of capital requirements can have both negative and positive consequences. It will reduce ill-advised lending, but it may result in lower credit availability, worsening the availability of liquidity and reducing economic growth. Aiyar et al. (2014) used UK bank data to evaluate the effect of capital requirements on banks’ lending behavior. In line with the previous literature, they found that the implementation of capital requirements has a negative impact on UK bank lending. Alhassan et al. (2013) investigated the lending behavior of the 25 largest banks in Ghana, 2005-2010. They find negative relationship between LTAs ratio and liquidity ratio. Kakes and Sturm (2002) also confirm that the liquidity ratio influences banks’ lending behavior. For smaller banks, the amount of liquid assets is more important, since they are more sensitive to monetary policy shocks. Although, large banks have in general a lower liquidity ratio, they can easier respond to monetary policy changes. Amidu (2006) reveals that in the Ghanaian banking sector, lending behavior is significantly affected by the central bank’s prime rate and inflation. Both have a negative impact on bank lending. The bank size and liquidity have a significant influence on the bank’s ability to react to money demand.
The literature shows that large, well-capitalized banks are less likely react to GDP and monetary policy shocks. Capital is influential in banks’ decision making. With the introduction of mandatory capital requirements, banks’ lending behavior changes. The same is true of (mandatory) liquidity ratio. Both are main drivers in banks’ lending behavior and are negatively correlated with LTAs ratio. Inflation and real GDP growth rates affect loan demand and, therefore, banks’ lending behavior. The volume of the loans or assets and the interest rates also influences bank lending.
12
3. Data and Methodology
This section explains the data and methodology. First, it demonstrates how the data is collected and how some variables are calculated. Second, it presents the statistics of the variables, and it gives a short explanation for these values. Finally, the section ends with a description of the methodology, the SCM.
3.1 Data
Quarterly aggregated panel data is collected from multiple databases for 2006Q1–2017Q4 for 25 European countries. Most of the data is gathered from the ECB’s Statistical Data Warehouse. The total outstanding loans for Monetary Financial Institutions (MFIs), excluding the European System of Central Banks, is retrieved from this database. The aggregated balance sheet of the MFIs sector is the sum of the harmonized balance sheets of all the MFIs. The total loans are a combination of the domestic loans and the loans to (other) Eurozone countries.4 The LTAs ratio is generated by dividing
total loans by total assets. The same has been done for the domestic loans and Eurozone loans. The capital ratio is calculated in a similar manner, namely capital and reserves divided by total assets. A logarithm of the total loans is taken to create a new variable that accounts for the size of the countries’ banking sector. This has been done because all the other variables are ratios. The size of the banking sector does not influence those ratios. In the absence of a variable that accounts for the size of the banking sector, countries such as Malta and Germany would have been treated exactly the same. Berger and Bouwman (2009) demonstrate that the size of the banks is of interest for changes in bank behavior. The interest rate and inflation are also gathered from the Statistical Data Warehouse. For the interest rate indicator, the long-term interest rate for convergence purposes is used, and for inflation, the Harmonised Index of Consumer Prices (HICP). The IMF database has been used to collect real GDP growth rates for the UK and the control group countries. The GDP growth rates are recorded on a quarterly basis and indicate the year-on-year percent change. The liquidity ratio is constructed by dividing the liquid assets by the total deposits. The ratio is constructed by dividing cash and securities by total deposits. This proxy is also used by Aspachs, Nier and Tiesset (2005) and Freedman and Click (2006). The data for calculating the liquidity ratio is gathered from the Eurostat European Commission database.
3.2 Descriptive Statistics
Table 1 summarizes the variables over the complete time span, 2006Q1-2017Q4.
4 No distinction has been made concerning the maturity of the loans. The counterparty sector is unspecified. This means that it includes MFIs and non-MFIs. Examples of non-MFIs are governments, households, and insurance corporations and pension funds (ICPFs).
13 Table 1: Averages variables
Variable European countries UK European without UK
Loans-to-assets (LTAs) ratio 60.22% 59.07% 60.27%
Domestic LTAs ratio 53.09% 44.99% 53.43%
Eurozone LTAs ratio 7.15% 14.07% 6.87%
Capital ratio 9.29% 7.79% 9.35%
Real GDP growth rate 1.81% 1.34% 1.83%
Liquidity ratio 83.11% 53.85% 84.37%
Inflation 1.93% 2.38% 1.91%
Interest ratio 3.74% 2.78% 3.78%
Logarithm total loans 12.5576 15.5305 12.4337
Notes: This table provides the averages of all the variables. The second column gives the averages of the 25 EU member countries. The third column illustrates the means of UK, and the fourth column shows the averages of all the European countries except the UK. In the appendix, a more extensive summary (Table A1) is given in the appendix, which takes into account standard deviation, number of observations, minimum value and maximum value. The data is gathered from ECB’s Statistical Data Warehouse, the IMF database and the Eurostat European Commission database. Since the same data is used for all the tables, the data sources are henceforth not repeated in the table notes.
Noteworthy is that the LTAs ratio for the UK is quite similar to the European countries’ average. Figure 2 plots the LTAs ratio over time for the UK and for all the control countries. The vertical red line represents the time of the referendum. The British LTAs ratio is quite stable over time and a little bit below the European average. Figure 2 shows that there are a few outliers, but most of the ratios are comparable to the UK’s.
Figure 2: Total loans-to-assets (LTAs) ratio for UK and all the control countries
Notes: This figure plots the total LTAs ratio for all the investigated countries. The black line represents the UK, and the 24 control countries are given by the grey lines.The vertical red line represents the time of the Brexit referendum. The data is gathered from ECB’s Statistical Data Warehouse, the IMF database and the Eurostat European Commission database. Since all the figures are created with the same data, the data sources will no longer be repeated. The same applies to the vertical red line, because it represents the time of the Brexit referendum in all the figures.
A noticeable difference is observed between the Eurozone LTAs ratios. The UK’s ratio is almost twice as high as the average of all European countries: 14.07% compared to 7.15%. The UK
14 thus lends relatively more loans to the Eurozone. This explains why the domestic LTAs ratio is lower for the UK compared to rest of Europe, since the total ratios were comparable. The logarithm of the total loans indicates the size of the market. The UK has the highest total number of loans in Europe. For this reason, the logarithm of UK is above the European average, and the highest observation, 15.6967, comes from the UK. The interest rate in the UK is, on average, almost 1% lower than the European rate. To explain the differences in interest rate between UK and the rest of Europe, a graph has been plotted that illustrates the interest rates over time. In Figure 3, it is clearly visible that the European average interest rate is higher than the UK’s, especially for 2008–2014. The UK has a higher inflation rate because of the lower interest rate. This relationship is in line with macroeconomic theory: Lower interest rates increase incentives to borrow more money. These incentives result in higher consumption and more investment by households and companies, causing the economy to grow and inflation to increase (Gylfason, 1981). However, as compared to Europe, the UK has a lower real GDP growth rate. The European average capital ratio and liquidity ratio are higher than the UK’s.
Figure 3: Interest rate UK and Europe
Notes: The figure plots the interest rates over the investigated time span for the UK and Europe. The UK is represented by the red line and the European average is represented by the blue line. The data is obtained from ECB’s Statistical Data Warehouse.
3.3 Methodology
The Synthetic Control Method (SCM) is applied to provide an answer to the research question. Abadie and Gardeazabal (2003) introduced this method, and it is further utilized and developed by Abadie et al. (2010) and Abadie et al. (2015). The SCM is a useful method to evaluate how bank behavior of British banks would have developed in the absence of the Brexit referendum. The synthetic UK follows (almost) the same path as the real UK prior the Brexit referendum and can therefore be considered as a doppelganger. The control group is formed by 24 member countries
15 from the EU.5 A synthetic UK is created by assigning weights 𝜔 = (𝜔
1, … , 𝜔24) to each country. The
weights are determined by minimizing the distance between the LTAs of the UK and the synthetic UK prior to the referendum. The gap that arises after the treatment period (June 24, 2016) between the real and the synthetic UK is the result of the Brexit referendum.
Before the weights 𝜔 = (𝜔1, … , 𝜔24) could be determined, the weights of the covariates
𝑣 = (𝑣1, … , 𝑣7) should be calculated. The covariates are a set of pre-treatment outcomes and other
time-invariant observed predictors. In this research, the covariates will be the pre-referendum LTAs ratios, interest rates, liquidity ratios, capital ratios, inflation, real GDP growth rates and a logarithm of the total loans. The pre-referendum period contains 41 quarters, namely 2006Q1-2016Q1. The weights 𝑣𝑚 are selected by the cross-validation approach. Intuitively, the cross-validation technique
selects the weights 𝑣 that minimize out-of-sample prediction errors. In most cases, the pre-treatment outcome will have the greatest weight, since higher weights are allocated to units with large predictive power on the outcome variable of interest. For this research, that will probably result in a high weight for 𝑣𝑙𝑜𝑎𝑛−𝑡𝑜−𝑎𝑠𝑠𝑒𝑡𝑠 (Abadie et al., 2014). When the weights 𝑣 are selected, the
weights 𝜔 are calculated, and the effect of Brexit can be plotted in a graph. The difference between the real UK and the synthetic UK is attributed to the Brexit vote.
Finally, to assess the significance of the estimates, a series of placebo studies will be conducted by iteratively applying the SCM to all the control group countries. In this way, all the control group countries will have a gap between the pre-Brexit and post-Brexit outcome. The significance of the result will decrease if the gap of the control countries is similar to the UK’s. Without any formal significance test, it is difficult to assess whether this effect is large enough to be significant. While a formal test for this does not exist, however, it is possible to approximate such a test by using RMSPE ratios (Abadie et al., 2003, 2010, 2015). The RMSPE ratio is defined as the post-treatment RMSPE divided by the pre-post-treatment RMSPE:
𝐑𝐌𝐒𝐏𝐄𝐫𝐚𝐭𝐢𝐨 =
𝐑𝐌𝐒𝐏𝐄𝐚𝐟𝐭𝐞𝐫 𝐁𝐫𝐞𝐱𝐢𝐭 𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐝𝐮𝐦
𝐑𝐌𝐒𝐏𝐄𝐛𝐞𝐟𝐨𝐫𝐞 𝐁𝐫𝐞𝐱𝐢𝐭 𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐝𝐮𝐦 ( 1 )
Comparing the RMSPE ratio of UK with all the countries in the control group will show whether the effect of the Brexit referendum for the UK is large.
The total LTAs can be divided in two different categories: domestic loans and Eurozone loans. The same analysis will be used for the two groups separately to check whether there are any differences between domestic and Eurozone lending behavior. Since the European loans of Romania were not available for 2006, the country has been withdrawn from the control group for the Eurozone loans analysis.
5 These countries are Austria, Belgium, Bulgaria, Cyprus, the Czech Republic, Germany, Denmark,
Spain, Finland, France, Greece, Hungary, Ireland, Italy, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Romania, Sweden, Slovenia and Slovakia.
16
4. Results
In this chapter, the results are presented and explained. First, the results of total lending are discussed. Then domestic lending and Eurozone lending are analyzed separately. The chapter ends with country placebo studies in which the significance of the results is checked.
4.1 Empirical Results
This section answers the research question, namely whether the Brexit referendum has impacted the lending behavior of British banks. A doppelganger of the UK is created, and the resulting post-Brexit gap provides clarity about the impact of post-Brexit on the British banking sector. The same analysis is conducted for domestic and European loans to examine whether there are different patterns in bank behavior between domestic and foreign lending.
4.1.1. Total Loans
Figure 4 plots the actual and the counterfactual LTAs ratio for UK. The LTAs ratio for the actual UK is given by the blue line, and the red line represents the doppelganger.
Figure 4: Actual and counterfactual loans-to-assets (LTAs) ratio for UK
Notes: The figure plots the actual LTAs ratio for the UK and the counterfactual LTAs ratio for the UK. The actual UK is given by the blue line, and the red line represents the doppelganger. The gap that arises due to the differences between the real UK and synthetic UK is plotted in Figure 5 below.
Figure 4 clearly shows that the Brexit announcement had a negative impact on the LTAs ratio of British banks. In the absence of Brexit, the lending behavior of British banks would have been quite similar to what is represented by the red line. The gap between the actual and the counterfactual UK is plotted in Figure 5. The gap reaches its peak at the end of 2017Q3, 10.96%. One quarter later, at the end of 2017, the gap decreases somewhat. Without Brexit, the British LTAs ratio would have been close to 71.12% at the end of 2017. The actual LTAs rate was 60.65%, resulting in a
17 LTAs gap of 10.47%. The decline in lending is in line with the expectations based on the literature review.
Figure 5: Gap between actual and synthetic UK
The figures demonstrate that the synthetic UK is a good fit for the actual UK. Table 2 confirms this representation. In this table, the means of the covariates for the actual UK, synthetic UK and all the control countries are given. It shows that the LTAs ratio for the real UK and its doppelganger is almost identical during the pre-Brexit period. The capital ratios, the interest rates and the real GDP growth rate are also very similar to each other. It is not very surprising that those covariates fit very well, because these variables have the highest covariate weight. The aggregated weights of these covariates are 98.63%. The weights of the covariates are given in the appendix, in Table A2.
Table 2: Covariates means for the pre-Brexit referendum period
Variables
UK Average of 24 European control
countries Real Synthetic
Loans-to-assets (LTAs) ratio 58.91% 58.95% 60.05%
Capital ratio 7.79% 7.76% 9.07%
Real GDP growth 1.23% 1.27% 1.60%
Liquidity ratio 54.00% 61.17% 83.73%
Inflation 2.50% 1.76% 2.06%
Interest rate 3.10% 3.14% 4.17%
Logarithm total loans 15.536 13.268 12.423
Notes: This table provides the values of all the variables for the period prior to Brexit. The second column gives the values of the real UK and the third column of the synthetic UK. The fourth column provides the average of the 24 control countries in the period before the Brexit referendum.
From Table A3 in the Appendix it follows that the synthetic UK is formed by Germany (40.1%), Czech Republic (21.1%), Finland (18%), Cyprus (14.3%), Austria (5.5%) and Spain (1%).
18 Together they form the best pre-Brexit fit for the UK’s banks’ lending behavior. Germany has the highest weight of all the countries, namely 40.1%. That is not very surprising since Germany is after UK the country with largest amount of loans. That explains why (of the control group) Germany’s logarithm of total loans is the closest to the UK’s. Table 3 presents the averages of both UK and Germany prior the Brexit referendum. It shows that the LTAs ratio of both countries is very similar. The LTAs ratio has the highest weight of all the covariates. This explains Germany’s high weight in the control group.
Table 3: Comparison between Germany and UK
Variable UK Germany
Loans-to-assets (LTAs) ratio 58.94% 58.63%
Domestic LTAs ratio 45.06% 51.43%
Eurozone LTAs ratio 13.88% 7.20%
Capital ratio 7.80% 5.14%
Real GDP growth rate 1.26% 1.49%
Liquidity ratio 54.07% 52.16%
Inflation 2.45% 1.48%
Interest ratio 3.06% 2.45%
Logarithm total loans 15.535 15.329
Notes: The table provides the average of all the covariates over the investigated period for the UK and Germany. That Cyprus is part of synthetic UK might come as surprise to many people. However, there is a strong relationship between the banking sectors of Cyprus and the UK. Cyprus was a British colony from 1878–1960 and is part of the Commonwealth of Nations. During the late 19th century, British or part-British banks established a dominant position in the banking sector all over the world. The biggest British banks all established a branch in Cyprus. In the 20th century, there was a global
development of rising local banks. This resulted in a decreasing market share for British banks. However, for Cyprus the rise and development of local financial institutions and their successful competition with the established British banks took place in a British colony, while for the most countries this was in the context of a nationalist government. In this way, the British banking sector secured some part of its dominant position. The part-British Imperial Ottoman Bank introduced modern banking to Cyprus and helped to construct a stable financial system. Barclays financed the establishment of an own Maltese central bank (Phylaktis, 1988). When the current financial situation is analyzed, many similarities are found between the UK and Cyprus. The banking sector of Cyprus is large compared to its economy. In 2010, the total assets-to-GDP ratio was 896%, while the EU average was only 357% and while the Eurozone’s was 334%. When the overseas operations of domestically owned banks are excluded, the ratio is still more than 700% for 2010. Cyprus is not the only country with a large banking sector, however; several European countries have a comparable or even bigger banking sector. The UK is one of these countries (Stephanou, 2011). The strong commercial and financial relationship between the UK and Cyprus is of crucial importance to Cyprus’ economy. The UK is also very important for Cyprus on a political level, since it holds two sovereign military bases on the island and since it is one of the three countries responsible for guaranteeing Cyprus’s independence (Oliver, 2017).
Czech Republic forms the synthetic UK for 21.1%. A few important similar characteristics explain the correlation between the UK and the Czech Republic prior to the Brexit referendum. Both
19 countries have their own currency and their own superior national central bank. In contrast to the Eurozone countries, their central banks are responsible for managing monetary policy, while for the Eurozone countries, it is the ECB’s responsibility. They both behave within the EU area but outside the Eurozone. Cernohorska (2015) has compared the banking sectors of the Czech Republic with those of the UK. Both countries use a common two-tier banking system. This system includes a central bank and commercial banking networks. The central banks use the same monetary policy regime. Both banking sectors meet the Basel 2 requirements and have very similar capital ratios. Their ratio of bank capital to assets is much higher than the minimums set in the Basel 2 accords.
The City of London, the primary central business district, is very important for the Finnish financial system. As reported by the Bank of International Statistics (BIS) 11.4% of foreign bank claims on Finnish banks comes from UK. Furthermore, the Finnish corporate governance model is quite similar to the UK’s (Vihriälä & Wyplosz, 2016), which explains why 18% of the synthetic UK is formed by Finland.
4.1.2. Domestic Loans
Table A1 in the Appendix shows the descriptive statistics. The main results are already discussed in Chapter 3, but for this part, it is useful to emphasize that the domestic LTAs ratios form the largest part of the total LTAs. In the UK, 76% of the total loans are domestic, while for the investigated European countries this figure is even higher, at 88%. Based on these numbers, it can be expected that domestic lending behavior follows the same path as the total lending behavior. This section examines whether this supposition holds true.
Figure 6 shows the domestic LTAs ratio for real UK and the counterfactual UK. It is very similar to the total lending behavior, which is in line with the expectations.
Figure 6: Actual and counterfactual domestic loans-to-assets (LTAs) ratio for UK
Notes:The figure plots the actual domestic LTAs ratio for the UK and the counterfactual domestic LTAs ratio. The actual UK is given by the blue line, and the red line represents the synthetic UK. The gap that arises due to the differences between the real UK and the counterfactual UK is plotted in Figure 7 below.
20 Figure 6 demonstrates that the Brexit announcement has had a negative impact on domestic lending. The observed pattern is similar to the total loans’. The gap between the actual UK and its doppelganger is plotted in Figure 7. The largest gap is observed in the last period, 2017Q4, namely 8.64%. Figure A1 in the Appendix plots the domestic LTAs. The figure shows that the UK’s ratio slightly decreased after Brexit, while the ratio for most countries increased. This explains how this gap arose. Furthermore, it is visible that the UK’s domestic LTAs ratio is below the European average.
Figure 7: Gap between actual and synthetic UK for domestic lending
The gap in the period prior to the Brexit referendum is small, illustrating that the counterfactual UK is a good fit for the actual UK. Table 4 confirms this fidelity. The domestic LTAs are almost identical for the pre-Brexit period, since the domestic LTAs ratio has a covariate weight of 93.90%. The capital ratio and real GDP growth are also similar. The logarithm of the total loans is used as a covariate. If a logarithm of the domestic loans were used, the results would be similar. The same countries would have formed the UK’s doppelganger, with only a few small changes in the countries’ weighting.
Table 4: Covariates means for the pre-Brexit referendum period for domestic lending behavior
Variables
UK
Covariates Weights Real Synthetic
Domestic loans-to-assets (LTAs) ratio 45.05% 45.07% 93.90%
Capital ratio 7.79% 7.66% 2.51%
Real GDP growth 1.23% 1.25% 2.72%
Liquidity ratio 54.00% 64.21% 0.30%
Inflation 2.50% 1.75% 0.11%
Interest rate 3.10% 3.48% 0.40%
Logarithm total loans 15.536 13.617 0.06%
Notes: This table provides the values of all the variables for the period prior Brexit for domestic lending. The second column gives the values of the real UK, and the third column of the synthetic UK. The fourth column provides the weights of the covariates in the domestic lending behavior analysis.
21 Table A4 in the Appendix shows how the counterfactual UK is formed. Again, Germany has the highest weight, at 31.9%. The synthetic UK is further formed by Belgium (24.6%), Spain (23%), Cyprus (12.2%) and Malta (8.3%). Germany’s and Cyprus’ weights are already explained in the previous paragraph. An explanation for Belgium is that it got a real GDP growth rate similar to that of the UK in the period prior the referendum: 1.28% versus 1.34%. Of all the EU countries, Spain invests most in UK’s banking sector. After the US, Spain the most foreign investments in UK’s financial services. It has such high investment because two of the largest Spanish banks, Sabadell and Santander, have many affiliates in the UK. Their British affiliates account for around one quarter of the banks’ profits and assets in 2015Q1. The UK also invests heavily in Spain, since it is the fifth-largest investor in the country (Garicano, 2016). Malta is, just as Cyprus is, part of the Commonwealth of Nations and a former British colony. It became independent in 1964. Malta and the UK still have a strong relationship, especially in the financial markets. Eurostat data shows that Maltese services exports to the UK amount to 13% of the Maltese GDP. A big part of these exports are financial services flows. Maltese banks have many outstanding loans to British citizens. Their exposure to the UK is equivalent to 7.4% of their assets (Rapa, 2017). This explains the Spanish and Maltese weights in the counterfactual UK.
4.1.3. Eurozone Loans
The same analysis is done for loans to the Eurozone. If a country is part of the Eurozone, their domestic loans are removed from the Eurozone loans. In this way, only the “foreign” loans are counted. Romania is excluded from the control countries because their Eurozone loans for 2006 are unavailable. The Eurozone LTAs for the UK and the remaining control countries are plotted in Figure A2 in the Appendix. The UK has the highest ratio, after Luxembourg. Of all the other countries, only Belgium’s Eurozone LTAs ratio is comparable to the British ratio. While the Eurozone LTAs ratio stayed constant for most countries after the Brexit announcement, the British ratio increased. Based on these observations, a positive gap between the UK and the counterfactual UK is expected. Figure 8 plots the actual and counterfactual Eurozone LTAs ratios. As expected, the actual UK is higher than the synthetic UK.
22 Figure 8: Actual and counterfactual Eurozone loans-to-assets (LTAs) ratio for UK
Notes: The figure plots the actual Eurozone LTAs ratio for the UK and for the synthetic UK. The blue represents the actual UK, and the counterfactual UK is given by the red line. The gap that arises due to the differences between the real UK and the counterfactual UK is plotted in Figure 9 below.
The Eurozone gap between the real UK and its doppelganger is plotted in Figure 9. Contrary to the previous results, a positive gap is found. The gap reaches its peak at 2017Q4: 3.85%. This is lower than the total loans gaps and domestic loans gaps, which is explained by the absolute Eurozone ratio being much lower than the domestic ratio (i.e., 13.9% compared to 45,1%). Not only does the UK’s Eurozone LTAs ratio increase, but also the total euro loans. The loans to the Eurozone increased from €1,325,367 to €1,415,871 million. The €90,500 million increase occurred between May 31, 2016, and December 31, 2017. At the same time, the British total assets stayed quite constant. The Brexit announcement has had a positive impact on lending to the Eurozone, while domestic lending has decreased. This trend is in line with the Berg et al.’s (2017) findings. They found that lending to domestic firms and by domestic firms decreased in the syndicated loan market. Only a limited and insignificant decrease in lending by international banks and firms in the British syndicated loan market was observed.
23 Figure 9: Gap between actual and synthetic UK for Eurozone lending
According to Table 5, the Eurozone LTAs ratio is exactly the same for the real and counterfactual UK. However, the figure above shows a definite difference in the ratio, resulting in gaps. The positive and negative gaps are equally large, however, resulting in an overall gap of zero. In contrast to the total and domestic loans, a logarithm of the Eurozone loans is taken instead of the total loans. As already shown, the Eurozone loans form a small part of the total loans. This can lead to big differences between countries’ Eurozone and total loans logarithm. Since the Eurozone lending behavior is investigated, it is better to take a logarithm of the Eurozone loans. Interest rate and capital ratio have large covariate weights. In line with the methodology theory, this results in small differences for Eurozone LTAs, capital ratio and interest rate between real and synthetic UK.
Table 5: Covariates means for the pre-Brexit referendum period for Eurozone lending behavior
Variables
UK
Covariates Weights Real Synthetic
Eurozone loans-to-assets (LTAs) ratio 13.86% 13.86% 19.59%
Capital ratio 7.79% 7.75% 55.53%
Real GDP growth 1.23% 1.66% 3.74%
Liquidity ratio 54.00% 108.32% 0.10%
Inflation 2.50% 2.19% 5.71%
Interest rate 3.10% 3.25% 13.17%
Logarithm European loans 14.088 11.475 2.16%
Notes: This table provides the values of covariates used in the Eurozone lending analysis. The second column gives the values of the real UK and the third column of the synthetic UK. The fourth column provides the weights of the Eurozone covariates.
The synthetic UK is largely formed of Austria (58.3%) and Luxembourg (29.5%). There are also small contributions from Germany (2.9%) and Hungary (9.3%). Austria has such a large share because it is, after Ireland, the country with the closest capital ratio to the UK’s. The capital ratio forms 55.5% of the covariate weights, so it is highly influential. Ireland is not included in the counterfactual UK, since its real GDP growth rate, liquidity ratio, inflation and interest rate are
24 completely different than the UK’s. The real GDP growth rate of Austria is a little bit higher, but it is very similar to the UK’s: 1.40% versus 1.34%.
The EU is very important for Luxembourg, since its foreign policy is mostly focused on multilateral international cooperation. The government is in most cases in favor of more free trade, due to the small size of the country. Despite its size, Luxemburg has a large banking sector. The banking sector is of crucial importance for its economy. Only five of the 143 credit institution in Luxembourg are domestic; the others have their roots in 30 different countries. This led to the highest internationalization rate in Europe. It results in a banking sector that is, compared to other countries, focused on international lending rather than domestic lending (Oliver, 2017; European Banking Federation, 2017). Figure A2 in the Appendix already showed that the UK has, after Luxembourg, the highest Euro LTAs ratio. Therefore, both countries are quite focused on the Eurozone and not only to their domestic market. This shared focus explains why Luxembourg has a 29.5% weight in the “European Synthetic” UK.
4.2 Placebo Studies
The placebo studies test whether the results are significant. They check whether the gap is indeed caused by the Brexit referendum. In this context, two different ways of conducting placebo studies are available: time placebo studies and country placebo studies (Born et al., 2017). For this research, country placebo studies are used. With country placebo studies, a doppelganger for every control country is created. The gaps that arise are plotted in figures. As long as control countries, where Brexit did not take place, do not have similar gaps, it can be concluded that the results for the UK are significant. It is difficult to determine significance by looking at a figure, so RMSPE ratios are calculated afterwards.
4.2.1. Total Loans
Figures 4 and 5 indicate that the SCM can provide a good fit for the British banks’ lending behavior prior to the Brexit referendum. This results in a RMSPE ratio of 1.14%, which is low. Of all the investigated countries, only Austria and the Czech Republic have a lower RMSPE ratio. A low pre-Brexit RMSPE ratio illustrates a good fit for the country’s doppelganger. Figure A3 in the Appendix plots the total LTAs gaps for UK and all the control countries. For some countries, it is very hard to create a doppelganger with similar banks’ lending behavior. This is visible from the poor fit between the real and the synthetic countries, resulting in large gaps prior to Brexit. The state with the poorest fit is Malta, with a RMSPE ratio of 8.635%. This poor fit is not very surprising, since it has the lowest LTAs ratio and lowest logarithm of total loans. Therefore, there is no combination of countries in the control group that can reproduce the lending behavior of Maltese banks.
If the UK’s doppelganger had failed to fit the country’s banking behavior prior to the Brexit referendum, much of the after-Brexit gap would be caused by lack of fit instead of the impact of the referendum. Crucial to placebo runs is that the gap prior the treatment be as small as possible. Countries with a poor fit are unable to provide information about the significance of the results, since these results are affected by a lack of fit. The Abadie (2010) paper is followed by excluding the countries of the control group that have a pre-Brexit RMSPE twice as large as that of the treatment country, UK. The following countries drop out: Bulgaria, Cyprus, Spain, Finland, Greece, Hungary, Ireland, Lithuania, Malta, Portugal, Romania and Slovakia. In Figure 10, the remaining countries are plotted (i.e., the countries with a well-fitted doppelganger). The gap of the UK is given by the black line, and the remaining control countries are given by the grey lines. Based on this figure, it is fair to
25 conclude that the results are significant, since the UK gap stands out. It suggests that the gap is caused by the referendum and not randomly. The country with the large positive gap is the Czech Republic.
Figure 10: Gaps for total loans-to-assets (LTAs) for UK and well-fitted control countries
Notes: This figure provides the gaps of the total LTAs for the UK and the well-fitted countries. The countries with a pre-root-mean-squared prediction error (RMSPE) of at least 2.28% are removed. The following control countries remain: Austria, Belgium, the Czech Republic, Germany, Denmark, France, Italy, Luxembourg, the Netherlands, Poland, Sweden and Slovenia. The control countries are given by the grey lines, and the UK is represented by the black line.
In Figure 11, the RMSPE ratios of all the countries are given with a good pre-treatment fit (i.e. the countries that have a pre-RMSPE of less than 2.28%). Figure A4 in the Appendix gives the RMSPE ratio of all 25 countries. The RMSPE ratio is calculated by dividing the post-Brexit RMSPE by the pre-Brexit RMSPE. If there is a big gap between those periods, it will result in a large ratio. From Figure 11, it follows that the UK has the highest ratio, after the Czech Republic. However, the Czech Republic’s gap is positive, while the UK’s gap is negative. Figure 10 already showed, and Figure 11 confirms, that no country comes close to UK’s gap. The British doppelganger gap stands out. It can thus be concluded that the results with regard to British banks’ lending behavior are significant.
26 Figure 11: RMSPE ratios for UK and well-fitted control countries for total lending
Notes: This figure provides the root mean squared prediction error (RMSPE) ratios of the UK and the control countries with a pre-RMSPE less than 2.28%. The following control countries remain: Austria, Belgium, Czech Republic, Germany, Denmark, France, Italy, Luxembourg, Netherlands, Poland, Sweden and Slovenia. The RMSPE ratio is calculated by dividing the post-referendum RMSPE with the pre-referendum RMSPE.
4.2.2. Domestic Loans
The results of domestic lending and total lending are quite similar. This section investigates whether the placebo studies are also similar. The pre-Brexit RMSPE ratio for the UK’s domestic lending is 1.40%. This ratio exceeds the total loan’s pre-Brexit RMSPE, which indicates that the total lending of the counterfactual UK is a better fit than the domestic lending counterfactual UK. This is confirmed by the fact that six countries have a lower pre-Brexit RMSPE score, while for total lending only one country had a lower ratio. The gaps between the UK and all the control countries are plotted in Figure A5 (see appendix). Again, there are control countries with large gaps in the period prior to the referendum. When countries with a pre-Brexit RMSPE ratio of at least 2.81% are excluded, countries with a bad-fit doppelganger drop out: Bulgaria, Cyprus, Finland, Greece, Hungary, Ireland, Lithuania, Luxembourg, Poland, Portugal, Romania, Slovakia and Slovenia. The 12 remaining countries are sketched in Figure 12. The results are similar to placebo studies of the total loans. Again, there is a significant positive gap for the Czech Republic, and the UK gap stands out. Based on the figure, the results of the UK for domestic lending behavior are significant. The RMSPE ratios should also be checked, however. The RMSPE ratios of the countries with a well-fitted doppelganger are given in Figure 13. It is similar to the figure of the RMSPE ratios of total lending (Figure 11). Only the Czech Republic has a higher RMSPE ratio, but they the Czech Republic has a positive gap. The UK’s RMSPE ratio is 1.5 times as large as Austria’s, which is the country that comes closest to UK’s gap. It can be concluded that, just like the total lending, the Brexit announcement has had a significant negative effect on the domestic LTAs ratio.
0,0 2,0 4,0 6,0 8,0 10,0 CZ GB AT NL SI BE LU IT SE DE DK PL FR
27 Figure 12: Gaps for domestic loans-to-assets (LTAs) for the UK and well-fitted control countries
Notes: This figure provides the gaps between the domestic LTAs for the UK and the well-fitted countries. The countries with a pre-root-mean-squared prediction error (RMSPE) of at least 2.80% are removed. The following 11 control countries remain: Austria, Belgium, the Czech Republic, Germany, Denmark, Spain, France, Italy, Malta, the Netherlands and Sweden. The RMSPE ratios are calculated by dividing the post-Brexit RMSPE with the pre-Brexit RMSPE. The RMSPE ratios for the UK and the 11 remaining control countries are given in Figure 13.
Figure 13: RMSPE ratios for the UK and well-fitted control countries for domestic lending
4.2.3. Eurozone Loans
The results of British Eurozone lending behavior differed from the total and domestic lending behavior. Therefore, it is expected that the placebo studies result will differ from the previous ones. The pre-Brexit RMSPE ratio for UK’s Eurozone lending is 0.825%. Compared to total lending and domestic lending, it looks small, but it should be considered that the UK’s Eurozone LTAs ratio is 13.86%. This means that the RMSPE ratio as a percentage of the Eurozone loans rate is 5.95%. This is higher than domestic lending (3.11%) and total lending (1.94%). This indicates that the total lending synthetic UK is the best fit, while the counterfactual UK for Eurozone lending is the worst fit. The fact that nine control countries have a smaller pre-Brexit RMSPE ratio for Eurozone lending confirms this.
0,0 2,0 4,0 6,0 8,0 CZ GB AT MT SE IT DE NL BE ES FR DK
28 Figure A6 in the Appendix plots all the Eurozone LTAs ratios. The country with the large positive gap is Luxembourg. This is unsurprising, since Luxembourg has by far the largest Eurozone LTAs ratio, and finding a combination of countries in the control group that can reproduce Luxembourg’s Eurozone lending behavior is therefore impossible. Again, the countries with a pre-Brexit RSMPE ratio twice as large as the UK are dropped. In this case, the countries with a pre-pre-Brexit RMSPE ratio of at least 1.65% are Belgium, Cyprus, Greece, Luxembourg and Portugal. Since Romania was not part of the Eurozone control group, 18 countries remain in the control group. Their gaps and the UK’s gap are plotted in the figure below. The figure shows that the UK’s gap no longer stands out. Hungary and Sweden have similar gaps. This would indicate the results are not significant. These gaps could arise due to a weaker pre-Brexit fit. To check this RMSPE ratios are plotted in Figure 15. Hungary’s RMSPE ratio is higher. This indicates that the results of the UK regarding Eurozone lending behavior are not significant. Spain has a higher RMSPE ratio as well, but that is mainly caused by the country’s small pre-Brexit RMSPE. It can thus be concluded that the lending behavior of British banks towards Eurozone loans increased after Brexit, but these results are not significant. This is again in line with Berg et al.’s (2017) findings, since British lending behavior towards international lending is insignificant.
Figure 14: Gaps for Eurozone loans-to-assets (LTAs) for UK and well-fitted control countries
Notes: This figure provides the gaps of the Eurozone LTAs for the UK and the well-fitted countries. The countries with a pre-root-mean-squared prediction error (RMSPE) of at least 1.65% are removed. The following 18 control countries remain: Belgium, Bulgaria, the Czech Republic, Germany, Denmark, Spain, Finland, France, Hungary, Ireland, Italy, Lithuania, Malta, the Netherlands, Poland, Sweden, Slovenia and Slovakia. The RMSPE ratios are calculated by dividing the post-Brexit RMSPE with the pre-Brexit RMSPE. The RMSPE ratios for the UK and the 18 remaining control countries are given in Figure 15.
29 Figure 15: RMSPE ratios for the UK and well-fitted control countries for Eurozone lending
0,0 2,0 4,0 6,0 8,0 HU ES GB IT NL LT SI IE SE AT FR CZ BG PL DK SK FI MT DE
