Impacts of Brexit referendum on European banks: evidence from Country-by-Country Reporting

• Název práce v češtině: Dopady referenda o brexitu na evropské banky: důkaz na základě Country-by-country Reporting
• Název v anglickém jazyce: Impacts of Brexit referendum on European banks: evidence from Country-by-Country Reporting
• Klíčová slova: Brexit, gravitační model, quasi-experimentální metody, banky
• Klíčová slova anglicky: Brexit, gravity model, quasi-experimental design, banks
• Akademický rok vypsání: 2020/2021
• Typ práce: diplomová práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: prof. Petr Janský, Ph.D.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 20.06.2021
• Datum zadání: 20.06.2021
• Datum a čas obhajoby: 15.06.2022 09:00
• Místo konání obhajoby: Opletalova - Opletalova 26, O109, AULA Michala Mejstříka č. 109
• Datum odevzdání elektronické podoby: 28.04.2022
• Datum proběhlé obhajoby: 15.06.2022
• Oponenti: prof. Ing. Karel Janda, Dr., Ph.D., M.A.

Seznam odborné literatury

Ricardo Mora & Iliana Reggio (2019) Alternative diff-in-diffs estimators with several pretreatment periods,
Econometric Reviews, 38:5, 465-486, DOI: 10.1080/07474938.2017.1348683

Baier, F.J., Welfens, P.J.J. The UK’s banking FDI flows and Total British FDI: a dynamic BREXIT analysis. Int
Econ Econ Policy 16, 193–213 (2019). https://doi.org/10.1007/s10368-018-00426-x

Aristeidis Samitas, Stathis Polyzos, Costas Siriopoulos, Brexit and financial stability: An agent-based simulation,
Economic Modelling, Volume 69, 2018, Pages 181-192, ISSN 0264-9993,
https://doi.org/10.1016/j.econmod.2017.09.019.

Dirk Schiereck, Florian Kiesel, Sascha Kolaric, Brexit: (Not) another Lehman moment for banks?, Finance
Research Letters, Volume 19, 2016, Pages 291-297, ISSN 1544-6123, https://doi.org/10.1016/j.frl.2016.09.003.

Santos Silva, João and Tenreyro, Silvana, (2006), The Log of Gravity, The Review of Economics and Statistics,
88, issue 4, p. 641-658.

Bouvatier Vincent, Capelle-Blancard Gunther and Delatte Anne-Laure. Banks in Tax
Havens: First Evidence based on Country-by-Country Reporting. European Commission
[online]. Discussion Paper 055, Brussels 2017 [cit. 11.06.2020]. Available at:
https://ec.europa.eu/info/sites/info/files/dp_055_en.pdf

Předběžná náplň práce v anglickém jazyce

Motivation:
On 23 June 2016, people of United Kingdom of Great Britain and Northern Ireland (from now on UK) decided, in
a referendum, to leave the European Union. This controversial act raised many questions and uncertain future for
both UK and EU citizens. In these days, financial crisis in 2009, global warming and Corona virus pandemic have
shown us that it is important for governments to team up or to collaborate to solve global issues. The European
Union is crucial agreement of nations of Europe to form a larger and more powerful unit based on peace and
cooperation. This unit can influence the international politics in much bigger scale than any separate member
state which also means that EU can address global issues considerably more effectively. The Brexit is a step
back in tackling global problems such as environmental problems, migration or regulating tech giants because the
EU loses a key player. However, there are losers and winners on both sides (UK and EU). For example, if tariffs
are to be imposed on imported goods, local producers benefit from it while importers from foreign countries loose
on the market. On the other hand, by the same logic, UK exporters are disadvantaged compared to local
producers in foreign countries. Therefore, Brexit plausibly may redistribute wealth not only in the UK itself but also
all over the European Union. The UK left the EU on 31 January 2020, nearly four years after the referendum. In
this master thesis, I will focus on only one type of players in the economy but the most important one – banks. I
will make use of yearly data on turnover in banks’ headquarters and subsidiaries abroad, Country-by-Country
Reporting (from now on CbCR), for several years before the referendum and all years after the referendum until
2020 (excluding 2020), the year of officially leaving EU. I will examine the banking sector and its reaction to
announcement of the UK leaving the EU both from the point of view of British banks as well as other EU banks.

There are several studies that focus on financial stability after Brexit. Samitas et al. (2018) used agent-based
modelling to study the effects of Brexit on the UK and the EU in terms of financial stability. They found out that
the consequences of Brexit will be worse for EU than for UK because the UK banking system is large enough to
recover from the predicted distress. Schiereck et al. (2016) analyze the reaction of banks’ shares and credit
default swap spreads to the announcement of Brexit and learn that spreads increased after the announcement
(however, not as much as in case of Lehman Brothers bankruptcy) and that share prices of EU banks fell more
compared to UK banks (notably, share prices fell more compared to Lehman Brothers bankruptcy). Baier &
Welfens (2019) find empirical evidence that, following the referendum, global FDI (foreign direct investment)
inflows increased while FDI inflows to the UK banking sector decreased.

The central research question of this master thesis is, how banks of UK and of EU react to Brexit announcement.
It is worth noting that banks of members of EU can benefit from a so called “passporting”. It means that
companies with residence in European Economic Area (EEA) can export their products into another member of
EEA without any direct authorisation by the importing member state. UK companies will no longer have this
privilege after leaving the EU. Although I will not be able to examine the effects of losing passporting directly due
to unavailability of data for years 2020 and 2021, the mere announcement of Brexit challenges the stability of
financial sector as suggested by literature (viz. above). Therefore, I will use CbCR dataset for European banks for
years 2013-2019, where years 2013-2015 represent the pre-announcement period and 2016-2019 is the postannouncement period. With this data, I will study the reaction of banks in terms of variables reported in CbCR.
The most important variable is turnover because it indicates level of activity of banks’ subsidiaries (if, for example,
we observe drop in turnover in a subsidiary between two years, it means that the bank decreased its activities in
the particular subsidiary). Additionally to turnover, In scope of the requirements, banks also report for each
country, where they have business, profit before tax, number of employees and tax paid. Some banks also report,
other types of taxes paid, subsidies received and some other informations. There are several cases of banks that
need to be considered. First, there are British banks that have business mainly in the UK (for example, in 2015,
Lloyds had more than 96% of total turnover reported in the UK), second, we have British banks that have a lot of
subsidiaries abroad either in Europe or in the rest of the world (e.g. in 2015, HSBC, the highest ranked European
bank in terms of assets, had more than 50 businesses in foreign countries including nearly all European
countries). Third, there are EU banks that either have subsidiaries in UK or do not. Banks of each type will
probably react in a different way to the Brexit announcement and it is unclear how these reactions will differ. UK
banks may concentrate more on activity in the UK and while they may decrease activity in the EU, they may
expand more to other non-EU or even non-European countries. Conversely, EU banks could have lower activity
in UK. I expect, however, that UK banks will record lower turnover in EU and vise versa simply because they
anticipate that in future, they will lose the passporting advantage. To empirically estimate the effect of Brexit announcement, I will use gravity model as well as difference-indifferences approach. Methodology is given in more detail below.

Hypotheses:
I will test the following hypotheses:
1. British banks report lower turnover in foreign countries as a consequence of the announcement of Brexit.
2. Banks with headquarters in the EU report lower turnover in the UK in the wake of the announcement.
3. British banks report higher turnover in the UK after the announcement of Brexit.

Methodology:
To test the stated hypotheses, I will make use of two econometric approaches.
First, hypotheses 1 and 2 are concerned with banks‘ activities abroad. To model exports and imports of any
business sector, it is convenient to use gravity equation which relates magnitude of trade, i.e. the dependent
variable (in our case turnover), to geographical distance between two economies and sizes of those economies
(e.g. GDP). Additionally, I include other standard independent variables that are usually used in gravity equation,
such as population, indication of common boarders, common language, colonial relationship, signature of
regional trade agreement or indication of global financial center. Furthermore, I will include a dummy variable
indicating tax haven because Bouvatier et al. (2017) and Moravec (2020) found empirical evidence that banks
report higher turnover in tax havens. This equation will serve as the baseline equation which should account for
all important factors to model the turnover. To estimate it, I will follow Silva & Tenreyro (2006) and use Poisson
Pseudo-Maximum Likelihood estimator (PPML). This method is convenient for my case because it estimates the
equation in multiplicative form, rather than in log-linear form. PPML is a better choice compared to log-linear OLS
because it does not drop observations with zero turnover and, unlike OLS, it is consistent also in case of
heteroscedasticity. As already suggested above, I will use CbCR data for years 2013-2019 for 45 largest
European banks in terms of assets. As a consequence, I must deal with panel form of the data. To do this, I will
use fixed effect approach which accounts for fixed heterogeneity present for each country. However, to make
sure, I will use Hausman test to see whether fixed effect or random effect is more suitable. To estimate the effect
of Brexit announcement, I will include a dummy variable indicating whether or not the announcement has already
been made. That is, the dummy variable is equal to one if, in corresponding or previous years, the announcement
has been made and zero otherwise. The percentage change of turnover related to Brexit can then be calculated
as exp(β)-1, where β is the estimated coefficient of the dummy variable. We can then isolate only British banks or
only EU banks and estimate the coefficient of the Brexit dummy variable for UK banks or for EU banks.

If we write the gravity equation in log-linear form, it will look as follows:
log(Turnover_{i,j,k,t}) = α_k + β_1log(GDP_{i,t}^{percapita}) + β_2log(Population_{i,t}) + β_3log(Distance_{i,j}) + β_4commonboarders_{i,j} + β_5commonLanguage_{i,j} + β_6Colony_{i,j} + β_7RTA_{i,j} + β_8GlobalFinCenter_i + β_9TaxHaven_i + γBrexit_t + u_{i,j,k,t}
Where the indices i,j,k,t stand for: country i is always the importing country and country j is always the exporter,
k stands for bank k and t is period t. Variables that are in logarithm are continuous ones and variables without
logarithm are dummy variables. We are interested in coefficient γ which is the effect of Brexit announcement.
Hypotheses 1 and 2 can then be tested by testing for the significance of the coefficient γ.

Second, hypothesis 3 is focused on local activity of banks (that is, activity in their home country). One could
simply estimate the effect of Brexit announcement by regressing turnover on the Brexit dummy variable while
controlling for important factors. However, this could potentially bring biased estimate because one cannot make
sure if the change in turnover is caused by the announcement or by a trend in the data or even by some other
factor (i.e. we do not know if we included all important factors and we may suffer from omitted variable bias). To
overcome this problem, I will use difference-in-differences (from now on DD) method which is a tool that is often
used to examine the effect of a policy or a treatment on one group while another group does not receive the
treatment and can therefore be used as the control group. In general, when we want to estimate effect of some
treatment, we need to calculate β = E(Y(t*) |D=1) – E(Y(t*)|D=0), where E(Y(t*)|D=1) is the expected outcome
after the treatment was done on the treatment group and E(Y(t*)|D=0) is the theoretical expected outcome in
case of no treatment on the treated group which cannot be observed. This is why we need to have the control
group with no treatment and we need to pose some assumptions about pre-treatment dynamics. Specifically, we
need the output variables of both groups (treated and control) to develop in a same way before the treatment and
the pre-treatment shocks to output cannot be dependent on the treatment. This is known as the common trends
assumption. The method performs poorly if the two control groups have different dynamics before the treatment.
Formally, the assumptions are well described in paper by Mora & Reggio (2017). To construct the model, I will
follow their proposed fully-flexible model to estimate the effect of Brexit announcement. This method is very
useful because it will allow me to estimate the effect without posing any prior assumptions before estimation and
the assumption can then be retrospectively tested.

In more detail, I will estimate the fully flexible model proposed by Mora & Reggio (2017) by OLS, where the
treated group will include all the British banks and the control group will include the rest of European banks.
Then, I will test the the common trend assumption using joint F-test with null hypothesis: H0 α(1) = 0, ..., α(t*-1) =
0, where α(1) = 0, ..., α(t*-1) are estimated coefficients of interaction terms t * T, where t is time dummy and T is
dummy indicating the treated group. If I cannot reject the null hypothesis it will mean, that the assumption of
common trend cannot be rejected. My estimated effect of the treatment (i.e. the announcement of Brexit) will then
be the α(t*) coefficient of interaction term t* * T, where t* is dummy variable indicating the period after treatment.
Third hypothesis can then be tested by t-test on α(t*).

Following Mora & Reggio (2017), the estimation DD equation will have the following form: Turnover_{i,k,t} = β_0 + β X + δBrexit_{i,t} + sum_{τ =2}^{T} (δ_t I{t, τ}) + sum_{τ =2}^{T} (α_τ I{t, τ} * {Brexit}_{i,t}) + u_{i,k,t}
X are all important faktors influencing turnover (those can be GDP per capita, population, financial center
dummy, tax haven dummy). Time variable dummy I{t, τ} is equl to 1 when time t equals τ. The effect of Brexit
announcement coefficient can be seen as coefficient α_T.

Expected Contribution:
By making this master thesis, I will contribute to four strands of literature. First, I will enlarge the recent literature
about Country-by-Country Reporting, such as Murphy (2015), Janský (2017) or Bouvatier et al. (2017). Second, I
will contribute to literature focused on gravity model - Tinbergen (1962), Anderson (1979), Anderson & van
Wincoop (2003), Silva & Tenreyro (2006). Third strand of literature are papers concerned with Brexit and its
financial consequences, such as Samitas et al. (2018) or Welfens (2019). And finally, I will build on literature
focused on DID estimation method, most notably Mora & Reggio (2017).

Outline:
The paper is going to be structured in the following way:
1. Introduction: In this section I will introduce the key points of the thesis, such as Brexit, CbCR data,
methodology and results that I will have came up with.
2. Literature: I will present the important literature for this paper.
3. Data & methodology: In this part, I introduce the CbCR data, present some stylized facts, I will present
the methodology of estimation and testing the effect of Brexit announcement on banking activity.
4. Results & discussion of results
5. Robustness: Here, I will describe potential alternative regressions and tests for robustness.
6. Conclusion