Project1 Number: 813261

Project Acronym: ABC-EU-XVA

Project title: Valuation Adjustments for Improved Risk Management

Periodic Technical Report

Part B

Period covered by the report: from [01/11/2018] to [01/11/2020]

Periodic report: 1st

1
The term ‘project’ used in this template equates to an ‘action’ in certain other Horizon 2020 documentation

1
1. Explanation of the work carried out by the beneficiaries and Overview of the progress

In this report, we'll provide an overview of the work of the ESRs in the project year from
1/2/2020 until 1/1/2021. After the report on the first project year, and the mid-term review
meeting, this is the second report for ABC-EU-XVA.

We first give a quick reminder:

ESR1 is Luis Souto, at the CWI in Amsterdam, the Netherlands. Since 1/8/2020, he
collaborates with ABANCA in A Coruna, Spain, however, he works from his home in the
Netherlands, due to the COVID19 measures in the respective countries.

ESR2 is Kristoffer Andersson, at the CWI in Amsterdam, the Netherlands. Since 1/10/2020
he collaborates with Belfius Bank in Brussels, Belgium, however, he works from home in the
Netherlands and Sweden, due to the COVID19 measures in the respective countries.

ESR3 is Graziana Colonna, at UDC in A Coruna, Spain. Since 1/9/2020 she collaborates with
EY in Amsterdam, the Netherlands, however, she works from home, in Italy, due to the
COVID19 measures in the respective countries.

ESR4 is Roberta Simonella, at UDC in A Coruna, Spain. Since 1/9/2020 she works with
UniPol in Bologna, Italy. She collaborates from her home in Italy, due to the COVID19
measures in the company.

ESR5 is Felix Wolf, at ULB, in Brussels, Belgium. Since 1/7/2020 he works with Rabobank,
in Utrecht, the Netherlands. He collaborates from his home in the Netherlands, due to the
COVID19 measures in the bank.

ESR6 is Kevin Kamm, at Alma Mater UniBo, Bologna, Italy. Since 1/9/2020 he works with
Banco Santander, in Madrid, Spain, however, he works from his home in Germany, due to the
COVID19 measures in the respective countries.

1.1 Objectives

This EID project aims to address a number of significant challenges arising from the
mathematical modelling, numerical computation and risk management, in the form of
valuation adjustments, of financial contracts. These adjustments represent a major focus of
the ongoing regulatory reform related to the recent global financial crisis. X-Value Adjustment
(XVA) refers generally to these valuation adjustments. The purpose of XVA is twofold: to
hedge possible losses due to a counterparty default event, and to determine the amount of capital
required by the bank under the new regulations. The "X" in XVA can be many different letters
nowadays, as financial industry has to deal with CVA (credit value adjustment), CollVA
(collateral value adjustment), DVA (debt value adjustments), FVA (funding value adjustment),

2
KVA (capital value adjustment), MVA (margin value adjustment), amongst others. They refer
to different parts of the chain of funding a financial transaction. In this EID project, four major
European banks (Banco Santander, Rabobank, Belfius Bank and Abanca), a major European
insurer (Unipol) and a major consulting and risk management company (EY, formerly Ernst &
Young) join efforts with five academic beneficiaries (from Universities in Delft, Brussels, A
Coruña and Bologna, plus CWI - the Center for Mathematics & Computer Science, in
Amsterdam) to educate 6 ESRs on various aspects of XVA. The standard formulas for the
calculation of XVA may not be sufficiently accurate under all (financial) circumstances and
may not take into account specifics of different financial portfolios.

This project lies at the intersection of financial mathematics, numerical mathematics and data
science, commonly referred to as Computational Finance. The objective is to perform multi-
disciplinary research and gain deep insight into a variety of aspects of financial counterparty
risk and their consequences. Moreover, we wish to incorporate (big) data and machine learning
techniques into the modelling and computation of XVA. From their projects, the ESRs will gain
expertise in counterparty risk, modern valuation adjustments and how to include these in the
financial mathematical models, in algorithms, software, (big) data. The ESRs will have an
excellent background to obtain high level jobs afterwards.

1.2 Explanation of the work carried per WP

The requirement to account for XVA brings the challenge of accurate modelling and efficient
pricing, even for commonly traded financial derivatives. This is due to the complicated nature
of different kinds of risk underlying XVA and the very procedures brought in to mitigate those
risks. The ESR projects are focused into 4 work packages (WPs). These WPs represent the
common stages in industrially-oriented mathematical research for financial applications. WP1
is “XVA modelling and data”, in which stochastic models for the quantities involved in the
XVA components are considered, selected and investigated. The choice of stochastic dynamics
and the measures (P and/or Q) have a major impact on the value adjustments, as well as on the
assessment of the risks associated. Proper handling of data is crucial to calibrate/validate
models. WP2 is “Numerical methods”, which covers the computations for the different value
adjustments we encounter in the ESR projects. In real-life industrial applications, a major
challenge is to develop efficient methods for accurate risk measures. In order to significantly
reduce the associated computational costs, as demanded by industry, WP3 “Software and
Hardware” is then connected to the implementation of the efficient algorithms on state-of-the-
art hardware, like on Graphics Processing Units (GPUs). WP3 also covers software aspects,
like specific programming languages, definition of a suitable software framework, and user
interfaces as desired by the industrial beneficiaries. This is also connected to machine learning,
where development already takes place in an advanced software framework. Test cases that are
very relevant to the involved industry will be defined and evaluated in the context of WP4
"XVA cases with real data". Each ESR will be working in each of the work packages. In all
work packages the industrial beneficiaries will have an active participation. Reporting will be
done by the academic beneficiaries. These WPs will give the ESRs an excellent overview of XVA
management in the financial industry.

Together, these four WPs gave the ESRs a fairly complete overview of aspects of risk
management in the financial industry. The choice of WPs is logical and relatively simple,
without too many details and subpackages, for clarity and convenience. In this first full project
year, the work by the ESRs has been mainly into WP1 and WP2, which we’ll report below.

3
1.2.1 Work Package 1

In work package WP1 “XVA modelling and data”, stochastic models for the quantities
involved in the XVA components are considered, selected and investigated. The choice of
stochastic dynamics and the measures (P and/or Q) have a major impact on the value
adjustments, as well as on the assessment of the risks associated.

In the context of the educational aspects within WP1, very early in the project, we had our first
ABC-EU-XVA Event.

Spring School on XVA modeling - A Pricing Framework for Valuation Adjustments

The Spring School which took place in Bologna (Italy) at the Department of Mathematics, on
May 16-17, 2019. The aim of the school was to provide self-contained lectures on XVA
modeling. Besides the ESRs, the school was open to a limited number (30) of academics and
researchers as well as practitioners. The invited speakers of the school were the renowned
professors Andrea Pallavicini (Banca IMI and Imperial College, London) and Marco
Francischello (Imperial College, London).

Andrea Pallavicini is the head of equity, FX and commodity models at Banca IMI, Milan, and
visiting professor at the Department of Mathematics of Imperial College, London. Over the
years he published several papers in financial modelling, theoretical physics and astrophysics.
He is the author of the books "Credit Models and the Crisis: a journey into CDOs, copulas,
correlations and dynamic models'', Wiley (2010), and "Counterparty Credit Risk, Collateral and
Funding with pricing cases for all asset classes", Wiley (2013). Marco Francischello is
Research Assistant of Mathematical Finance at Imperial College London, after having studied
there for his PhD under the supervision of Prof. Damiano Brigo. His research focuses on how
to incorporate market imperfections such as default risk, funding costs, collateral agreements,
and capital constraints into the valuation of derivative products.

The school provided a detailed analysis of the valuation of derivative products in presence of
default risk, collateral agreements, funding costs, and capital constraints. The courses offered
both a theoretical and an applied perspective, developing a pricing framework and showing its
applications and links with the current market practices regarding valuation adjustments.
Starting from credit and default risk, the concepts of Credit and Debt Valuation Adjustments
(CVA and DVA) were introduced. Collateralization as a natural way to reduce default risk was
discussed and applications of our valuation framework to Wrong Way Risk and Gap Risk were
shown. To make models more realistic, funding costs and the associated Funding Valuation
Adjustment (FVA) were introduced. The lecturers discussed the funding strategies available to
an investment bank and highlighted the economic arguments behind the FVA. Furthermore, the
framework was linked to different market practices regarding discount rates and netting sets
evaluation. Backward Stochastic Differential Equations were derived from the valuation
equation and some numerical implementations were presented.

The courses also illustrated how initial margins and capital constraints influence the valuation
of derivative products. The last part dealt with initial margins and their role in the valuation of
a derivative product and their relation with the Margin Valuation Adjustment (MVA).
Pallavicini showed a model for regulatory capital constraints and how banks alter their
investment decisions to abide to these constraints. Finally, he illustrated how the model gives a
novel interpretation to what the industry calls Capital Valuation Adjustment (KVA).

4
Some photos regarding the Bologna Spring School:

In the context of WP1, we will now discuss the research that has been performed by the 6
EWSRs.

ESR1 works on the modeling of wrong way risk. Credit risk is intended not only as default
risk, but also as migration risk, that is when the credit-worthiness of a given counterparty gets
worse, and recovery risk, i.e. the risk of not being able to recover the exposure at default
entirely. The XVA framework is the Basel III (and IFRS 9) response to this. Within the XVA
setting, a cutting-edge topic is the joint modelling of the probability of default (PD) and the loss
given default (LGD). Empirical studies have shown that the PD and the LGD tend to be
comonotonic, giving rise to a dangerous WWR, especially during crises. PD and LGD have
often been modelled as independent random variables, thus underestimating the risk. If we also
add the error due to the use of risk-neutral probability measures instead of the real-world ones,
it is easy to understand why the regulator requires more effective approaches to risk valuation.
We will deal with the dependence between PD and LGD, and take into consideration the
discrepancies between risk-neutral and physical measures.

The approach in ESR1’s work is novel in the sense that reinforcement urn processes, that are
tractable stochastic models, are being considered for the modeling. This is not yet been done in
the literature and highly promising. ESR1 already has a scientific report written in this direction,
and is in his present work (as of 1/12/2020) working on a more detailed model in this research
direction. This work is expected to be finalized mid 2021.

ESR2 has worked in the context of WP1 on portfolios of derivatives, written between two
parties. The first party is referred to as the bank and is considered to be default-free. The second
party, which may default, is referred to as the counterparty. The Credit Valuation Adjustment
(CVA) is the difference between the risk-free portfolio value and the risky portfolio value,
where the risky portfolio value is defined as the portfolio value when taking default risk of the
counterparty into account. While there is no ambiguity of the risk-free portfolio value, it is not
completely clear how the risky portfolio value should be computed. The question is whether
the exercise policy should be adjusted for the fact that the counterparty may default. For
instance, if the counterparty ends up in financial distress, it is reasonable to assume that the

5
other party would be more willing to exercise the callable derivatives, in order to lower its
exposure to the counterparty. In the case of a portfolio, the situation is more complex, and
depends on contractual details such as the close-out and netting agreements. One consequence
is that, in the presence of a netting agreement, the exercise decisions can no longer be made
individually. In general, for a risky portfolio, it is not possible to describe the value of a single
derivative, but only the value of the entire portfolio. This is an interesting problem since almost
all existing algorithms rely on exercise decisions made in isolation and risky derivative values
that can be added up to obtain the risky portfolio value. In practice, this means that the
counterparty currently may be paying a CVA which is based on a sub-optimal exercise strategy.
Even more problematic is that the overestimation of the CVA is higher for counterparties that
already are under financial distress.

As novel piece of research work by ESR2, that is currently being done, is modeling spreads in
market quotes and the limit order book in the context of the FX market. This research is
performed in order to assess the risk that arises due to the extensive use currently of algorithmic
trading. The aim is to contribute with an elegant mathematical problem formulation, on the
basis of forward-backward stochastic differential equations (FBSDEs), in which machine
learning play a prominent role in the numerical solution.

ESR3 has worked so far on counterparty risk modeling in the XVA context. Counterparty risk
is a specific type of credit risk, as it is the risk of default by the counterparty in a financial
derivatives contract. The counterparty fails to meet its obligations on that trade. Consider an
over-the-counter (OTC) option, sold by a hedger to an investor. If the option expires in-the-
money, the hedger owes the intrinsic value to the investor. Counterparty risk is the credit risk
that the hedger will not be able to fulfil its obligation to the investor (for example, the hedger
may have gone bankrupt). Bilateral counterparty risk gained importance in the wake of the
global financial crisis. Different techniques for the valuation of derivatives and portfolios under
counterparty risk have been developed. These have in common that, in order to calculate the
derivative value, a partial differential equation (PDE) needs to be solved. Alternatively, an
expected value has to be computed in an integral representation formula.

The dimension of the PDE depends on the number of stochastic market variables involved in
the model: the bigger the number of stochastic variables, the higher the dimension of the PDE.
In low dimensions, deterministic methods, such as the finite element method, are state-of-the-
art and give precise results. Developing algorithms for solving high-dimensional PDEs has been
an exceedingly difficult task, due to the ”curse of dimensionality”. Monte Carlo methods are
often appreciated due to their flexibility and applicability in high dimensions. The idea of
ESR3’s work is to combine these two instruments, exploiting the power of Monte Carlo
methods and the preciseness of a PDE solution. In particular, ESR3 starts from the fundamental
work of Christoph Burgard and Mats Kjaer, where, using hedging arguments, they derive two
PDEs, one linear and one nonlinear, in order to estimate the value of a risky derivative between
two defaultable parties, the hedger and the investor. The PDEs are one-dimensional PDEs (since
the stochastic market variable is only one, the underlying asset of the derivative) with two
constant parameters, the default intensities of both the hedger and the investor. ESR3 models
these default intensities as stochastic processes. The aim in her modeling is to combine a finite
element algorithm with a Monte Carlo approach, in a so-called hybrid modeling approach. This
is an approach to deal with the curse of dimensionality due to an increasing number of stochastic
processes. ESR3 has finalized a paper on this model, including numerical solutions, and a
second paper is on its way.

6
In the second piece of research, she has included DVA (Debt Value Adjustment, i.e., the
symmetric part of the valuation adjustment, connected to a parties’ own default probability)
into the modeling. This adds new complications and especially the differences in the modeling
results in ESR3’s first two papers are highly interesting to estimate the importance of including
DVA in the XVA framework.

Moreover, in the cooperation with EY the newly defined compounded overnight interest rate is
being considered and the differences between this new rate and the commonly used (but slowly
discarded) LIBOR rate. The impact of the new model on prices of interest rate derivatives and
their pricing is also up next.

ESR4 Recently, attention has been given to the extension of valuation adjustments from a single
currency to a multi-currency setting. ESR4 focuses on the pricing of European options under
valuation adjustments in a multi-currency setting. Moreover, stochastic intensities of default
are assumed and underlying assets denominated in different currencies are involved, with the
additional hypothesis of a zero default intensity for the hedger. The price of a derivative should
reflect the hedging costs transmitted by the hedger, in a world where a high percentage of
uncollateralized transactions is characterized by the presence of an investor (risk taker) and a
hedger (risk hedger). The exchange rates (or FX rates) are deterministic, so that foreign
exchange risk is not present and only the investor is defaultable.

Based on all assumptions, ESR4 infers linear and nonlinear partial differential equation (PDE)
formulations of the problem for the XVA price. Therefore, we are in a multi-dimensional
setting, where the involved stochastic factors are the underlying assets and the credit spread of
the investor.

ESR5 works on collateralization, a market mechanism which efficiently reduces credit risk
from over-the-counter derivatives by covering their outstanding exposure with low-risk
securities, like cash or bonds. Collateralization introduces new challenges to the pricing of
collateralized assets. Discounting needs, for example, reconsideration, as the interest rate paid
on the collateral inherits the role taken by the risk-free rate in the conventional setup.
Importantly, the pricing measure Q, associated with the risk-free rate in the standard model, is
instead determined by the collateral rate, which reflects the (positive or negative) interest paid
on the collateral account. The optimal choice of collateral securities is known as the cheapest
to deliver (CTD) collateral. Collateral posting strategies, which minimize the price of the asset,
are subject to constraints given by the credit support annex (CSA) of each asset. Under full
substitution rights, the entire collateral account can be switched from one collateral security to
another at any time. ESR5 considers the case of full substitution rights, assuming an
instantaneous exchange of collateral in continuous time. The stochastic CTD valuation then
depends on the distribution of an integral of the maximum of different stochastic processes,
which does not give analytically tractable dynamics, and require approximation. The focus of
ESR5’s modeling approach is on a conditionally independent approximation of the involved
processes at interpolation points, which leads to improved analytical tractability in a second-
order model, while taking into account the correlation structures observed in the market.
ESR6 has proposed and analyzed a stochastic version of the Magnus expansion for the solution
of linear systems of Ito stochastic differential equations (SDEs). The Magnus expansion
provides an exponential representation of the solution of a first-order linear differential
equation. The exponent is aggregated as an infinite series, whose terms involve multiple
integrals and nested commutators. This expansion has been successfully used to construct

7
algorithms for the numerical integration of matrix linear differential equations, as they inherit
from the Magnus expansion the preservation of qualitative traits of a problem.
ESR6 has proved existence and provided a representation formula for the logarithm associated
to the solution of the matrix-valued SDEs. The Magnus expansion also provides a novel method
for solving stochastic PDEs (SPDEs) for which
ESR6 presents some preliminary tests utilizing massive parallelization on a GPU to accelerate
the evaluation of such a model. SPDEs are the next new tool to use in the context of
contemporary risk management and XVA.
Following this approach, ESR6, in close collaboration with Banco Santander, is working on a
novel idea for the introduction of SPDEs in the context of so-called rating triggers, i.e. the
consideration of rating changes of a counterparty and not only a direct change to default. At
this stage of development, apart from choosing the right model, there are many difficulties to
overcome, such as a lack of data on rating changes in the risk-neutral world (Q-side), imperfect
data in the real world (P-side) due to the withdrawal of firms over time and inconsistencies in
the rating matrices published by the rating agencies, and a mismatch between the severity of
default risk on the P side and the Q side, i.e. usually the default risk implied by CDS is higher
than the default risk published by the rating agencies. In this context, the change of measure
and the corresponding Radon-Nikodym derivative becomes a crucial tool and the model must
take into account the different weighting of the probabilities of default, while allowing a
tractable change of measure and keeping the calculation effort feasible.
Summarizing, extensive work by all 6 ESRs has taken place in the context of work package 1.
Truly innovative approaches have been defined, and the first results of these models are
currently written up and sent out for publication. It is clear that the work packages 1 and 2
“Numerical methods” are closely connected, as new models need to be evaluated numerically.
The papers that are currently being written by all ESRs, are based on a combination of modeling
and numerical evaluation. However, in the coming period there will also be additions to this
work package, as some of the ESRs will also dive into a different aspect of modern risk
management in their follow-up work and papers
1.2.2 Work package 2

Work package WP2 “Numerical Methods” covers the computations for the different value
adjustments we encounter in the ESR projects. In real-life industrial applications, a major
challenge is to develop efficient methods for accurate risk measures.

In the context of education in WP2, a major Event has taken place, the ICCF2019. The 3rd
International Conference on Computational Finance (ICCF2019) was held in A Coruña from
July 8 to 12, 2019 (see http://iccf2019.udc.es/ ). Approximately 120 academic experts and
professionals from the field of Quantitative Finance and their computational aspects
participated in ICCF2019, in order to present recent advances of great interest in topics related
to the financial and insurance sectors. Participants came from 24 different countries. The
program is found at http://iccf2019.udc.es/index.php/preliminary-program and included 12
world- renowned plenary speakers:

• Stéphane Crépey, Université d'Évry-Val d'Essonne, France

• Griselda Deelstra, Université Libre de Bruxelles, Belgium
• Bruno Dupire, Bloomberg, USA
• Peter A. Forsyth, University of Waterloo, Canada

8
 • Julien Guyon, Bloomberg, USA
• Monique Jeanblanc, Université d'Évry-Val d'Essonne, France
• Rüdiger Kiesel, Universität Duisburg-Essen, Germany
• Andrea Pallavicini, Banca IMI, Italy
• Olivier Pironneau, Université Pierre et Marie Curie, France
• Christoph Reisinger, University of Oxford, United Kingdom
• Nizar Touzi, École Polytechnique, France
• Antony Ware, University of Calgary, Canada

The plenary conferences of Stéphane Crépey, Monique Jeanblanc and Andrea Pallavicini
were on current aspects of the XVA, the central theme of the EID ABC-EU-VVA. In
addition, thematic minisimposiums and communications were held.

An Industrial Day was held at the headquarters of Afundación, with lectures given by three
world- renowned professionals: Bruno Dupire (Head of Quantitative Research in Bloomberg),
Julien Guyon (Senior Quant in Bloomberg) and Andrea Pallavicini (Head of equity, FX and
commodities in IMI Banking), which talked about major issues in the financial sector. Next, a
round table was held on “Current hot topics in financial industry”. In it, in addition to Bruno
Dupire (Bloomberg), María R. Nogueiras (Head of Collateral Risk Analytics at HSBC
London), Lech Grzelac (Senior Quant of Rabobank), Peter Forsyth (extensive experience with
the sector) and Cornelis Oosterlee, who chaired the round table. Also to be noticed was the
presence of a relevant number of young researchers, 19 of them opted for the Journal of
Computational Finance award for the best work presented by them during the congress (see
http://iccf2019.udc.es/index.php/jcf-young-researcher-award ). Two special volumes of
articles related to conference presentations will be published in the Journal of Computational
Finance and International Journal of Computer Methods. This edition of Coruña is part of the
activities of the EID ABC-EU-XVA and gives continuity to the ICCF series, that held in
Greenwich (2015) and Lisbon (2017), emerging as an initiative within the ITN Strike network
developed between 2013 and 2016. On this occasion, ICCF2019 is also a satellite congress of
the world congress of applied and industrial mathematics ICIAM2019 (see
https://iciam2019.org/ ), in Valencia from July 15 to 19.

Photos of ICCF2019 in A Coruña

9
In the context of WP2, ESR1 works on Reinforced Urn Processes (RUPs) which represent a
flexible class of Bayesian nonparametric models. They are suitable for dealing with possibly
right-censored and left-truncated observations. A reliable estimation of the RUP hyper-
parameters is missing in the literature. In many applications, data is only available in an
incomplete form due to the nature of the experiment being conducted or the application. For
example, in medical studies where patients are observed over a limited period of time, the event
of interest—next stage of a disease or even death—may not occur during the observation period.
In this case we say that observations are right-censored. Furthermore, since in many cases
patients do not start being observed from the time they are born, this data will also be left-
truncated in case the missing time window is of interest, like in survival studies and risk
management. Although truncation and censoring have both been extensively studied when there
is only one target variable, much less progress has been made in higher dimensions, even just
for the bivariate case. The bivariate case is of special relevance for the many applications it has,
in which at least one variable is subject to either truncation or censoring . Given that left-
truncation and right-censoring fall under the umbrella of incomplete data, many authors have
opted for an approach based on the EM algorithm to tackle the problem. To the best of our
knowledge, nobody has proposed an extension of the EM algorithm to include bivariate left-
truncation. ESR1 proposes an extension of the Expectation-Maximization (EM) algorithm for
RUPs, both in the univariate and the bivariate case. Furthermore, a new methodology
combining EM and the prior elicitation mechanism of RUPs is developed: the Expectation-
Reinforcement algorithm. Numerical results showing the performance of both algorithms are
presented for several analytical examples as well as for a large data set of Canadian annuities.
A paper related to this work by ESR1 has been submitted for publication.

ESR2 has worked in the context of WP2 “Numerical methods” on the use of machine learning
techniques, particularly Artificial Neural Networks (ANNs), for the computation of Credit
Value Adjustment (CVA). Particularly, ESR2 proposes a neural network-based method for
approximating the expected exposures and potential future exposures of Bermudan options. In
a first phase, the method relies on the recently developed Deep Optimal Stopping algorithm
(DOS), which learns the optimal stopping rule from Monte-Carlo samples of the underlying
risk factors. Cashflow paths are then created by applying the learned stopping strategy on a new
set of realizations of the risk factors. Furthermore, in a second phase the cashflow-paths are
projected onto the risk factors to obtain approximations of pathwise option values. The
regression step is carried out by ordinary least squares as well as neural networks, and it is
shown that the latter produces more accurate approximations. The expected exposure is
formulated, both in terms of the cashflow-paths and in terms of the pathwise option values and
it is shown that a simple Monte-Carlo average yields accurate approximations in both cases.
The potential future exposure is estimated by the empirical α-percentile. Moreover, it is shown
that the expected exposures, as well as the potential future exposures can be computed under
either, the risk neutral measure, or the real-world measure, without having to re-train the neural
networks. This work has resulted in a research paper which is submitted for publication.
Moreover, in a second research paper, ESR2 generalized the neural network-based method for
CVA computations towards a portfolio of derivatives. In particular, ESR2 focuses on portfolios
consisting of a combination of derivatives, with and without true optionality, e.g., a portfolio
of a mix of European- and Bermudan-type derivatives. CVA is computed, with and without
netting, for different levels of wrong-way risk (WWR) and for different levels of credit quality
of the counterparty. It is shown that the CVA is overestimated with up to 25% by using the
standard procedure of not adjusting the exercise strategy for the default-risk of the counterparty.
For the Expected Shortfall of the CVA dynamics, the overestimation was found to be more than
100% in some non-extreme cases.

10
ESR3 has worked so far on counterparty risk modeling in the XVA context. To calculate the
derivative value, a partial differential equation (PDE) needs to be solved. The dimension of the
PDE depends on the number of stochastic market variables involved in the model: the bigger
the number of stochastic variables, the higher the dimension of the PDE. Developing algorithms
for solving high-dimensional PDEs has been an exceedingly difficult task, due to the ”curse of
dimensionality”. Monte Carlo methods are often preferred due to their flexibility and
applicability in high dimensions. The idea of ESR3’s work is to combine these two numerical
methods, exploiting the power of Monte Carlo methods and the preciseness of a PDE solution.
In particular, ESR3 starts from the fundamental work of Christoph Burgard and Mats Kjaer,
where, using hedging argu-ments, they derive two PDEs, one linear and one nonlinear, in order
to estimate the value of a risky derivative between two defaultable parties, the hedger and the
investor. The PDEs are one-dimensional PDEs (since the stochastic market variable is only one,
i.e. the underlying asset of the derivative) with two constant parameters, the default intensities
of both the hedger and the investor. These models have been extended to higher dimensions in
García, and numerically solved for one and two spatial dimensions in Arregui, Salvador
Mancho and Vázquez. Starting from these previous ideas, ESR3 models these default intensities
as stochastic processes, including them either as a new spatial dimension in the PDE or as a
stochastic coefficient. The aim is to combine a finite element algorithm with a Monte Carlo
approach, a so-called hybrid modeling approach. ESR3 has used a Multi-Level Monte Carlo
(MLMC) method, introduced by Michael Giles. This method relies on repeated random
sampling that are taken on different levels of accuracy. It can be considered an evolution of the
basic MC method. The MLMC method can significantly decrease the computational cost of
ordinary Monte Carlo methods by taking many samples with a low accuracy and corresponding
low cost, and only a very few samples are taken at high precision and corresponding high cost.
The Multi-level Monte Carlo method deals with the stochastic factors appearing in the PDEs,
and a Lagrange-Galerkin methodology allows to solve the PDEs. Because there are two
stochastic factors in the PDEs, the expected values of the (stochastic) solutions of the two PDEs
are computed. A paper on this hybrid numerical method applied to a PDE with one spatial
dimension is currently being finalized. In a second paper, a PDE model in two spatial
dimensions is being considered in this hybrid modelling and numerical approach.

ESR4 works on the extension of valuation adjustments from a single currency to a multi-
currency setting. ESR4 focuses on the pricing of European options with valuation adjust-ments
in this multi-currency setting. Moreover, stochastic intensities of default are assumed and
underlying assets in different currencies are involved, with the additional hypothesis of a zero
default intensity for the hedger. Based on assumptions and previous work by García, Lope and
Palomar, ESR4 poses formulations of the XVA pricing problem in terms of linear and nonlinear
PDEs. For this purpose, ESR4 employs hedging and no arbitrage arguments and makes a choice
for the mark-to-market value at default, that leads either to a nonlinear problem, if the mark-to-
market value is equal to the risky derivative price, or to a linear problem. In a second step, ESR4
deduces a formulation of the problem in terms of expectations with the goal of applying a Monte
Carlo method for computing the total valuation adjustment, following previous work by
Arregui, Salvador-Mancho and Vázquez, for the one currency setting. The choice of this
numerical method for the approximation of XVA is because the Monte Carlo Method is not
affected by the curse of dimensionality, that arises when employing other numerical approaches
to solve multi-dimensional PDE problems. This argument prioritizes the use of Monte Carlo
like techniques here. Summarizing, ESR4 has proposed PDEs models and expectations-based
formulations for the pricing of financial options and total valuation adjustment in a multi-
currency setting, where the involved assets are denominated in different currencies, thus leading
to multidimensional problems. In order to compute the price of the valuation adjustments, she

11
has employed the Monte Carlo method. The numerical examples help in the analysis of the
XVA behavior and its dependence on the underlying assets and the investor’s credit spread. A
first paper is almost finished for the case of deterministic FX rates. A second one is ongoing, in
which stochastic FX rates are incorporated and the corresponding model has been posed.

ESR5 works on collateralization, a market mechanism which efficiently reduces credit risk
from over-the-counter derivatives by covering their outstanding exposure with low-risk
securities, like cash or bonds. The optimal choice of collateral securities is known as the
cheapest to deliver (CTD) collateral. ESR5 considers the case of full substitution rights,
assuming an instantaneous exchange of collateral in continuous time. The canonical evaluation
of the stochastically based CTD option depends on the distribution of the maximum of the
stochastic processes, which is not analytically tractable. ESR5 shows that a second order Taylor
expansion of this distribution provides an accurate approximation under market conditions and
introduces a common factor model for the cross currency collateral spreads, which enables fast
and semi-analytical approximations of the involved moments. Regarding WP2, ESR5 presents
numerical results to the common factor approximations of the first and second order Taylor
expansions, with the spreads being modelled under the T-measure. For each currency, the mean
reversion speed parameters, the volatility parameters, the long-term mean curves, the initial
values and the instantaneous correlations require calibration and determine the parameters of
the common factor approximation scheme. Further, the number of additional currencies and the
maturity need to be chosen.

To calibrate the cross-currency collateral spread model on market data, cross-currency option
prices are employed which are not liquidly traded in the market. Alternatively, historical data
can be used when assumptions on the forward spreads linked to the “short” spreads are made.
ESR5 considers artificial data of realistic magnitude for the mean reversion speed and volatility
parameters to show scope and limitations of the common factor approximation model. The
common factor model is restricted to non-negative correlations. As correlations cannot be
obtained from the market, this assumption is tested against the historical correlations of cross-
currency basis spreads. Except for short maturity exceptions, which can be attributed to
statistical noise, this assumption appears justified. A research paper on this topic is being
finalized by ESR5.

ESR6 works, in the context of WP2, on a numerical scheme for the Magnus expansion. He has
proposed and analyzed a stochastic version of the Magnus expansion for the solution of linear
systems of Ito stochastic differential equations (SDEs). A new method is proposed for the
numerical solution of stochastic partial differential equations (SPDEs) based on spatial
discretization and application of the stochastic Magnus expansion. A notable feature of the
method is that it is fully parallelizable. In his first paper on this topic, ESR6 derived the
stochastic Magnus expansion for Ito SDEs and proved its convergence up to a positive stopping
time. The main result also provides an asymptotic estimate for the optimal stopping time.
Additionally, the Magnus expansion is tested for the numerical solution of stochastic
differential equations, confirming its potential as a highly parallelizable and accurate method.
Numerical tests in order to assess the accuracy of the numerical schemes confirm this.
Preliminary tests have also been presented for the case in which the Magnus expansion provides
a novel method for solving SPDEs. From a numerical perspective, some of the main advantages
and drawbacks of the Magnus expansion are: Pros: the calculation of the matrix logarithm is
very fast by means of the expansion, and it can be done very rapidly multiple times. The
expansion is highly parallelizable, so that it is possible to employ Graphics Processing units for
the computation. Cons: The calculation of the matrix exponential can be slow with standard

12
libraries, and like its deterministic counterpart, the stochastic expansion works only up to a
finite time. Furthermore, the methodology is memory demanding in high dimensions. At an
early stage in the search for a suitable model related to rating triggers, ESR6 proposed to extend
the discrete ratings commonly used in the industry to a continuous framework, which allow
simple Monte Carlo-based implementations of transition probability densities. Since a
stochastic transition probability density is already a 5-parameter function, it requires
calculations on a GPU to guarantee feasible calibration times to market data. In a next step,
more sophisticated models are sought in order to correctly capture market behaviour with the
idea in mind to use the stochastic Magnus expansion as the tool to deal with computational
efficiency.

Summarizing, important research work has been performed in the context of WP2 by all 6 ESRs.
We encounter modern and advanced versions of the Monte Carlo method (multi-level variants,
and also in the context of expectation-maximization, and reinforcement-maximization
techniques), PDE discretizations, hybrid combinations of PDEs and Monte Carlo methods, and
even advanced machine learning and data-driven neural network technologies. Truly innovative
approaches have thus been defined, and the first papers are currently written up and sent out for
publication. It is clear that the work packages 1 and 2 “Numerical methods” are closely
connected. ESR1 is currently writing his third research paper, where the first two have been
sent to renowned journals for publication. ESR2 has also submitted two research papers to top
journals. ESR3 has almost finalized her first research paper, which will also soon be sent out
for publication and the same is true for ESR4 and ESR5. ESR6 has one paper submitted and
currently works on his second research paper. All submitted papers can be found in the open-
access arxiv.org repository, as well as on the ABC-EU-XVA publication webpage.
1.2.3 Work package 3

In the context of WP3 “Software and Hardware” a Crash course on Python Programming,
the second ABC-EU-XVA Event, was organized in A Coruña, Spain. Python is a powerful and
highly productive multipurpose object oriented programming language: good for scripting,
prototyping and building complex libraries, as well as a glue between different languages.
Thanks to this power and versatility it has spread among all environments with large software
requirements, such as engineering and finance, achieving huge software ecosystems in these
fields, among many others. In the frame of the European Industrial Doctorate ABC-EU-XVA,
the Crash course on Python Programming was organized at the University of A Coruña on July
4 and 5th, 2019. The aim of this course was to give an introduction to the Python programming
language, with applications in scientific and financial computing. The course was 12 hours long
in a hands-on modality: students had the opportunity to follow the teachers explanations and to
program small examples and codes. It was taught by Prof. Iñigo Arregui (associate professor at
University of A Coruña), Prof. José A. García (associate professor at University of A Coruña)
and Dr. Álvaro Leitao (who was post-doctoral researcher at University of Barcelona and is now
a post-doctoral researcher in University of A Coruña). The main attendants were the six ESRs
of ABC-EU-XVA, who just started their doctoral stage at their respective universities.
Moreover, it was also open to a small number of other students and researchers. Finally, a
professor on applied mathematics, another doctoral student on applied mathematics and master
students of physics also attended the course. The program of the course and additional
information are available at the web site http://dm.udc.es/crashpython2019/

Photos of the Python crash course Event.

13
1.2.4. Work package 4

WP4 "XVA cases with real data" is devoted to the evaluation of test cases that are relevant
to the involved industry. These test cases will be defined and evaluated in the context of the
secondments that have recently started. Each ESR will be working on a specific relevant test
case. In all work packages the industrial beneficiaries will have an active participation, but a
focus is on WP4. The work in this WP is for the coming period.

14
1.2.5 Work package 5

WP5 is on the management and general education.

Starting with the educational aspect, at CWI Amsterdam, an Event was organized, in
Amsterdam, Sept. 2 - 5 2019 (with soft skills, vlog communication skills and Basel III training).
Program, The CWI Event on Soft Skills, Vlogging and the EY course on Basel III
Monday 2 September, 10:00 - 17:00hrs, CWI, Science Park, Amsterdam
*09:00-17:00hrs “Taking Charge of your PhD project I”
* 19:00hrs dinner Restaurant Dauphine, Amsterdam
Tuesday 3 September, 10:00 - 17:00hrs
* “Taking Charge of your PhD project II”, with an actor involved
Wednesday 4 September, 9:00 - 17:00hrs
* Vlog course 'How to make a video-vlog with your smartphone', CWI
Relevant information regarding this course was given beforehand, and an app needed to be
downloaded ahead of the course.
Thursday 5 September, 9:00 - 17:00hrs
Visitor address: EY headquarters, Antonio Vivaldistraat 150, Amsterdam
* Basel III course:
- Basics of Banking (and regulatory trends)
- History; from Basel I until Basel III (1988 - 2013)
- Latest updates from Basel; finalization of Basel III (2017)
- XVA, and the Basel impact on the CEO/CFO agenda
- Basel game
Friday 6 September, 10:00hrs
"A mathematical guidance through Amsterdam"

The PhD start up course is a well-thought course; the instructor had a lot of experience with
different PhD students and understands the difficulties and also how to tackle them. This is also
a very interactive course in the sense that the instructor challenges the participants to question
what their goals are and what needs to be done to achieve them. There is a part where the ESRs
should work together with an actor.

Vlogging - this course is important because of outreach and public appearance to a larger
audience. It was explained how to create a video professionally and how to present a scientific
message to a specific audience in the best way. It was possible to decide what audience the
ESRs wanted to “talk to”. The instructor was good in helping us with our specific tasks.
Basel course - The course was somewhat technical from an economic perspective, with a lot of
content. This is because the course was also used for internal EY education. The students can
keep the information for future reference.
At the end of this course week, all ESRs have received three Attendance Certificates.
This was also a good week for ESR Team building!

In the context of WP5, also the Kick-Off Meeting and the Mid-Term review meeting took place.

15
After the start of the EU project on XVA, called ABC-EU-XVA, November 1st 2018, two
important activities took place first. First of all, the application text for the recruitment of the 6
PhD candidates was established, and the text was circulated internationally to receive CVs of
excellent candidates for the 6 ESR positions. We already then decided to start with these ESRs
by approximately May 1st 2019. As a second important activity we planned the formal Kick-
Off meeting for the project, in Amsterdam, the Netherlands, February 5th 2019. For this
occasion we reserved a private room at restaurant Enoteca located at The Manor Hotel
Amsterdam: ENOTECA Linnaeusstraat 89 1093 EK Amsterdam.The program of this evening
involved short presentations about the participants in the project. The aim was to get to know
each other better, and to define the project consortium. All participants of ABC-EU-XVA
introduced their companies/university groups and departments during this Kick-Off meeting.
Also the expectations of the project were communicated.

Photo of the Kick-Off Meeting:

Recruitment within EID ABC-EU-XVA

Here, we will report about the recruitment procedure. We started with the recruitment
December 1st 2018 (before that date we prepared the application text, the communication and
advertisement by the usual channels). The vacancies were posted on our webpage:
http://fineng.ewi.tudelft.nl/ABCXVA/Applications/index.html. We have also advertised on
Euraxess, https://euraxess.ec.europa.eu/jobs/360974. All advertisements were also placed on
all institutional webpages, Academic Transfer and related sites, All-acad.com, Academics
positions, Informatics Europe, The six positions have been advertised together, next to the
individual calls for candidates on our institutional webpages. Deadline for applications was
set to January 15th, 2019.

For ESR1 and ESR2 we have received 36 + 53 applications, respectively, for ESR3 and 4 in
total 16+10 applicants applied, for ESR5 we had 25 applicants, and for ESR6 24 applicants
wrote us a mail. There were quite a few candidates without the proper preknowledge, but
generally a high number of suitable applicants remained. After a pre-selection of the
candidates, based on their CVs, for each ESR position we had a list of approximately five to
eight favorable candidates. Interviewing these candidates took place in two stages. The first

16
stage was a general interview with the potential candidates by means of skype. After these
pre-selection interviews, there were basically three favorite candidates remaining for each of
the positions. These were interviewed a second time. Moreover, on several lists with
favorites, the same candidates appeared on top. These candidates could choose their favorite
positions first. After this round, essentially three ESR positions were filled with the applicants
of our choice. The other three positions were filled after a session in which the PIs, together
with the connected industrial beneficiaries, interviewed several candidates in person.

The day after the Kick-Off meeting, i.e., Wednesday February 6th 2019, we invited the
candidates to Amsterdam, for a presentation plus application interview, in the CWI. This
interview round, we asked for short presentations about their MSc work.. By March 1st 2019,
it became clear that all positions had been offered to the desired top candidates and that all
these candidates had accepted the ESR positions. They got to know each other already mid-
May when our first ABC-EU-XVA Event took place in Bologna. Two out of six ESRs are
female. All candidates have been jointly recruited with the responsible industrial beneficiary,
so all selected ESRs are known at the industrial site. The start date for all 6 ESRs has been
May 1st 2019. All ESRs have also signed agreements with the respective Graduate Schools at
their universities. They have all started with Graduate School courses. Next to this, they have
all written the EID ABC-EU-XVA Career Plans. These plans have a standard format, which
was placed in an Appendix of the Consortium Agreement.

Details regarding the Supervisory Boards

Also in the context of WP5, we have established the two Boards, governing our EID, i.e. the
Supervisory Board (SB) and the Research and Training Board (RTB). These Boards form the
main feedback mechanisms about the quality of education and research and to make suggestions
for further education and research. As mentioned in the Consortium Agreement, the RTB is
composed of the academic partners plus one of the PhD students (it is Felix Wolf). All
beneficiaries are present in the SB, plus one of the PhD students (Graziana Colonna), plus two
professors from outside the consortium who are scientific experts on the topics of our project.
The two academic members of our SB are: prof. Laura Ballotta, CASS Business School,
London, https://www.cass.city.ac.uk/faculties-and-research/experts/laura-ballotta and prof.
Karel In't Hout, U. Antwerp, Belgium, https://www.uantwerpen.be/nl/personeel/karel-inthout/.
The SB meets typically once a year, when an Event takes place. These can be Skype meetings.

Prof. Karel In’t Hout and prof. Laura Ballotta, the scientific experts in our SB.

17
Another important event that was organized in the context of WP5 was the Mid-Term Review
meeting, in Brussels, 23-01-2020, followed by the Board meetings, at ULB, Université Libre
de Bruxelles, 24-01-2020. The Mid-Term Meeting was held in the beautiful Salle Marie-
Thérèse at the Academy Palace, located at Hertogsstraat 1 Rue Ducale, 1000 Brussels. EU
officer Dr. Filippo Gagliardi attended the first day, talked to the ESRs in a private meeting as
well as with the coordinator of the project.

Thursday, 23 January 2020 (first day)

09.00 – 09.20 Welcome to Brussels (Griselda Deelstra); introduction by REA project

officer and ABC-EU-XVA coordinator (Cornelis Oosterlee)
09.20 – 10.00 Tour du Table: All scientists-in-charge present their teams and role in the
network
10.00 -- 10.30 REA project officer presentation (monitoring, reporting and purpose of mid-
term meeting)
10.30 – 10.50 Coffee / tea break
10.50 – 11.45 Coordinator’s report: presentation of the network and progress
11.45 – 12.15 Brief presentations by the ESRs (part I, 3 ESRs)
12.15 – 13.15 Lunch

13.15---13.45 Brief presentations by the ESRs (part II, 3 ESRs)

13.45 – 15.00 Brief presentations of industrial beneficiaries

15.00 – 15.20 Coffee / tea break

15.20 – 16.20 Restricted session of EU officer with the ESRs

16.20 – 16.40 Restricted session between EU officer and the coordinator

16.45 Feedback from REA project officer and open discussion

Conclusion of the meeting: we’ll walk to the restaurant and we conclude there with

19.00 Common dinner with ABC-EU-XVA participants

Friday, 24 January 2020 (second day)

09.30 – 10.15 General discussion with all ABC-EU-XVA members about Mid-Term
Review, and a look into the near future
10.15 – 11.00 Details about ESRs moving to the industry; and internal group discussions
11.00 – 11.20 Coffee / tea break
11.20 – 12.20 RTB meeting followed by Supervisory Board Meeting ABC-EU-XVA
12.20 – 13.00 Sandwich lunch
18
The feedback on the project from the Mid-Term review meeting was very positive. The project
is very well on its way, the ESRs gave very good presentations. The organization is up and
running. The Events organized by the project were considered very positive.
Photos: Two group photos at the Mid-Term Review meeting

19
Our scientific experts, profs. Laura Ballotta and prof. Karel In’t Hout were also present at the
Mid-Term Review meeting.

Day 2: The internal Boards meeting and group discussion

On-line workshop 2/7/2020: Machine Learning in Quantitative Finance and Risk

Management

We also report about the on-line workshop “Machine Learning in Quantitative

Finance and Risk Management“, which took place in the context of ABC-EU-XVA. The

20
workshop was organized by ESR2 (Kristoffer Andersson) and Cornelis Oosterlee, from
CWI, with a speaker from ABC-EU-XVA (Andersson) and all ESRs attended the workshop
in the audience. We had 120 participants in the morning session and 80 participants in
the afternoon session. The contents were highly interesting from the ABC-EU-XVA
perspective. The workshop took place, on-line, via zoom.us, and the programme was as
follows:

Thursday July 2nd 2020,

The workshop times are CET, Central European Time:

10:00 AM: (Keynote) Christoph Reisinger (U. Oxford): “Deep xVA solver — A neural
network based counterparty credit risk management framework”
11:00 AM: Kristoffer Andersson (CWI): “Learning exposure profiles for portfolios of
exotic derivatives”
11:50 AM: Shashi Jain (IISc Bangalore): “Universal static hedging using a shallow neural
network”

13:30PM: Anastasia Borovykh (Imperial Coll., London): “To interact or not? On

convergence properties of interacting particle optimization”
14:20PM: Shuaiqiang Liu (TU Delft): “Deep learning for large time-step simulations of
stochastic differential equations”
15:15PM: (Keynote) Yuying Li (U. Waterloo, Canada): “Asset allocation without pain:
learning dynamic strategies directly from market data”

The slides from the workshop can be found under

https://portals.project.cwi.nl/abcxva/events/slides-from-the-speakers

The video recordings of the presentations are found here:

https://www.youtube.com/playlist?list=PLtBl3dSNhbhgoh3hWkezs-K9HPXvNcTPh

1.3 Impact

It is good to mention that the XVA teams are identified as the front-office groups that grew
fastest within the banks. Although credit value adjustments have been calculated since the early
2000s, mainly to assess the impact of counterparty risk in capital at a trade level, recent
regulatory requirements ask for well-educated people on the theoretical aspects as well as on
their practical computational issues. XVA has become crucial from the regulatory viewpoint,
so that human resources need to be trained properly, not only for banks but also for other
companies, like consultants, and for the regulatory institutions. In this setting, it is of a high
importance to train ESRs and PhD students related to these risk management applications. It is
not easy these days to enter a financial company without a decent background in financial
mathematics. The ESRs in ABC-EU-XVA will have the unique opportunity to cooperate and
work in the industry, focusing on outstanding current topics like XVA and counterparty risk,
which is favorable for their industrial careers. The financial, analytical, computational and
problem-solving skills acquired in the research training are transferable to many areas far

21
beyond their own scientific specialization. The broad expertise training will improve the
researchers' career prospects and options for employability, also in different public and private
sectors. This will be strengthened by the additional training in project management skills (team
leadership, diversity and multicultural aspects), organizational skills (events, time management,
fund raising), improvement of language and presentation skills (to specialists, to managers).
The projects and educational events shall stimulate creativity and the entrepreneurial mindset
of the ESRs. So far, the following papers has been sent for publication to highly respected
journals with international circulation, see also https://portals.project.cwi.nl/abcxva/overview-
research-programme/publications ,

ESR1, Luis Souto:

1. Joint and Survival Annuity Valuation with a Bivariate Reinforcement urn process.

By a Bi-variate Reinforced Urn Process (B-RUP), a novel way of modeling the dependence of
coupled lifetimes is introduced, with application to the pricing of joint and survivor annuities.
In line with the machine learning paradigm, the model is able to improve its performance over
time, and it allows for a priori information, like experts’ judgement, to complement empirical
data. Using a well-known Canadian data set, the performances of the B-RUP are studied and
compared with existing literature.
2. Estimation for univariate and bivariate reinforcement urn processes under left-
truncation and right-censoring
Reinforced Urn Processes (RUPs) represent a flexible class of Bayesian nonparametric models
for dealing with possibly right-censored and left-truncated observations. A reliable estimation
of their hyper-parameters is missing in the literature. We therefore propose an extension of the
Expectation-Maximization (EM) algorithm for RUPs, both in the univariate and the bivariate
case. Furthermore, a new methodology combining EM and the prior elicitation mechanism of
RUPs is developed: the Expectation-Reinforcement algorithm. Numerical results showing the
performance of both algorithms are presented for several analytical examples as well as for a
large data set of Canadian annuities.
ESR2: Kristoffer Andersson:

1. A deep learning approach for computations of exposure profiles for high-dimensional

Bermudan options.
In this paper, we propose a neural network-based method for approximating expected exposures
and potential future exposures of Bermudan options. In a first phase, the method relies on the
Deep Optimal Stopping algorithm (DOS) proposed in [1], which learns the optimal stopping
rule from Monte-Carlo samples of the underlying risk factors.

2. Deep learning for CVA computations of large portfolios of financial derivatives

In this paper, we propose a neural network-based method for CVA computations of a portfolio
of derivatives. In particular, we focus on portfolios consisting of a combination of derivatives,
withand without true optionality, e.g.,a portfolio of a mix of European- and Bermudan-type
derivatives.CVA is computed, with and without netting, for different levels of WWR and for
different levels of credit quality of the counterparty. We show that the CVA is overestimated
with up to 25%by using the standard procedure of not adjusting the exercise strategy for the
default-risk of the counterparty. For the Expected Shortfall of the CVA dynamics, the
overestimation was found to be more than 100% in some non-extreme cases.

22
ESR6: Kevin Kamm:
1. On the stochastic Magnus expansion and its application to SPDEs
We derive a stochastic version of the Magnus expansion for the solution of linear systems of
Ito stochastic differential equations (SDEs). The goal of this paper is twofold. First, we prove
existence and a representation formula for the logarithm associated to the solution of the matrix-
valued SDEs. Second,we propose a new method for the numerical solution of stochastic partial
differential equations (SPDEs) based on spatial discretization and application of the stochastic
Magnus expansion. A notable feature oft he method is that it is fully parallelizable. We also
present numerical tests in order to asses the accuracy of the numerical schemes.

The ESRs will be given ample opportunity of exposure by giving presentations at workshops
and conferences. Publications arising from the research will attract international attention in
industry and academia. International experience and proficiency in foreign languages are
important ingredients in the transnational European job market. Through this network the
transnational mobility will be extended to the doctoral level, increasing the career prospects of
the researchers. It also develops an early mind setting in each one’s family, necessary to make
transnational mobility successful.

However, the COVID19 crisis in Europe severely disturbed these travel plans of the ESRs in
2020. All ESRs are improvising these days regarding their secondments, and are happy to be at
research work, and to find their connection to the industrial partners, and to their academic
supervisors, from their homes. Working from home appears to be really tough in such a project,
where several of the PhD students are not well connected to research groups, and where
travelling is impossible. Hopefully, things get back to normal again soon, so that the ESRs are
allowed to travel to the industrial partners and to different countries.

For now, we’ll plan an on-line presentation cycle by the ESRs, where the next meeting will take
place, on-line, 21/1/2021.

1.4. Access provisions to Research Infrastructures

No access to research infrastructures has been provided under the grant.

1.5 Resources used to provide access to Research Infrastructures
Not applicable.

23
2. Update of the plan for exploitation and dissemination of results

As mentioned earlier in this report, all ESRs are working more-or-less from their homes, as the
industry, as well as the universities, is reluctant to open their buildings under the COVID19
measures. This is fully understandable, and we have to deal with this situation.

The plans regarding the research output in the form of scientific publications remains
unchanged. The pace of the scientific progress however will be influenced by the COVID
situation. Communication is more difficult with supervisors and also with the partners. It is
clear, that the connection to the industrial beneficiaries is not ideal, in these times. We are
pragmatic and aim, most of all, for successful PhD dissertations for our ESRs.

The COVID times also has impact on the public outreach in our project. “Days of open doors”
in universities and research institutes were cancelled this year, conferences and workshops
didn’t take place. There is a start of on-line workshops, and, although it is not the same, we’ll
also organize an online workshop, 21/1/2021, in which our ESRs can present their work. This
on-line workshop replaces the workshop that was planned early February 2021 in the context
of the Actuarial and Financial Mathematics Conference, AFMATHCONF, (which will not take
place next year).

There will also be changes regarding the planning of the vlogs that were proposed in the context
of our project. A first set of vlogs by our ESRs can be found at our webpage, see

https://portals.project.cwi.nl/abcxva/overview-research-programme/vlogs-about-the-eu-
project/vlogs

These are nice, and positive vlogs. We expected a sequence of these 2-3 minutes vlogs over the
years, about the work, the moving to another city, the academic employer, the mid-term review
meeting, the industrial employer, and, eventually, about the PhD defence and the project
finalization. However, as can be understood, the COVID19 crisis in Europe severely disturbed
these vlog plans. All ESRs are improvising these days regarding their secondments, and are
happy to be at research work, and to find their connection to the industrial partners, and to their
academic supervisors, from their homes. Moreover, vlogging from home doesn’t make sense
in terms of publicity. So, we have decided not to vlog in 2020, and, hopefully, there will be
vlogs of our ESRs on 2021, when they are allowed to travel to the industrial partners, and when
things get back to normal (hopefully) again.

3. Update of the data management plan (if applicable)

No changes
4. Follow-up of recommendations and comments from previous review(s) (if applicable)
A recommendation from our scientific experts in the project was to have more presentations
from the ESRs. We have had several on-line meetings in which individual ESRs could present
their work. Coming 21/1/2021, we’ll have an on-line workshop for a full day in which the ESRs
will all present.
5. Deviations from Annex 1 and Annex 2 (if applicable)

24
5.1 Tasks
The ESRs cannot travel to the industrial beneficiaries and partners, due to the COVID measures.
As such, the work modus is completely different from what was anticipated.
This is a big change, putting a lot of pressure to all members of this EID, particularly to the
ESRs. This working mode also impacts the research tasks, as, for example, it is not possible to
work with certain types of (confidential) financial data, that is only available on a company’s
laptop. There are on-line weekly contacts to the industrial hosts and also to the academic
supervisors. We’ll manage, but for an EID this is clearly not ideal. As soon as changes to this
modus are possible again, we’ll reconsider and see how to adapt the working modus.
5.2 Use of resources
This is an MSCA, so this item is not applicable.

25