## Financial Risk graduate

Are you in the final phase of your master and interested in the problems financial organisations face to become both financially stable and liquid in this fast-changing global environment? Then a Graduate Internship at Deloitte Financial Risk perfectly suits you!

As a Graduate Intern at Deloitte Financial Risk you will get intensive support from your internship supervisor, who will be there for advice. While you work on your thesis, you are part of the Financial Risk talent pool. This means that besides writing your thesis you will also have the opportunity to work on projects for our clients or help with research assignments. Moreover, an internship at Deloitte gives you access to our extracurricular activities, such as

- Training sessions
- Drinks
- Teambuilding activities

At Financial Risk, a service line within Deloitte Risk Advisory, our goal is to become the undisputed advisor within the financial services industry. Different teams focus on

- Market, Credit & Actuarial Risk;
- Capital Management, Liquidity & Treasury Risk;
- Accounting & Financial Reporting Risk.

If you combine writing your thesis with an internship at Deloitte, you will receive a compensation that is competitive in the market. The total amount of compensation you get depends on the number of days you will be with us. In addition to the internship compensation, we offer travel allowance in case you do not have a student transport card. You will also be provided with a laptop during your internship period.* *Read more about our employment benefits.

### List of thesis topics

Below you find a list of thesis topics we have available at the moment, divided into the different teams of Financial Risk. Each project can obviously be tailored to your specific interests and needs! Click on one of the topics to read more about the topic.

## Market, Credit & Actuarial Risk

- Credit Value Adjustment (CVA) Risk Charge
- BGM Libor Market Model in Risk Neutral and Real World Measure
- Modelling Client Behaviour of Savings in a negative interest rate environment
- Migration matrix design IFRS9
- Application of machine learning in non-life insurance pricing
- Risk adjustment determination under IFRS 4 phase II
- Integrated Non-Life Insurance Modelling
- Bayesian Variable Selection in Linear Regression
- Rule extraction algorithms
- Identification of high-cost claims
- Machine learning in required capital estimations
- Stochastic correlation models for calculating capital requirements
- Risk mitigating techniques under new accounting standards for insurance contracts
- Application of machine learning to reduce financial reporting risk
- Risk adjustment determination under IFRS 17
- AI Insurance management board
- Asset performance based liability valuation
- Dynamic pricing in life insurance
- Aggregated risk pricing with machine learning
- Variability of risk premiums in non-life insurance
- Development of NINR reserve over the policy lifetime
- Analysis of methodology and assumptions prescribed by PRIIPs regulation

## Capital Management, Liquidity & Treasury Risk

### Market, Credit & Actuarial Risk

**Topic:** Credit Value Adjustment (CVA) Risk Charge

**Area of expertise:** Market Risk

**Abstract:** The Basel Committee recently proposed adjustment to the CVA framework within Basel III. The aim of these proposals is to achieve more alignment with accounting rules (IFRS 13), using risk-neutral measures, as well as a more alignment with the Fundamental Review of the Trading Book (FRTB). One of the proposals involves the removal of the Internal Modelling Approach for CVA. Banks can now choose between the Standardised Approach and the Basic Approach. One question that rises: What is the impact of the adjustments on capital requirements for CVA? Topics closely related are: valuation of derivatives, collateral, netting, portfolio effects, exposure modelling, IFRS 13, Basel III, FRTB.

**Topic:** BGM Libor Market Model in Risk Neutral and Real World Measure

**Area of expertise:** Market Risk

**Abstract:** Pricing interest rate derivatives is performed in the real-world, but what happens if we need “real-world” risk metrics for the risk departments? The solution is a combination of real-world and risk-neutral simulations. Possible directions for research are 'Implement of change of measure in the BGM model', 'Calibration of the utility function that links the risk neutral and the real world measures' and 'Application of Longstaff-Schwartz to avoid nested simulations and to increase the computational performance of the model'.

**Topic:** Modelling client behaviour of savings in a negative interest rate environment

**Area of expertise:** Credit Risk

**Abstract:** Currently, in current negative interest rate environment, banks are faced with a tough questions – can we lower our savings rates below zero? And how will clients react if we do so? As savings is an important funding source for banks, banks want to understand the risk whether clients will leave the bank once they set the savings rates at zero (or even below zero). Common industry practice however in modelling and measuring client behavior in a negative interest rate environment however is not yet established. The objective of the thesis is hence to contribute to this industry practice. And establish a methodology of modelling client behavior of savings in a negative interest rate environment.

**Topic:** Migration matrix design IFRS9

**Area of expertise:** Credit risk

**Abstract:** For IFRS 9, banks need to estimate lifetime expected credit losses for credit facilities with significantly increased credit risk in order to determine the provisions that need to be held. Common industry practice is to use migration matrices to extend 1 year Probability of Default (PD) to multi-year PDs/ cumulative PDs by multiplying migration matrices under Markov Assumptions. However, the matrix also has to be conditioned on the state of the economy to incorporate the forward looking information as required by IFRS9. The question is whether Markov properties still hold when a matrix is conditioned on forward projections of the economy, i.e. migration matrices for future periods are conditioned on historical data. Furthermore, is it possible to measure the impact if the Markov properties don’t hold anymore and is there a possible remediation? Finally, in terms of migration matrix design, is it possible for migration matrices to also incorporate the possibility of cures or should default be defined as an absorbing state?

**Topic:** Bayesian variable selection in linear regression

**Area of expertise:** Credit Risk

**Abstract:** Financial institutions are performing regressions for many different models. Selecting the best set of regression parameters is often performed manually by the developer of the model. Model developers sometimes say; “It’s more of an art than a science”. Bayesian variable selection allows for the selection of a parameter set that is in fact mathematically optimal. Which, at the very least, is a great way to challenge the “art” of parameter selection. Credit risk is a possible direction for research. Low default portfolios (LDP) account for a large share of total bank lending. Due to the scarcity of default observations and subsequent need for numerous assumptions in order to model probability of default (PD) for LDP’s, model calibrations are needed, which on their turn, introduce significant model risk. This is usually absorbed by applying high level conservatism. Bayesian variable selection could offer a reduction of model risk.

**Topic:** Rule extraction algorithms

**Area of expertise:** Credit Risk – Advanced credit scoring

**Abstract:** Modern machine learning techniques (e.g. random forests, neural networks) deliver a great improvement in terms of predictive power in comparison with conventional techniques (e.g. generalized linear models, logistic regression) and are increasingly used to replace or enhance risk models. However, their complexity makes these kind of models hard to interpret and validate. Can we use rule extraction algorithms (deducing a set of prediction rules such that these prediction rules have (approximately) the same predictive power as the underlying model, but the rules are easier to interpret) for advanced machine learning models, to make the inner workings/decision making of the algorithm transparent and interpretable.

**Topic:** Machine learning in required capital estimations

**Area of expertise:** Actuarial Risk

**Abstract:** Insurance companies are required to retain a part of their own funds as capital to withstand losses in (severe) stress scenarios. The amount of capital insurance companies need to hold reflects a loss event that could occur, on a statistical basis, once every 200 years. This is according to the Solvency II capital directive, which applies to all European (re)insurance companies. To calculate the required capital, also known as the Solvency Capital Requirement (SCR), insurance companies need to take their complete risk profile into account. E.g. insurance companies are exposed to market risk, life underwriting risk, etc. To be able to calculate the SCR for their entire portfolio, insurance companies use an approximation approach. In general, the value of assets and the insurance liabilities are approximated for several economic shock scenarios (i.e. quantiles of a loss distribution). The most commonly used approximation approaches is termed ‘curve fitting’ which essentially means that the value of an insurer’s assets and liabilities are estimated under particular (non-)economic shock scenarios by means of a polynomial regression. Given the value of an insurer’s assets and liabilities under a subset of predefined, e.g. equity shocks, a curve can be fitted through these points, constituting a loss distribution. The combination of loss distributions for multiple risk drivers allows the insurer to estimate its total capital requirement by taking into account all risks and the connection between them. In practice, for more non-linear and complex risk drivers (e.g. credit spread risk), the quality of the polynomial fit could be improved upon. In this thesis you will investigate whether machine learning algorithms do a better job at finding a suitable, i.e. better fitting loss distribution than traditional regression approaches and, hence, are better able to approximate the ‘true’ capital requirement in a 1-in-200 loss scenario.

**Topic:** Stochastic correlation models for calculating capital requirements

**Area of expertise:** Actuarial risk

**Abstract:** Insurance companies are required to retain a part of their own funds as capital to withstand losses in (severe) stress scenarios. The amount of capital insurance companies need to hold reflects a loss event that could occur, on a statistical basis, once every 200 years. This is according to the Solvency II capital directive, which applies to all European (re)insurance companies. To calculate the required capital, also known as the Solvency Capital Requirement (SCR), insurance companies need to take their complete risk profile into account. E.g. insurance companies are exposed to market risk, life underwriting risk, etc. Insurance companies deal with many risk factors which have to be taken into account in their models. These risk factors may be correlated, which means that the risk factors may move in the same direction or in an opposite direction. Nowadays insurance companies who apply an internal model to calculate their SCR tend to use stochastic simulation models where they use static (deterministic) correlations between risk factors. The number of risk factors they take into account may be larger than 100. It depends mostly on the level of granularity on which the risk factors are taken into account. In the current situation insurance companies calculate the correlations between the different risk factors deterministically and keep them constant in their models for an entire year. The aim of this thesis is to determine if there is added value in making the correlations between risk factors dependent on time, i.e. stochastic. As part of this research you will focus on questions like:

• Do correlation parameters vary over time and can a statistical distribution be found that appropriately models their behaviour?

• Does it make a difference whether to use linear or rank correlations?

• What is the impact on SCR as a result of introducing time dependent correlation factors?

• What are the technical/computational challenges for applying stochastic correlations and can these be mitigated?

**Topic:** Risk mitigating techniques under new accounting standards for insurance contracts

**Area of expertise:** Actuarial Risk

**Abstract:** On January 1th, 2021 a new accounting standard (IFRS17) will be introduced which will change the way insurers show their financial results to the world. Both profits as well as losses need to be reflected in a more transparent manner including the movement over time. Investors and analysts will, for the first time ever, be able to compare the dividend generating capacities of insurers within the market. However, these new accounting standards set the principles on which the financial reporting must be based. And principles give room for interpretation. Based on Deloitte’s current understanding you will build further on the development of risk mitigating techniques like reinsurance contracts and derivatives to provide insight on how the SII and IFRS17 are complementary or contradictory.

**Topic:** Application of machine learning to reduce financial reporting risk

**Area of expertise:** Actuarial Risk

**Abstract:** On January 1th, 2021 a new accounting standard (IFRS17) will be introduced which will change the way insurers show their financial results to the world. Both profits as well as losses need to be reflected a more transparent manner including the movement over time. Investors and analysts will, for the first time ever, be able to compare the dividend generating capacities of insurers within the market. However, these new accounting standards set the principles on which the financial reporting must be based. And principles give room for interpretation. This thesis topic looks deeper into different options for the moment of transition, insurance policies must be reflected as if they were under IFRS17 from the start. However, there are many different option to determine the starting value (fully retrospective, modified retrospective and fair value). The choices made at the transition date determine the business strategy going forward. As a thesis student you will apply the machine learning techniques to find the optimal balance between the value at the moment of transition (Jan 1th 2021) and all possible scenario’s possible thereafter. Application of machine learning techniques are especially interesting for students with a quantitative back ground while the application towards insurance provides a clear link with the actuarial risk advisory practice of Deloitte.

**Topic:** Risk adjustment determination under IFRS 17

**Area of expertise: **Actuarial Risk

**Abstract:** The insurance industry has been subject to many regulatory developments in recent years and these developments are still ongoing. The traditional framework of actuarial calculations has been further expanded. The introduction of concepts as market value and market consistency ultimately led (as mutual agreements between insurers) to the Market Consistent Embedded Value (MCEV) approach to determine the economic value of insurance obligations. One of the components required to determine market value under IFRS 17 is the so-called 'risk adjustment', this component is comparable to the 'risk margin' under Solvency II but not exactly the same. The choice for the methodology to determine this risk adjustment is free, options include the Cost-of-Capital approach or a method based on conditional tail expectation. In addition to the own chosen method, each insurer must also report a risk adjustment based on a Confidence Level methodology. The graduation program has the following objectives: 1. Examining the possible methods that can be used to determine the risk adjustment under IFRS 4 phase II, and 2. Analyze the differences between the different methods and the consequences of different assumptions on the outcomes.

**Topic:** AI Insurance management board

**Area of expertise:** Actuarial Risk

**Abstract:** Do you believe that you have what it takes make an impact in the board room? Are you ready to challenge the board of an insurance company on their decisions? Are you quantitative but also able to see the bigger picture? In short, do you have what it takes to create a virtual board? The introduction of Solvency II provides more information to the management board. Additional risk and performance metrics have been added, reducing the complexity of running an insurance company to an optimization strategy of the new performance metrics. In this graduation program students take this concept one step further and develop a model that can replace a management board as a whole.

**Topic: **Asset performance based liability valuation

**Area of expertise:** Actuarial Risk

**Abstract: **This thesis internship is the ideal combination between financial risk management, asset modelling and gaining insurance insights. The Matching Adjustment (MA) under Solvency II provides the possibility for insurers to make a cash flow match between assets and liabilities for a balance sheet approach to valuation. The starting point of this thesis internship will be to explore the characteristics of insurance liability cash flows. For the selection framework of assets the MA guidelines can be leveraged. The core of your thesis will be the asset selection mechanism which will be able to solve for the optimal balance sheet composition under MA and other real world restrictions. The thesis internship concludes with a case study where the outcomes of the thesis are tested as if the insurer would have to report under IFRS17. This topic is especially interesting for students with a financial or actuarial back ground while the application in the case study will make your first mark in the insurance sector.

**Topic: **Dynamic pricing in life insurance

**Area of expertise:** Actuarial Risk

**Abstract:** Dynamic pricing is becoming more common in the Dutch insurance landscape. Recent developments have been particularly focusing on non-life insurance products. In this thesis subject the goal is to investigate the potential gains for life insurance products when using pricing techniques from the non-life sector. Apply general linearized models or machine learning techniques to predict profitability of Term life or annuity products with predicting variables such as age or smoker status. One of the areas that needs further research is the price elasticity, so what effect will a price increase (or decrease) on a portfolio segment have on the corresponding sales volumes. Build a simplified insurance product to illustrate the effects.

**Topic:** Aggregated risk pricing with machine learning

**Area of expertise:** Actuarial Risk

**Abstract:** Risk pricing is at the heart of non-life insurance for pricing and profitability analysis. Where in typical businesses the costs of the goods sold is known before offering the product to the market, in insurance a large and crucial part of the costs is known only after offering the insurance policy: the claims. Risk pricing is the art of estimating the expected claim amounts given the policyholders’ characteristics. Traditional risk pricing approaches take a frequency-severity approach per product and peril. This strategy does not take into account the interactions which may exists between frequency and severity models, perils or products. Machine learning enables the user to model the insurance risk in a single model, while capturing the interactions which exist between the various covers. This research aims at investigating the added value of modelling the risk premium at an aggregated product, peril and risk premium level. Research questions are aimed to build a business case for implementation of machine learning on an aggregate level. To this end, the results will be compared with traditional GLM results, both quantitatively and qualitatively.

**Topic:** Variability of risk premiums in non-life insurance

**Area of expertise:** Actuarial Risk

**Abstract: **Dynamic pricing aims at finding an optimal balance between the customers’ risk profile, their price elasticity and the competitors’ position in the market. Based on these models, the expected customer lifetime value can be maximised to find the optimal premium. The estimated risk profile typically is a point estimate of the expected future claims. By investigating the variability of the policyholder’s risk profile, a distribution of the customer lifetime value can be estimated. This helps the non-life insurer in balancing competitiveness and profitability of premiums, further optimising their product offering.

**Topic:** Development of NINR reserve over the policy lifetime

**Area of expertise:** Actuarial Risk

**Abstract: **The actuarial cycle links the core actuarial activities of non-life insurance: pricing, reserving, risk and performance management. However, in practice little communication exists between the models in each of these individual elements. Deloitte has the vision and ambition to integrate the pricing, reserving, risk and performance management and simultaneously incorporate individual policy risk. We attain this goal by introducing a new type of reserve in addition to the well-known IBNR and RBNS reserve: the Not Incurred and thus Not Reported (NINR) reserve. The NINR is the estimate of the expected future claim amount on a policy. It thus replaces the current premium reserve. You will study what factors influence the development of the NINR over the policy lifetime; which could consist of the remaining lifetime of the policy, whether the policyholder already made a claim and individual policyholder characteristics.

**Topic:** Analysis of methodology and assumptions prescribed by PRIIPs regulation

**Area of expertise:** Market Risk

**Abstract: **As of January 1, 2018 the so-called Packaged Retail and Insurance-based Investment Products (PRIIPs) regulation came in to place, impacting the banking, insurance and asset management industries. The PRIIPs regulation lays down regulatory technical standards with regard to key information documents (KIDs), which aim to improve the retail investor’s understanding of PRIIPs, the risks involved, and the comparability of those products. The legislation prescribes that parties offering PRIIPs should provide KIDs to investors. The KIDs are 3 pages long and contain general considerations, product information, performance indicators and risk indicators. The key aspects of the performance and risk indicators are market risk, credit risk, liquidity risk and costs. The regulation prescribes how to calculate these performance and risk indicators. For example, with regards to market risk a historical simulation based on the (historical) principal components of the interest rate curves is used to derive performance and risk indicators for interest rate derivatives. Deloitte advices its clients on the implementation of the PRIIPs regulation, in particular the calculation of these performance and risk indicators. Deloitte observes that model choices made by the regulator are debatable from a technical perspective and that some parts of the regulation are even left for interpretation. Deloitte would like to investigate the implication of the model choices and points of interpretations from an academic point of view. This could provide banks, insurers and asset managers with more insights in the potential shortcomings of the methodology and assumptions prescribed by the PRIIPs regulation and thus also the information communicated to its investors.

**Do you have other interesting ideas? Let us know! You are always welcome to present your topic to us.**

### Capital Management, Liquidity & Treasury Risk

**Topic:** Funds Transfer Pricing

**Area of expertise:** Capital Management

**Abstract:** Bank products (savings, mortgages, loans) are priced with a specific pricing framework to transfer costs from saving products and long-term funding to loans (e.g. mortgages). Several financial market developments affected the price (and pricing) of bank products, such as more stringent requirements for capital and liquidity. The wholesale funding has also changed (in terms of access, pricing and volatility). The balance sheet structure and risk appetite regarding various risks (including credit, liquidity, interest rate risk) are important components affecting pricing as well. In order to have a strong business model with a healthy long-term interest rate margin, insights and forecasts for bank product prices are important. Research and analysis on the components of product pricing gives the bank insight to steer the balance sheet.

**Topic:** Capital requirements and total loss-absorbing capacity of (inter)national banks

**Area of expertise:** Capital Management

**Abstract:** In 2014, a European directive was issued (EU Bank Recovery and Resolution Directive (BRRD)). The reason for this directive was that the financial crisis has shown that there is a lack of adequate tools to effectively deal with failing banks. During the crisis Member States were forced to save banks that were "too big to fail". The purpose of the BRRD is to prevent as much as possible the need for such actions. For this purpose, the banks' loss absorbing capacity must be increased and, if necessary, the provision of own and debt to bankruptcy (bail-in). The two requirements related to the loss absorbing capacity are TLAC (Total Loss Absorption Capacity) and MREL (Minimum Requirement for own funds and Eligible Liabilities). What levels of TLAC and MREL do different banks currently have? Which products meet the requirements of the BRRD and what is its impact on banks? How does bail-in work and does the contribution to the Deposit Guarantee System (DGS) play a role?

**Topic:** Value based accounting of banking book products under stress

**Area of expertise:** Capital Management

**Abstract:** Many banking product are valued on a market value basis. However, this market value isn’t always as stable or as representative for the “real” value of the product. Especially in times of stress, market values tend to be extremely volatile. Bid/Ask spreads tend to widen. Incorporating these new market values in the banking book can lead to large differences in value, even negative, while the real losses often tend to be less severe. How good is this policy? Why should or shouldn’t banks incorporate these market values in their banking book? How realistic are the values? Is there a reasonable band with for market value of a product?

**Do you have other interesting ideas? Let us know! You are always welcome to present your topic to us.**