11 September 2015 MVA, KVA: modelling challenges Moez MRAD Head of Credit & XVA Quantitative Research moez.mrad@ca-cib.com Views and opinions expressed in this presentation are the personal ones of the speaker. They may differ from current, past or future employer views, policies and interpretations Crédit Agricole Corporate and Investment Bank is authorised by the Autorité de Contrôle Prudentiel et de Résolution (ACPR) and supervised by the ACPR and the Autorité des Marchés Financiers (AMF) in France and subject to limited regulation by the Financial Conduct Authority and the Prudential Regulation Authority. Details about the extent of our regulation by the Financial Conduct Authority and the Prudential Regulation Authority are available from us on request.
Some CCR Challenges / Expected XVA Fundamental Review of the Trading Book (FRTB) - 2018: CVA VaR calculations for IR and FX will require the computation of several sensitivities on EPE profiles. Failure to meet the requirements for the FTRB approach will result in a fall back to the standardized approach (SA- CVA) and higher RWA/Capital costs as a result. Initial margin requirements (MVA) 2016: BCBS: All financial firms that engage in non-centrally cleared derivatives must exchange initial and variation margin as appropriate to the counterparty risks posed by such transactions. The standardized approach is add-on based / may be replaced by SA-CCR later on. Failure to come up with an internal model will result in a fall back to the standardized approach. Capital consumption: SA-CCR review (2017) / Leverage ratio (2018) / Net Stable Funding Ratio (2018) / Liquidity Coverage Ratio (2019). It is important to estimate the KVA: the present value of the future capital costs due to the various charges Internal Model: To get the term structure of UL we have to nest a VaR like computation (the UL at the desired confidence level and time horizon) within the Monte Carlo simulation. Page 1
Initial Margin and MVA IM posted has to be funded, IM received cannot be re-hypothecated. Margin Value Adjustment (MVA): Expected future funding cost due to IM. IM reduces over-collateralization which lead to less CVA. IM also reduces the need for capital as it reduces counterparty credit risk. Credit risk turned into liquidity risk and credit-related costs into funding costs. Internal Model based method Calibrated for 99% confidence interval, Margin Period of Risk at least 10 days. Compute IM per netting set and asset class. Sophisticated enough to cope with regulatory requirements. Brute force approach: Bump risk factors and revalue. It needs a high computational effort. Can not be achieved without use of specific technology (GPU and/or other recent technologies). Numerical techniques based approach: Compute analytic derivatives (sensitivities) per trade. Possible for the trades eligible for CCPs. Use sensitivities with historic market moves to compute the exposure distribution. Derive VaR or Expected Shortfall directly from the distribution (similar to parametric VaR). More complex products can be handled in a similar way (semi-exotics) or replaced by a regression. Page 2
Capital Requirements and KVA (I) Main sources of capital consumption for derivatives: Counterparty Credit Risk (CCR). CVA risk charge. Market risk (VaR and Stressed VaR). Capital consumption depends on: Counterparty nature: CCPs and many non-financials are exempt from CVA charge and/or have lower CCR. CSA (via CVA risk charge and credit exposure): high volume, complexity or disputes increase MPR and hence CVA risk charge. collateralization and initial margins reduce CCR charge. Regulator and jurisdiction Exemptions from CVA risk charge vary in different jurisdictions Availability of CDS protection to reduces CVA risk charge Received Initial Margin (and/or independent amount) Diffusion Models, measures, calibration Page 3
Capital Requirements and KVA (II) Why is capital required on top of the other XVA (Credit risk, Funding cost )? Other XVA adjust accounting for the expected loss (or expected cost) over the life of a transaction as seen at t= 0. Unexpected loss arise as well. They do not impact accounting as not a fair value (a drift) but rather a risk. These unexpected losses have to be covered by allocated Bank s capital, for an amount corresponding to the desired confidence level the Bank wants being able absorbing these unexpected losses at. Capital value Adjustment (KVA): Up fronted value of unexpected losses How to upfront unexpected losses? Up front revenues over the whole life of the transactions / portfolio. Getting the initial unexpected loss, even if not straightforward, isn t sufficient What is required is the whole forward profile of the time and market dependent term structure of UL A calculation difficulty arises here To get the term structure of UL means we have to nest a VaR like computation (The UL at the desired confidence level and time horizon) within the Monte Carlo simulation getting us the EL in the first place. If the VaR to be compution relies itself on a numerical procedure (MC or historical VaR), we multiply in a potentially explosive way the number of valuations to be carried Brute force approach or smart numerical methods might be considered when applicable Page 4
Numerical approaches for sensitivities computation AAD XVA are functions with lots of input variables (such as market data) Sensitivities are approximations of the mathematical derivatives. Derivatives can be computed via: Finite differences (FD). Adjoint Algorithmic Differentiation (AAD). AAD can be implemented via: Operator overloading (advised for XVA) Source transformation All the sensitivities will be displayed in one single sweep: same computation time for 10 or 1000 sensitivities! Page 5
AAD Pros and Cons Pros Sensitivity to n inputs is computed in constant time! Well suited to Monte-Carlo simulations Multithreading/parrallel computing compatibility C++ compliant Cons Memory consuming! Implementation of ADD is quite tricky in practice Does not improve (significantly) sensi precision Not as intuitive as FD Robustness Efficient for processes with high computation cost like XVAs estimation. Page 6
AAD for XVA AAD methodology can be divided into three fundamental steps: 1. Tree construction: the XVA estimation can be seen as a sequence of basic operations i.e. a pricing tree. 2. Tree evaluation: the pricing tree is evaluated for each simulation scenario. 3. Adjoint Backward propagation: sensitivities are computed in a backward mode: the tree is browsed from output (XVA) to inputs (market data). Thus, it is necessary to store: All the operators of the pricing tree (done in step 1). Results of these operators for each Monte Carlo Simulation (step 2). All the computation time for the three steps should not exceed 4 times the standard pricing computation time in theory. In practice, closer to 8. Page 7
Performance Tests 800 sensitivities in AAD performance in CVA single evaluation units Construction Evaluation AAD propagation 0,893 0,936 3,269 Page 8
MVA computation (I) SIMM Extended to MVA Compute sensitivities for each scenarios and each time step (analytically and/or via AAD) Use of sensitivities in order to get valuation changes: 1. First choice consider historical VaR shocks; and use sensitivities to estimation portfolio valuation changes; then sort scenarios and compute the 99% quantile. 2. Second choice is to use parametric VaR techniques: compute the square root of the sum of squared portfolio valuation change, compute the root mean square multiply it by a normal standard deviate corresponding to 99% and derive a normal or Gaussian covariance VaR approach. Page 9
MVA computation (II) MVA by replication Compute a VaR inside a Monte carlo simulation: 1. Use Longstaff-Schwartz regression function for all products of the portfolio. 2. Longstaff-Schwartz in its original form require Augmentation in order to estimate VaR efficiently. Two augmentation methods: Early State Augmentation (Wang & Calfish (2009)): Simulations start earlier than today. Preserve path continuity (needed for portfolio with bermudan products). Shocked State Augmentation (Green & Kenyon (2014)): Apply one VaR shock at each simulation date, on each path. Simpler, parsimonious, does not require extra simulation. Only for non bermudan products (example cleared ones). Apply shocks and compute VaR for each time and each scenario. Page 10
KVA computation (I) KVA is up fronting future capital consumption K(t), 0< t < T. KVA = 0 T h t B 0, t p C,B t E Qt K t dt K t = K MR t + K CVA t + K CCR t KVA = KVA MR + KVA CVA + KVA CCR KVA MR can be computed using standardized Approach (Basle II): K MR is not dependent on market conditions. Internal Model: Nested VaR computation (Similar to MVA) KVA CVA can be computed using standardized Approach (Basel III): We end up by up-fronting EAD (t). Internal Model: Nested CVA Credit VaR computation (Similar to MVA) Page 11
KVA computation (II) KVA CCR can be computed by up fronting K CCR (t) = weight 12.5% EAD(t) Weight is computed via: (1) standardized or (II) Foundation or Advanced IRB approaches. EAD t can be computed as in: 1. Current Exposure Method: MtM + add-on / some netting is allowed for add-on. 2. standardized approach : formula depending on transaction, collateral and their risk. 3. SA-CCR: tries to mimic the IMM approach but still depend on MtM at simulation time. KVA computation complexity similar to CVA or DVA (especially in CEM). 4. Internal model: Exposure at default EAD t = 1.4 EEPE(t) Effective expected positive exposure EEPE t = 1 T t 1 T t 1 0 max EE(t + u)dv 0<u<v Each time and scenario we need to estimate EEPE (nested Monte Carlo!) In practice solved by American Monte Carlo techniques or Branching Process (see Andreasen (2015)) Page 12
Disclaimer 2015, CRÉDIT AGRICOLE CORPORATE AND INVESTMENT BANK All rights reserved. This research report or summary has been prepared by Crédit Agricole Corporate and Investment Bank or one of its affiliates (collectively Crédit Agricole CIB ) from information believed to be reliable. Such information has not been independently verified and no guarantee, representation or warranty, express or implied, is made as to its accuracy, completeness or correctness. This report is a commercial communication as defined in article 6 of the Directive 2000/31/CE of 8 June 2000. For the avoidance of doubt, it is not a communication à caractère promotionnel within the meaning of the Règlement General AMF. It is provided for information purposes only. Nothing in this report should be considered to constitute investment, legal, accounting or taxation advice and you are advised to contact independent advisors in order to evaluate this report. It is not intended, and should not be considered, as an offer, invitation, solicitation or personal recommendation to buy, subscribe for or sell any of the financial instruments described herein, nor is it intended to form the basis for any credit, advice, personal recommendation or other evaluation with respect to such financial instruments and is intended for use only by those professional investors to whom it is made available by Crédit Agricole CIB. Crédit Agricole CIB does not act in a fiduciary capacity to you in respect of this report. Crédit Agricole CIB may at any time stop producing or updating this report. Not all strategies are appropriate at all times. Past performance is not necessarily a guide to future performance. The price, value of and income from any of the financial instruments mentioned in this report can fall as well as rise and you may make losses if you invest in them. Independent advice should be sought. In any case, investors are invited to make their own independent decision as to whether a financial instrument or whether investment in the financial instruments described herein is proper, suitable or appropriate based on their own judgement and upon the advice of any relevant advisors they have consulted. Crédit Agricole CIB has not taken any steps to ensure that any financial instruments referred to in this report are suitable for any investor. Crédit Agricole CIB will not treat recipients of this report as its customers by virtue of their receiving this report. Crédit Agricole CIB, its directors, officers and employees may effect transactions (whether long or short) in the financial instruments described herein for their own accounts or for the account of others, may have positions relating to other financial instruments of the issuer thereof, or any of its affiliates, or may perform or seek to perform securities, investment banking or other services for such issuer or its affiliates. Crédit Agricole CIB may have issued, and may in the future issue, other reports that are inconsistent with, and reach different conclusions from, the information presented in this report. Crédit Agricole CIB is under no obligation to ensure that such other reports are brought to the attention of any recipient of this report. Crédit Agricole CIB has established a Policy for Managing Conflicts of Interest in relation to Investment Research which is available upon request. A summary of this Policy is published on the Crédit Agricole CIB website: http://www.ca-cib.com/sitegenic/medias/doc/91928/2011-politique-gestion-conflits-interets-ca-cib-va.pdf.. This Policy applies to its investment research activity. None of the material, nor its content, nor any copy of it, may be altered in any way, transmitted to, copied or distributed to any other party without the prior express written permission of Crédit Agricole CIB. To the extent permitted by applicable securities laws and regulations, Crédit Agricole CIB accepts no liability whatsoever for any direct or consequential loss arising from the use of this document or its contents. France: Crédit Agricole Corporate and Investment Bank is authorised by the Autorité de Contrôle Prudentiel et de Résolution ( ACPR ) and supervised by the ACPR and the Autorité des Marchés Financiers ( AMF ). United Kingdom: Approved and/or distributed by Crédit Agricole Corporate and Investment Bank, London branch. Crédit Agricole Corporate and Investment Bank is authorised by the ACPR and supervised by the ACPR and the AMF in France and subject to limited regulation by the Financial Conduct Authority and the Prudential Regulation Authority. Details about the extent of our regulation by the Financial Conduct Authority and the Prudential Regulation Authority are available from us on request. United States of America: This research report is distributed solely to persons who qualify as Major U.S. Institutional Investors as defined in Rule 15a-6 under the Securities and Exchange Act of 1934 and who deal with Crédit Agricole Corporate and Investment Bank. Recipients of this research in the United States wishing to effect a transaction in any security mentioned herein should do so by contacting Crédit Agricole Securities (USA), Inc. (a broker-dealer registered with the Securities and Exchange Commission). The delivery of this research report to any person in the United States shall not be deemed a recommendation of Crédit Agricole Securities (USA), Inc. to effect any transactions in the securities discussed herein or an endorsement of any opinion expressed herein. Italy: This research report can only be distributed to, and circulated among, professional investors (operatori qualificati), as defined by the relevant Italian securities legislation. Spain: Distributed by Crédit Agricole Corporate and Investment Bank, Madrid branch and may only be distributed to institutional investors (as defined in article 7.1 of Royal Decree 291/1992 on Issues and Public Offers of Securities) and cannot be distributed to other investors that do not fall within the category of institutional investors. Hong Kong: Distributed by Crédit Agricole Corporate and Investment Bank, Hong Kong branch. This research report can only be distributed to professional investors within the meaning of the Securities and Futures Ordinance (Cap.571) and any rule made there under. Japan: Distributed by Crédit Agricole Securities Asia B.V. which is registered for securities business in Japan pursuant to the Law Concerning Foreign Securities Firms (Law n 5 of 1971, as amended), and is not intended, and should not be considered, as an offer, invitation, solicitation or recommendation to buy or sell any of the financial instruments described herein. This report is not intended, and should not be considered, as advice on investments in securities which is subject to the Securities Investment Advisory Business Law (Law n 74 of 1986, as amended). Luxembourg: Distributed by Crédit Agricole Corporate and Investment Bank, Luxembourg branch. It is only intended for circulation and/or distribution to institutional investors and investments mentioned in this report will not be available to the public but only to institutional investors. Singapore: Distributed by Crédit Agricole Corporate and Investment Bank, Singapore branch. It is not intended for distribution to any persons other than accredited investors, as defined in the Securities and Futures Act (Chapter 289 of Singapore), and persons whose business involves the acquisition or disposal of, or the holding of capital markets products (as defined in the Securities and Futures Act (Chapter 289 of Singapore)). Switzerland: Distributed by Crédit Agricole (Suisse) S.A. This report is not subject to the SBA Directive of January 24, 2003 as they are produced by a non-swiss entity. Germany: Distributed by Crédit Agricole Corporate and Investment Bank, Frankfurt branch and may only be distributed to institutional investors. Australia: Distributed to wholesale investors only. This research, and any access to it, is intended only for wholesale clients within the meaning of the Australian Corporations Act. THE DISTRIBUTION OF THIS DOCUMENT IN OTHER JURISDICTIONS MAY BE RESTRICTED BY LAW, AND PERSONS INTO WHOSE POSSESSION THIS DOCUMENT COMES SHOULD INFORM THEMSELVES ABOUT, AND OBSERVE, ANY SUCH RESTRICTIONS. BY ACCEPTING THIS REPORT YOU AGREE TO BE BOUND BY THE FOREGOING. 11/05/15 Page 13