Basel 2.5 Model Approval in Germany Ingo Reichwein Q RM Risk Modelling Department Bundesanstalt für Finanzdienstleistungsaufsicht (BaFin)
Session Overview 1. Setting Banks, Audit Approach 2. Results IRC 3. Results CRM 4. Results Stress VaR
Session Overview 1. Setting Banks, Audit Approach 2. Results IRC 3. Results CRM 4. Results Stress VaR
Basel 2.5: Specific Interest Rate Risk Trading book portfolios plain vanilla credit trading products (contains no correlation risk, no securitizations or ntd) Calculation Method VaR svar IRC CRM SA X X X securitizations and ntd not eligible for CTP (X) (X) X CTP CRM securitizations and ntd X X X other financial instruments X X (X) X SA securitizations and ntd (X) (X) X other financial instruments X X X X Abbreviations: X / (X) mandatory/optional waiver optional (partial use) VaR : value-at-risk IRC : incremental risk charge CTP : correlation trading portfolio svar : stressed VaR CRM : comprehensive risk measure SA : standard approach
Market Risk Models in Germany Before 2012 After Basel 2.5 Banks with model approval Interest Rate Risk 13 12 general 13 12 specific 7 6 Equity Risk general 12 11 specific 10 10 FX 9 9 Commodities 6 6 IRC Banks Plus two CRM approaches Banks with VaR (1-day/99%) ranging from 1 to 80 million EUR
On-site Examinations On-site examinations to assess IRC and CRM approaches Audit assignment (BaFin mandates Deutsche Bundesbank) Definition of audit scope, preparation through supervisory meeting ( pre-visit ) Writing decree of examination to the bank Sending letter of audit assignment to Deutsche Bundesbank On-site examination by Deutsche Bundesbank Deutsche Bundesbank writes examination report Final assessment Request and analysis of bank's written comments (hearing) Assessment of audit findings and preparation of the administrative act (approval, sanctions, qualitative factor, terms and conditions) Tracking (findings, terms and conditions)
Statistics: On-site inspections Basel 2.5 IRC Weeks (on-site) Examinors Bank 1 3 5 Bank 2 term 1 4 6 term 2 2 4 Bank 3 9 9* Bank 4 5 6 Bank 5 5 6 Bank 6 5 6 Bundesbank spent approx. 300 man weeks on-site Extent determined by bank s complexity *not all examinors were on-site all the time Earliest start date: Jun 2010 CRM Weeks (on-site) Examinors Latest end date: Sep 2011 Bank 1 6 7 Bank 2 9 14*
Session Overview 1. Setting Banks, Audit Approach 2. Results IRC 3. Results CRM 4. Results Stress VaR
IRC requirements Mandatory for preliminary approval Primary focus of on-site examinations Complete and adequate capturing of credit exposure positions Appropriate stochastic modeling, one-year-horizon, 99.9% probability Default and migration risks (e. g. probabilities) Appropriate parameterization (e. g. segmentation, correlation) Qualified model validation Not mandatory for preliminary approval Integration into risk management processes (Use Test) Reporting Limitation Not in focus
Capturing of Positions Findings caused by ambiguous rules text positions in institution s own debt Clarification by EBA guideline on IRC expected positions in sovereign bonds (should be included) Findings caused by the requirement to exclude securitizations detecting and excluding securitizations and ntd s was difficult data enrichment needed Other Findings caused by new interfaces and processes (e. g. update and review of relevant parameters to identify ntd s)
IRC Model in a Nutshell 1. Factor Model 2. Migration Matrix 3. Valuation All German banks are basically using the same factor model approach! Differences: choice of factors, calibration, valuation
systemic factor idiosyncratic factor Factor Model Choice of factors position sector intra-sector issuer asset correlation often similar to S&P Credit Pro segmentation only few findings: inappropriate choice of factors (e. g. not all available information used, bank could not justify simplification) Calibration of factor model (inter and intra sector correlations) All German banks use Gaussian Copula! Variety of approaches, from pure reliance on default and migration time series to a mixture of equity time series (for fine tuning) and default time series (for the level), but no pure reliance on equity time series for both (inter- and intra-sector correlations) additionally reliance on IRB parameters and expert judgment
Factor Model Findings regarding the calibration of factor model Weaknesses in all approaches detected Caused by requirement to base the calibration on observable market data (e. g. justification needed for expert judgment detail) Representativeness of market data for banks portfolio Appropriateness of mixture approaches (How to combine different data?) Example based on pure default and migration calibration Pairwise asset correlations difficult to compare (among banks) due to different choice of factors inter-sector correlation * intra-sector correlation = pairwise correlation
Migration Matrix Example: Extract of an migration matrix based on Moodys S&P data (1 year migration probabilities) Bank s approaches Estimation based on data bases (Moody s DRS or S&P Credit Pro) Example: Lando (2002) Direct usage of published matrices Scaling by matrix generator Findings Representativeness of market data for banks portfolio Granularity not appropriate (e. g. no distinction between gov and corp)
(Re-)Valuation under Rating Migration Example: Re-Valuation for BBB Instrument Banks approaches All banks use valuation grids Variety of approaches Use of credit spreads (current or mean) Use of default intensities (current or mean) Findings Simplification of valuation approaches Appropriateness of fallback solutions Need for fallback solutions due to lack of market data!
(Re-)Valuation under Default Example: Moody s DRS database Bank s approaches Direct usage of Market data (e.g. Markit) Expert judgment or IRB Based on databases Stochastic and deterministic RR Findings Inconsistencies, simplifications Market data is rare, impact on IRC numbers is huge! Appropriateness of fallback solutions
Liquidity Horizon Example: 2 Trades, 3m and 6m LH Banks approaches Set everything to 12m Draw correlated factors on 3m bases, replace migrated positions by clones with original rating Possibilities to replace migrated or defaulted positions Same or new issues (same or new idiosyncratic Factor) ->Different level of diversification Reset systematic factors or keep values Findings Medium impact on IRC numbers: 3m->12m means increase of 10 to 20% Liquidity assessment (appropriateness of process) Replacing of positions
Validation Banks approaches Analysis of basic model assumptions Sensitivities and scenario analysis (assess impact on IRC numbers) Use of different calibration data to assess model error Use of stress tests to assess appropriateness of IRC numbers (e. g. default of one or two of most important issuers) Findings No validation concept in place Basic model assumptions not validated
Use Test Banks approaches Inclusion of IRC numbers in regular reporting Reporting of overall numbers, stand alone numbers for relevant portfolios, stand alone for relevant issuers, contributions No limit setting on IRC numbers so far No Findings Regulatory requirements not fixed yet. Use test not in focus of inspections Deutsche Bundesbank conducted fact finding only
Additional Findings Documentation Rating assignment process Insufficient update and review process Appropriateness of fallback solutions Modeling of funds, lack of look-through Assigning a lower rating is not always conservative! Look-through is even more important for IRC!
Interesting Issues Loss distribution with one dominating issuer: Error of R² estimation: (mean in red, quantiles (15%, 85%) in green) Impact of rating migration compared to default: 7-20% of IRC number IRC compared to current market risk number (10 day VaR): 2,8-6,5
Conclusion (in My Personal View) Many issues with position capturing Can be fixed easily. Major challenges for the implementation of the IRC-model are Insufficient market data for calibration of factor model calibration of recovery rates bond valuation after severe downgrade calibration of rating matrices Model Validation Not easy to solve! no backtesting available, appropriateness of model outcome difficult to assess Alternatives needed like benchmarking, but: All banks are using the same approach!
Conclusion (in My Personal View) High capital charge (compared to current market risk), but: Benefit for (daily) risk management under discussion Limitation? Useful supplement to current VaR model with respect to issuer concentration Many issues found, but: Strong dependency on Rating Agencies How can overall model soundness be ensured? Next steps? Regulatory requirements for use test Analyzing strengths and weaknesses of different approaches
Session Overview 1. Setting Banks, Audit Approach 2. Results IRC 3. Results CRM 4. Results Stress VaR
CRM requirements Mandatory for preliminary approval Primary focus of on-site examinations Complete and adequate capturing of credit exposure positions Appropriate stochastic modeling, one-year-horizon, 99.9% probability All price risks Appropriate parameterization (e. g. segmentation, correlation) Qualified model validation Process to assess eligibility of trades for Correlation Trading Portfolio Regulatory Stress Testing Not in Focus: Standardized Approach (Floor Calculation) Similar to IRC Additional for CRM
Process to assess eligibility Two types of criteria: Clear requirements Static (instrument features): Assessed once, stable during lifetime of an instrument Dynamic (liquidity of a trade): Frequent assessment needed Approaches for liquidity assessment: Interpretation necessary Combination of several liquidity reports (e. g. DTCC (American Clearinghouse) quarterly report, Markit CDS liquidity report) and expert judgment Findings Quarterly report requested! Robustness of processes Soundness of criteria, appropriate combination of criteria Short-term nature of assessment, volatile CTP composition
CRM Modelling Approaches Correlations: How to aggregate risk factors? Bank 1: Based on existing approaches Historical Correlations Re-Use of path generators from Internal Model Method (IMM) for Counterparty Credit Risk, Risks are Credit Spreads, FX, Interest Rates Uncorrelated (supported by quantitative analysis) Additional risk factors like Base Correlations, Recovery Rates and Index Bases Uncorrelated (supported by quantitative analysis) Bank 2: Re-Use of IRC approach (Extension: Stochastic Recovery Rates) plus Credit Spreads and Base Correlations with Mean Reversion Processes Both banks implemented a modular approach (in line with Basel interpretive issues).
CRM Modelling Approaches Findings Simplifications not justified, e. g. Correlation assumptions not sufficiently supported by market data (for some combinations of risk factor classes) Only 5 year credit spreads etc. Missing risk factors (FX and interest rate risk) Inappropriate update frequency of parameters (for the mean reversion processes) Simplifications in pricing models Monte Carlo with full revaluation to fulfill the cross gamma requirement
CRM Stress Tests Quarterly report requested! No severe Findings regarding Stress Tests. Both banks implemented regulatory Stress Test (based on Basel Stress Test Guidance). Are the Stress Tests useful to review own fund requirements? 600 500 Regulatory CTP 250 200 Extended CTP Credit Risk Balanced 400 150 300 200 100 100 50 0 date 1 date 2 date 3 date 4 0 date 1 date 2 date 3 date 4 CRM Stress Test CRM Stress Test Most severe regulatory Stress Test compared to CRM risk number (99.9%, 1 Year)
Conclusion (in My Personal View) similar to IRC High capital charge (compared to contribution of CTP to VaR model), but: Benefit for (daily) risk management under discussion CTP has extraordinary regulatory attention! Many issues found, but: How can overall model soundness be ensured? Next steps? Follow-up inspections on model validation (including backtesting) Regulatory requirements for use test Analyzing strengths and weaknesses of different approaches
Session Overview 1. Setting Banks, Audit Approach 2. Results IRC 3. Results CRM 4. Results Stress VaR
Stress VaR requirements No On-site Inspections on Stress VaR so far! Assumption: Risk model at least principally capable of calculating stressed VaR Necessary for the approval: Suitable quantitative process for the choice of the stress period Adequate process for the calculation of the Stress VaR providing proper market data of the stress period Appropriate proxy concept Starting from 2011: Integration of Stress VaR into normal scope of model audit plan Stress VaR subject to routine on-site inspections
Lehman- Insolvency 4 3.5 svar/var Stress VaR Results 3 2.5 2 Reason for the variation not fully understood. 1.5 1 0.5 0 Error bars indicate Min/Max in five weeks Observation Period Not acceptable! Bank 11 Bank 10 Bank 9 Bank 8 Bank 7 Bank 6 Bank 5 Bank 4 Bank 3 Bank 2 Bank 1 01.12.2009 23.08.2009 15.05.2009 04.02.2009 27.10.2008 19.07.2008 10.04.2008 01.01.2008
Stress VaR: Choice of observation period Approaches For Historical Simulation models: Time window that contains the three most severe losses for the current portfolio composition Parametric models: Choose Maximum VaR for historical parameterizations, daily, weekly, monthly, quarterly calculations Two Step approaches Pre-selection of possible time window: Restrict analysis to most important risk factors Take into account all risk factors in the final selection Findings Explored time window to narrow, e. g. restricted to financial crises Missing relation to current portfolio, e.g. simplified portfolio representation But: What should a bank do if 1998 is the most stressful year?
Conclusion (in My Personal View) The Stress VaR concept is an appropriate quick fix to the market risk model: Provides persistent memory of critical market situations ->Dampens cyclicality of model based capital requirements Easy and quick implementation But: Lack of market data will increase! No solution yet! Solution could be to do historical model calibration only for systemic risk factors, but: It doesn t get boring! Let s wait for the fundamental review of market risk models
The End Thank you for your attention! Contact details: email: ingo.reichwein@bafin.de phone: +49 228 4108 1852