Model Risk Management: IACPM Fintegral 2016 Benchmarking Results Jörg Behrens, Kai Pohl Dublin, 19 May 2016
Statisticians, like artists, have a bad habit of falling in love with their models George Box, Statistici 2016 fintegral consulting www.fintegral.com 2
This Presentation Thank you and Juliane / IACPM: 54 out of 56 firms responding! Your survey, questions, answers Identification of Model Risk as relevant topic Survey devised jointly with IACPM MRM steering committee Analysis of responses and follow up Major findings and some surprises Thoughts for future Thanks to Kai Pohl have a speedy recovery! 2016 fintegral consulting www.fintegral.com 3
Rationale of the Survey The Rise of Model Risk Management Models underpin the modern financial system Regulation, mostly triggered by OCC SR 11-7 Paradigm shift from mathematics to model application and management Few people are modelers but everyone has an interest Don t wait for regulators or auditors to tell you how to manage your models... 2016 fintegral consulting www.fintegral.com 4
Billions USD No. Respondents Survey Demographics 54 financial institutions participated in the survey Share by Region Share by Number of Employees 30 30% 54% 16% Americas Asia Europe 20 10 0 <30 30-80 80-150 >150 1,000 s of Employees Share by Balance Sheet Size Share by Nature of Institution >500 300-500 200-300 100-200 50-100 <50 4% 10% 10% 76% Bank / Investment Bank Insurer ECA/IFI Other 0% 20% 40% 60% 2016 fintegral consulting www.fintegral.com 5
and relevant Peer Groups Global Systemically Important Banks (G-SIB) vs non G-SIB Used as a proxy for size. G-SIB institutions are generally larger (17 G-SIB s). CCAR vs non CCAR Regulatory pressure is considered to be highest in the US (23 CCAR banks). Geographic location Americas, Europe and Asia. Type of financial institution Bank, insurance, supranational, corporate. 2016 fintegral consulting www.fintegral.com 6
Model Validation vs. Model Risk Management Tasks & Ownership Model risk validation: model build and parameterisation MRM: focus on processes around model use Management Separation between model risk management and validation fluid Unclear or inconsistent split of tasks between teams or team hierarchy Management owned by Model development team (78%) Independent MRM team (69%) Business unit (63%)? Measurement 0% 20% 40% 60% 80% 100% Development / Business MRM Validation Audit 1.1. Model Risk Validation established but not always clearly separated from -management 2.2. Model Risk Management not clearly associated with 3-lines-of-defence-model 3.3. Possibly conflicting views between pure CPM vs group risk functions 2016 fintegral consulting www.fintegral.com 7
Scope of Model Risk Management Scope of MRM ranges from credit (the obvious) to less obvious such as fraud, marketing or HR processes Review Activity of MRM Credit risk models Market risk models Operational risk models Model overlays Tactical tools Other 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1. Selection bias: major focus on credit risk 2. Awareness about interlinkage between Market/Credit/OpRIsk (Capital, Risk Appetite...) 3. Model overlays/overrides: judgemental components: key problem for model risk! 2016 fintegral consulting www.fintegral.com 8
Relative Significance of Model Risk Types Easily measured model risks are more thoroughly analysed Types of analysed model risks Data Implementation Statistical Parameters 87.0% 85.2% 85.2% 85.2% 7.4% 5.6% 9.3% 7.4% Misuse 64.8% 13.0% User/process Interpretation 63.0% 61.1% 5.6% 14.8% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Currently Next 3 years 1.1. Historically: main focus of model risk on implementation check (benchmarking prototype vs system) 2.2. Heavy focus on quantitative points (3 of 4) 3.3. New focus on qualitative components (lower 3 categories) 2016 fintegral consulting www.fintegral.com 9
Impact of Model Risk and Capital Buffer On a scale of 1 (low) to 4 (very high), participants assessed the impact of model risks on different categories US banks tend to include a capital buffer for model risk 100% 80% 60% 40% 20% Capital Buffer 0% Americas Europe Asia 1.1. CCAR enforces good MRM to avoid extra capital charge 2.2. Geography more important than CCAR vs non-ccar 3.3. Participants more concerned about impact of risks on P&L and Reg Cap than on reputation 2016 fintegral consulting www.fintegral.com 10
Understanding of Model Risk Considered either good or moderate US regulated banks tend to feel their resources are adequate, however... Judged only on size, large banks tend to view their resources as rather inadequate Understanding of Model Risk Adequate Resources? Adequate Resources? 16 Adequate Resources? 35 40 14 13% 39% 9% 37% Very good Good Moderate Beginner 30 25 20 15 10 5 35 30 25 20 15 10 5 12 10 8 6 4 2 0 Non CCAR CCAR 0 Non G-SIB G-SIB 0 CCAR / Non G-SIB Non-CCAR / G-SIB Yes No Yes No Yes No 1.1. MRM at CCAR banks might benefit from significant CCAR budgets 2.2. G-SIBs seem to be more realistic about their (lack of) MRM resources 3.3. CCAR non G-SIB seem to be prepared for CCAR really? 2016 fintegral consulting www.fintegral.com 11
Measurement of Model Risk Statistical model risk and assumption based model risk can be quantified. «Other» mainly refers to scorecard approaches / hybrids. Sensitivity Analyses Back Testing Benchmarking We do not quantify model risk Other 0% 10% 20% 30% 40% 50% 60% 70% 1. 1. 85% of the firms claim to quantify model risk (top 3)... 2. 2. Maybe PD focus due to IACPM bias 2016 fintegral consulting www.fintegral.com 12
Model Risk Aggregation Risk Appetite for Model Risk? In spirit of OpRisk (SREP, e.g. Defined in terms of failures) No standard procedure for the aggregation of model risk exists Simple techniques facilitate interpretation and implementation. Advanced approaches can be more precise. Aggregation Techniques Simple addition error contributions Pass-through sensitivity Error propagation techniques Other Theory: error propagation 0% 5% 10% 15% 20% 25% Practice: pragmatic approaches including summation (simple or sensitivity based) 2016 fintegral consulting www.fintegral.com 13
and Link to Risk Appetite About 39% firms have defined a Risk Appetite for Model Risk (wow!) More than that, i.e. 46%, do aggregate model risk Definition of Model Risk Appetite Aggregation of Model Risk Non CCAR 4 26 Non CCAR 10 21 CCAR 7 17 CCAR 9 15 0 5 10 15 20 25 30 No model risk appetite Model risk appetite defined 0 5 10 15 20 25 No aggregation of model risk Aggregation of model risk Easy task for 7% to link the two? 1. Over 70% of CCAR banks define a risk appetite level, but only a small minority of non-ccar institutions (mainly through score cards) 2. Scope probably narrow and focussed on particular model type (e.g. PD mode), not total capital 2016 fintegral consulting www.fintegral.com 14
Thoughts for future MRM So far so good: Clarify what is model risk management Get some insight into what others are doing Soften shock caused by MRM regulatory requirements Further points of interest: Improving identification of unmanaged model risks (statistical, procedural, assumption) Improve model risk aggregation and link to risk appetite Develop and defend thoughtful approach to MRM Actively develop your MRM approach and do not wait for the regulator to tell you how to manage your models 2016 fintegral consulting www.fintegral.com 15
Fintegral Consulting London Zürich Frankfurt consulting@fintegral.com www.fintegral.com Dr. Jörg Behrens Managing Partner - Direct: +41 44 552 1106 E-Mail: joerg.behrens@fintegral.com - Fintegral Consulting AG Brandschenkestrasse 150 CH - 8002 Zürich Kai Pohl Director London Office - Direct: +44 7753 456 894 E-Mail: kai.pohl@fintegral.com - Fintegral Consulting Ltd 23 Austin Friars London EN2N 2QP