Session 190 PD, Model Risk Management and Controls
Moderator:
Chad R. Runchey, FSA, MAAA
Presenters:
Michael N. Failor, ASA, MAAA
Michael A. McDonald, FSA, FCIA
System Access and Change Control
SOA 2014 Annual Meeting
Agenda
Systems vs. Models
Reasons for System Access and Change Control Change Control Components
Leading Practices
Systems Development Life Cycle Spreadsheet Specific Issues
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control Change Control Components
Leading Practices
Systems Development Life Cycle Spreadsheet Specific Issues
Systems vs. Models
Characteristics of Actuarial Systems
Systems often contain a myriad of options and settings in which to build and run actuarial models
Systems provide the underlying code (actuarial formulas) needed to run projection models
– “Open systems” provide users the ability to modify system code / formulas – “Closed systems” prevent user-level changes to underlying system code
System coding errors can affect every model built on the system (similar to DNA coding errors)
Systems vs. Models (cont.)
Characteristics of Actuarial Models
Often run on projection systems
Defined by selecting options, settings and inputs accommodated by the underlying projection system (similar to genes in DNA)
Models may be independent of each other
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control
Change Control Components Leading Practices
Systems Development Life Cycle Spreadsheet Specific Issues
Reasons for System Access and Change Control
Modeling Demands (“givens”)
Models will continue to evolve
Models must be quickly modified to remain viable
Model flexibility cannot be at the expense of controls
Reasons for System Access and Change Control (cont.)
Modeling Demands → Increased Systems Development
More code development in more applications
– Vender provided “open code” actuarial projection systems – Use of Excel (VBA)
– SQL interfaces – C++
– SAS
Some major drivers
– Stochastic modeling – Solvency II
– VM-20
– System consolidations – Product creativity
Actuarial Modeling Controls:
A Survey of Actuarial Modeling Controls
in the Context of a Model-Based Valuation
Framework
(December 2012)
Reasons for System Access and Change Control (cont.)
Sponsored by the SOA
Research and Analysis by Deloitte
SCOR Participated in the Project Oversight Group
Compared Current State of Controls Against
Those Expected To Be in Place for Model-Based Valuation (MBV) Approaches
Modeling Governance Theme Score Current State Synopsis Governance Standards 3 While many companies employ a variety of model governance policies, few companies have a holistic, formal and documented model governance structure. General Modeling Process 3 Many companies have multiple models and modeling platforms and few companies incorporate a model steward role in the modeling processes. System Access and Change Control 4 Model changes are not generally governed by a formal change process. Model Assumption Management 3 Assumptions are regularly reviewed and updated, but with few controls in place to ensure assumptions are approved and input appropriately. Model Input Management 2 Many companies use automated feeds from admin systems for model inputs of liabilities. Other model inputs are often less automated.
Model Output Management 2 Model output used for financial reporting purposes is generally well controlled, while model output for analysis and other purposes is generally less controlled.
On a scale of 1 to 5 (1 being the best), system change control received a “4” – the worst score of all the categories. Note that system coding changes fall under this category.
Reasons for System Access and Change Control (cont.)
Majority of the survey respondents had no formal change control process
Under system access and change control, the SOA sponsored report
recommends two key next steps to move from the current state to leading practices:
“Implement a formal change management process for governing model code changes and model updates”
“Determine the calendar for internal model releases to ensure consistency of the model of record across the organization”
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control
Change Control Components
Leading Practices
Systems Development Life Cycle Spreadsheet Specific Issues
Change Control Components
Code Changes Are Often Performed by Actuarial Staff
IT involvement beneficial
Adoption and modification of existing IT control processes
Code Comparison Tools
May be provided in modeling system software
Third party code comparison tools work with VBA (Excel and MS-Access)
Code Changes Should Be Peer Reviewed
Rigor varies from company to company
SOA 2013 Valuation Actuary Symposium, Session 14 (Reviewing and Validating Actuarial Models), “Systems Peer Review” presentation
Change Control Components (cont.)
Model Steward
Must have top-down support
Roles vary from organization to organization
Secures production models
Archives models
– When changes are made – Periodic scheduled archival
Applies version controls
Assures that procedures have been properly followed
Change Control Components (cont.)
Clearly Defined System Access Controls
Levels
– Run model – Read only
– Change model inputs – Change code
Locations
– Production grids – Testing servers
Models and Systems
– Actuarial projection systems – Specific models
– Spreadsheets – Data
Change Control Components (cont.)
Test Beds (or Test Packs)
Need to be kept up to date
Test beds have their limitations
Management / Departmental Approvals
Valuation
Pricing
Risk Management
Others
Communication and Documentation
Start to finish
High-level descriptions down to comments in the code
Archived and linked to: – Production models – Test results
Change Control Components (cont.)
Different Modeling Environments
Centralized models
– Model of record is maintained
– Simultaneous changes by different coders
– Requires a formal code aggregation process
– Integration testing
Non-Centralized models
– No single model of record
– Pricing models vs. valuation models
– Potentially divergent modeling results
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control Change Control Components
Leading Practices
Systems Development Life Cycle Spreadsheet Specific Issues
Leading Practices*
Typically Have Four Different Steps in the Control Process
1. Establish procedure to identify and prioritize model changes (change request process)
2. Evaluate coding changes in a test environment and analyze impact on financial results
3. Perform additional testing on the model code changes (depending on nature of model change)
– Regression testing – Sensitivity testing – Peer reviews
4. Document and seek formal approval
– Document changes and rationale, results and discussion, any sensitivity tests – Appropriate parties review documentation and either approve or deny release – If approved, schedule system release for production use (try to coordinate
Leading Practices* (cont.)
Modeling Teams Responsible for Managing Change Requests
Determine change request procedures
Maintain and publish logs of all change requests
Prioritize change requests
IT Involvement in the Code Change Process
For open code systems and centralized models
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control Change Control Components
Leading Practices
Systems Development Life Cycle
Systems Development Life Cycle
What Is a Systems Development Life Cycle (SDLC)?
Clearly defined process for planning, creating, testing, and deploying systems and systems modifications.
SDLC Benefits
Risk reduction
Well defined roles and responsibilities
Improved communication and documentation
Auditability
Are Actuaries Employing SDLC Methods?
Systems Development Life Cycle (cont.)
Change Request Log
Vital for any SDLC process
Formalized documentation of requested system changes – Reasons for the change?
– Is it a system correction or a system enhancement? – Date when change needs to be in production?
– Consequences if change is not made?
– Other relevant details (i.e., change request ID, date, etc.)
Each change request is assigned a priority level based on the above
Regression T est BUILD Unit Test Technical Specification Functional and
System Design System Test
Business
Requirements User Acceptance Test
Systems Development Life Cycle – ‘V’ Methodology* (cont.)
Business Requirements (BR)
Produced by the actuaries requesting the change
Details the required changes
Includes calculation methodologies (i.e., reserve requirements)
Functional and System Design
Describes model design and structure as a solution to the BR
Can be a source for an actuarial programmer to go directly to code
Technical Specification
Documentation at a more detailed “coding” level
Systems Development Life Cycle – ‘V’ Methodology* (cont.)
Unit Testing
Ties back to technical specification
Low-level testing of code segments in isolation
System Testing
Ties back to functional and system design
Testing on a single policy or model point
Scenario testing
End-to-end testing
Stress testing
User Acceptance Testing (UAT)
Ties back to business requirements
Run in production mode
Checks for reasonableness of results
Regression Testing
System Access and Change Control
Systems vs. Models
Reasons for System Access and Change Control Change Control Components
Leading Practices
Systems Development Life Cycle
Spreadsheet Specific Issues
At the Far End of the “Flexibility vs. Control” Paradigm
Reduce the Number of Spreadsheets
Periodic Archival
Inventory and Document Workbook Components
VBA / SQL code
How to run VBA / macros and use spreadsheet
Input data sources and links
Description of outputs
Microsoft add-ins
– Random number generators – @Risk
Spreadsheet Specific Issues (cont.)
Apply Workbook Access Controls
Need to lock down models
Limit directory access
Protect Workbooks
Apply Change Controls
Workbook versioning
VBA
– Code management methods apply – Third party code comparison tools
Spreadsheet Specific Issues (cont.)
Additional Testing Concerns
Workbook testing should be Excel version specific
VBA testing may be affected by Excel session settings
Common Risks
Difficult to analyze large complex spreadsheets
Key person risk
Model Risk Management and
Controls
Model Risk Management
Chad Runchey, FSA MAAA
Ernst & Young LLP
Model risk management
•
Background
•
Drivers of model risk
•
Model risk management framework
What is model risk?
… possible adverse consequences (including financial loss) of
decisions based on models that are incorrect or misused. – Federal
Reserve Supervisory Letter 11-7
•
Can have many causes:
• Design errors
• Inappropriate or incorrect use • Model calculation errors
• Poor data
• Misunderstood or poorly communicated assumptions or results
•
Increases with greater model complexity, input/assumption
What is driving the increased
focus?
•
More model-driven financial reporting and
analysis
•
Regulatory environment
• Solvency II – internal capital models
• Federal Reserve – comprehensive scope
• NAIC – limited requirements
Model risk management framework
Pr oc es s & C on tr ol s Enterprise Governance M ode l G ov er na nc e Model LifecycleModel Review Components
1st Line of Defense
(Model Owners) 2
nd Line of Defense (Model Governance & Validation)
.
Roles and responsibilities
• 1stline: Owns model
lifecycle and related activities
• 2ndline: Establishes
model risk management framework and
standards, provides review and challenge, and can perform independent model testing
• 3rdline: Performs
independent testing and verification of 1st
and 2ndline activities
Co nc ep tua l S oundne ss Da ta & IT In fr as tr uc tu re Model Development Model Implementation Model Operation Performance Monitoring Model Change Management
Business Purpose Inde pe nde nt T es ting
Model risk policy
•
A policy formalizes the framework and clearly
states the roles and responsibilities of each
group within the organization.
•
Other key components of a policy include:
• Model definition
• Risk-rating approach
• Documentation standards
First line of defense: Model owners
• Business purpose
• Model development and implementation
• Ensure design, theory and logic underlying model is supportable and documented
• Ensure that model components work as intended, are appropriate for intended business purpose, and are conceptually sound and mathematically and statistically correct
• Compare with alternative theories and approaches
• Demonstrate data is suitable for model and consistent with theory behind approach • Test model’s accuracy, robustness and stability, assessing potential limitations and
documenting results
• Model operation
• Performance monitoring
• Assess performance over time under various conditions and as model applications change • Gather insights from observation of outcomes and incorporate into feedback loop
Second line of defense:
Independent risk management
• Model Governance
• Assess model access rules
• Assess change management protocols
• Conceptual Soundness
• Ensure model methodology is theoretically correct and fit for purpose • Ensure applicability of assumptions and other inputs
• Data & IT Infrastructure
• Assess applicability, accuracy, completeness and availability of all input data, including data mapping and transformation procedures
• Assess the sufficiency of the IT infrastructure
• Process and Controls
Common challenges
•
Governance and policy
•
Model definition
•
Model validation
• Management expectations
• Independence
• Sustainability
•
Model documentation
Contact information
Chad Runchey
(212) 773-1015
Session 190 PD, Model Risk Management and
Controls
Michael McDonald
SOA 2014 Annual Meeting
Overview
•
Model Validation
•
Model Controls
•
Current Trends in Valuation Transformations
•
Useful Links
Overview
•
Model Validation
•
Model Controls
•
Current Trends in Valuation Transformations
•
Useful Links
Model Validation
Why is it needed?
•
Reserve and capital valuations are moving to model-based
approaches
•
Complexity of models has increased likelihood of material
modeling errors
•
Increased scrutiny from regulators and senior management
•
Basis for Actuarial analysis of inforce exposure
Modeling methodology
Calculations
Model Validation
Model Review
Model Validation
Model Review
•
Identify model setup deficiencies vs. best practices
•
Consistency
•
Transparency
•
Organization
• Intuitive object/variable names
• Hard coded values are stored in central location
•
Well documented
Model Validation
Model Vetting
•
Validate Model results
•
Vetting tools should be independent of model
•
Build simplified calculator
•
Compare results of simplified scenario for reasonability
•
Build robust calculator for interim calculations
•
Projected mortality calculation
•
Projected NAAR calculation
•
Dynamic lapse assumption
Model Validation
Sensitivity Analysis
•
Review against pricing model and/or historical
sensitivities
SENSITIVITY RECONCILIATION Pricing Sensitivity x Differences Inforce Distribution a Inforce Aging b Actuarial Assumptions c Economic Assumptions dModel Validation
Sensitivity Analysis
•
Review against pricing model and/or historical
sensitivities
SENSITIVITY RECONCILIATION Pricing Sensitivity x Differences Inforce Distribution a Inforce Aging b Actuarial Assumptions c Economic Assumptions d Unknown Differences eModel Validation
Case Testing
•
Roll forward inforce – compare reserves vs baseline
projected
time
Extract aged for 5 years Inforce Aging Process
Model Validation
Case Testing
•
Stress-test dynamic assumptions
•
Single policy – unique attributes
Overview
•
Model Validation
•
Model Controls
•
Current Trends in Valuation Transformations
•
Useful Links
Model Controls
•
Documentation
•
Model User manual/Documentation manual
•
Internal model documentation
•
Issue log
•
Version Controls
•
Linear development
•
Standardized model naming
•
Model comparison tools
Model Controls
•
Model Development Validation
•
Change summary
•
Testing plan with expected impact
•
Change management checklist
•
Peer review
•
Impact quantification
Model Controls
•
Inclusion control
•
User access
•
Important for both production and development
environment
•
Consolidate calculations into one model
•
Eliminate spreadsheet models!
•
Too flexible
Overview
•
Model Validation
•
Model Controls
•
Current Trends in Valuation Transformations
•
Useful Links
Valuation Transformation
•
Legacy Design
•
Extracts and assumptions reside in business function
area
•
Model output consolidated in back-end spreadsheets
•
Next Generation Design
•
All model assumptions reside in external databases (pricing
and valuation)
•
Models reference tables populated with admin data through
ADMIN EXPERIENCE MONITORING Policy Data Queries Analysis PRICING Model Hypothetical Policy Data Assumptions Product features Actuarial VALUATION Model Policy Data Assumptions Product features Actuarial Feedback Feedback FINANCE PRICING TEAM
Extract Extract Extract Extract
Historic Versioning FINANCIAL REPORTING SYSTEM • Consolidates, organizes, aggregates financial results DATA WAREHOUSE Policy Info Claims/Pre miums/Exp enses Pricing Distribution ADMIN FINANCE ASSUMPTION DATABASE Product Features Actuarial Assumptions Scenario s EXPERIENCE MONITORING Queries Analysis PRICING Model VALUATION Model PRICING TEAM
Next Generation Illustration
Next Generation Illustration
FINANCIAL REPORTING SYSTEM DATA WAREHOUSE Policy Info Claims/Prem iums/Expens es Pricing Distribution ASSUMPTION DATABASE Product Features Actuarial Assumptions ScenariosValuation Transformation
•
Robust automated controls and processes
•
Actuarial staff can focus time on analyzing results
•
Non modeling staff can review model assumptions
•
More granular model output available for analysis (ie.
seriatim projections)
•
Financial statement info from single source
•
Increased transparency
Overview
•
Model Validation
•
Model Controls
•
Current Trends in Valuation Transformations
•
Useful Links
Useful Links
SOA Survey – Model Control
http://www.soa.org/Research/Research-Projects/Life-Insurance/Actuarial-Modeling-Control.aspx
EU Study – Model Processes and Controls
http://www.google.com/url?sa=t&rct=j&q=&esrc=s&frm=1&source=web&cd=6 &ved=0CDsQFjAF&url=http%3A%2F%2Fwww.actuaries.org.uk%2Fsystem%2F files%2Fdocuments%2Fpdf%2Factuarialprocessescontrolswplife200905rpt.pdf&ei =qpPzU_C7GrjLsQS29IDIAg&usg=AFQjCNHhVm8nlNxsrg5Hyjzwod0XTUS Sjw&sig2=mWsOJGTI1aq6un5CWkE43w