Transcription
The Model Risk Manager's andModel Validator's ToolboxMathWorks Computational Finance ConferencePaul PeelingSeptember 27th 20211
MathWorks helps you manage model risk with a platform andtechnology for your entire organization.Model GovernanceModel ValidationModel DevOps2
Many teams, users and stakeholders collaborate to bring amodel from research to production.The 1st and 2nd lines ofdefence have welldefined roles andresponsibilities.The business, quantsand IT are involvedthroughout.3
It is difficult to reach a sustainable and cost-effective MRMstrategy if tools and processes are not coordinated. Poor quality models Regulatory scrutiny High cost Inconsistency Frustrated users Low automation4
Risk Management is a complex system with interconnected parts.5
The MathWorks Model Risk Management Solution supports all users andevery step of a model’s lifecycle.Model Inventory & Repository (MIR)Centralized access to models, lineage, audit trail, risk scoring, andmodel risk reportingMMDMDEModel Development Environment (MDE)MIRMEEMRE-Explore, develop, back-test, and document models andmethodologies-Improve transparency and reproducibility of model developmentprocess-Create reusable model templates-Auto-generate model documentationMTVEModel Review Environment (MRE)-Perform independent model reviews-Perform interactive what-if and sensitivity analysis on modelparameters-Comment and flag various aspects for response and resolution6
Monitoring andPerformanceAssessmentModel InventoryDeploymentandImplementationDefinition andDevelopmentModelInventoryReview andValidationBuild and Test,QualityAssurance7
The Model Inventory is the point of entry, showing the completemodel landscape across business lines.Centralized access tomodel riskmanagement data andprocesses throughMATLAB Online Server.Customizable views,providing aggregatedand drill-downinformation.8
Every model is tracked, linked to code and documentation, andinformation is maintained as the model evolves.Customizable fieldsand links to externaldatabases.Workflows for modelcreation, review anddeployment.Integrated with codeand document controlsystems.9
The Model Inventory is the centralized application to perform allmodel risk management activities.Risk TieringImpact AnalysisPlanning10
Monitoring andPerformanceAssessmentModel efinition andDevelopmentModelInventoryReview andValidationBuild and Test,QualityAssurance11
The Model Development Environment produces documentationas you explore data and build models.Richly annotated codeas a basis fordocumentation.Interactive controls andvisualizationspromoting modelinsight and challenge.12
We provide reusable and customizable templates for every stepof the model development processScript Snippets andLive Tasks coveringevery step.Project structure andWord templates.Automation,consistency, reusabilityof model artefacts.13
Candidate models are trained, compared and calibrated in theExperiment Manager.Reproducibility ofmodel builds.Tracking of validationmetrics and annotationof results fordocumentation.Encompass existingworkflows aroundlearner and modellingApps.14
Monitoring andPerformanceAssessmentModel tion andDevelopmentModelInventoryReview andValidationBuild and Test,QualityAssurance15
Developed models are submitted through the Model ReviewEnvironment to be assessed and approved.Access to up-to-datemodel code anddocumentation througha browser (MATLABOnline Server).Model analysis can beexecuted in-place tosupport een 1st and 2ndlines of defence.16
Quantitative information required for internal and regulatorydocumentation is automatically produced.Information is producedby running quantitativetests automatically onmodels.Supplementalinformation populatedfrom inventory andmodel documentation.17
Monitoring andPerformanceAssessmentModel on andDevelopmentModel16InventoryReview andValidationBuild and Test,QualityAssurance18
Rigor and trust in models is established through a Model TestEnvironment accessible through CI/CDQuantitative unit andperformance test suitescovering regulatoryreporting requirements.Interoperability withPython and Jupyter.19
Monitoring andPerformanceAssessmentModel inition andDevelopmentModelInventoryReview andValidationBuild and Test,QualityAssurance20
Approved and tested models are deployed to production with aREST API supporting discovery end execution.Authorize and auditmodel usage.Horizontally andvertically scalable.Integrate with businesssystems with no recoding.Language andimplementationagnostic.21
Monitoring andPerformanceAssessmentModel MonitoringDeploymentandImplementationDefinition andDevelopmentModelInventoryReview andValidationBuild and Test,QualityAssurance22
Metrics used in model review and validation are monitored onproduction models.Build and deploy usingApp Designer andMATLAB Web AppServer.Alerts when metrics falloutside of approvedusage.Dashboards, KPIs andmetrics accessible tomodel users andstakeholders.23
MATLAB seamlessly interoperates with open source and third-partytechnology platforms across the modeling life-cycleCloudData SourcesMonitorDashboardsData feedsBig evelopmentStreamingMATLAB Parallel ServerMATLABOnlineDatabasesMATLAB Production ServerFilesMEETraining, simulation, optimizationMATLAB PlatformModel DevelopmentModel InventoryCI / CDMonitoringMIRMMDAzure DevOpsMDEDomain specific toolboxesRiskDeepLearningData exploration, preprocessing, model developmentReportingModel ReviewMTVESharing, deployment, integrationMATLAB WebApp ServerWeb AppMREModel Validation and Testing24
Key Benefits of MathWorks MRM Solution Unified system of technologies addressing keybusiness, modeling, workflow, and governance needs– Manage model risk with automation and transparency Modeling platform integrated across 1st & 2nd lines ofdefense, covering research to production Eliminate inefficiencies, reduce cost/time Enhance communication Accelerate regulatory approvalmodelriskmanagement@mathworks.comPerform end-to-end modeling faster, better, cheaper25
6 The MathWorks Model Risk Management Solution supports all users and every step of a model's lifecycle.-Explore, develop, back-test, and document models and methodologies-Improve transparency and reproducibility of model development process-Create reusable model templates-Auto-generate model documentationModel Development Environment (MDE)-Perform independent model reviews
