DFA Implementation Issues

  • View

  • Download

Embed Size (px)


DFA Implementation Issues. CAS Risk and Capital Management Seminar Toronto, Ontario, Canada July 8-9, 2002. Speakers. Jen Ehrenfeld, ACAS, MAAA American Re-Insurance Company Gerald Kirschner, FCAS, MAAA Classic Solutions Risk Management, Inc. Elizabeth Wiesner, FCAS, MAAA - PowerPoint PPT Presentation


  • DFA Implementation Issues

    CAS Risk and Capital Management SeminarToronto, Ontario, CanadaJuly 8-9, 2002

  • SpeakersJen Ehrenfeld, ACAS, MAAAAmerican Re-Insurance CompanyGerald Kirschner, FCAS, MAAAClassic Solutions Risk Management, Inc.Elizabeth Wiesner, FCAS, MAAAAccident Fund Company

  • Large Company PerspectiveJen Ehrenfeld

  • DFA Implementation IssuesAgendaLessons Learned withBuilding the ModelParameterizing ModelPresenting ResultsCorporate Structure

  • Building The ModelRight resources in-housePriority of DFA Model cant do this and your day job tooRecognition of Investment Up Front before any Returns

  • Parameterizing ModelCompany Organization vs. Risk\Line SegmentationPolitically Sensitive Issues vs. RealityHighlights areas of concern\Due Diligence on own CompanyIf unable to model risks well, how well are they being monitored and managed.

  • Parameterizing Model - ContinuedRisk = New Planned Loss Plus Historical ReservesPlanned Losses at least as variable as ReservesReserve Modeling Symmetric distribution or Adverse Development more likely

  • Mean vs. Mode vs. MedianPlan Loss Ratio Where does it lie on the distribution?There are three measures of central tendency.1. Mode - the most frequent observation2. Median - the 50th percentile3. Mean - the average value of the distribution

  • Mean vs. Mode vs. MedianSurveying Actuaries All Three Given as AnswersExample: Plan Loss Ratio = 0.75Standard Deviation around Plan = 0.30Clearly, Sensitive to Standard Deviation

  • Mean vs. Mode vs. Median

  • Correlation Where to start?Empirical Measures of Correlation Dont Result in Intuitive AnswersInitial Sensitivity Testing Anything below 15% - Basically IndependentAnything above 50% - Basically DependentWe let intuition take over.

  • Correlation ContinuedFor each Line of Business Combo:Selected Correlation LevelIndependent 0%Ex: PR Catastrophes with HealthcareLow - 15%Ex: WC with International LinesMedium 30%Ex: Different Lines written within same Organizational Div High 50%Ex: Same Line different types WC Treaty & WC Fac

  • Correlation ContinuedImportant to do Follow-Up Sensitivity TestingImportance of Correlation to StudyWhat if all lines 100% IndependentWhat if all lines 100% DependentMunich counterparts using an average correlation for every line combinationNot sure of implications May be ok for Total Liability analyses, but not for line comparisons.

  • Correlation Continued

  • Correlation ContinuedOnce Correlations are defined, if change detail of analysis, need to adjust correlationsAnswer should stay the same no matter what detail you review the companyBreak down into more detail between line correlations must decrease to get the same overall variance.Gets hairy fast

  • Presenting ResultsREASONABILITY CHECK!!!Cant DFA tell you that???Not too much at once - even for very savvy audiencesDFA, in some respects, may be too much of a leap from current practicesStart simple - Get buy-in Expand AnalysesPlan VariabilityReinsurance Analyses

  • Corporate StructureNeed Dedicated ResourcesFrom time of building model implementationNeed Interaction & Coordination with All Areas of CompanyNeed High Exposure & Support from Top Management Need to be able to have a Tangible Impact

  • Success vs. FailureFailure - After describing a work issue or project, someone instantly is reminded of a Dilbert cartoon.

    Success - Not receiving any Dilbert cartoons for a week.

  • Large Company / Outside Consultant / Software VendorPerspectiveGerald KirschnerClassic Solutions Risk Management, Inc.www.csrmi.com

  • Key item to rememberThere is no silver bullet - A DFA model can NOT do anything and everything

  • Large Company Perspective

  • Internal Obstacles unrealistic timelinesImplementation of a DFA model should be measured in years and not weeks or even months.first year figure out what youre trying to do usually requires a narrowing of scopesecond year improve efficiency of processthird year start adding to the process

  • Internal Obstacles desire to cross tie with other systemsThere may be other systems in the company that are doing valuations or projections of parts of the company, and those may overlap with parts of the DFA model scopeExpecting or demanding that the two tie out in a precise manner may be unrealistic, given the assumptions being used by the different systems

  • Internal Obstacles overly detailed modelingJust because data exists to allow you to model at the nth degree of detail, dont necessarily do it.Where company data does not allow you to create a logical set of assumptions, dont overwork the assumptions that you can make.

  • Getting the consultants involvedcompany preparationknow what you want to accomplish and be realistic as to short and long term goalsbe willing to be flexiblebe ready to invest significant time and resourceshave a small-scale test case that can be used in a trial runtake advantage of your trial run to learn the softwares strengths and weaknesses

  • Outside Consultant / Software Vendor Perspective

  • Getting the consultants involvedconsultant preparationListen first, talk secondAsk lots of questions about company:company expectationsdesired use(s) of modelunique aspects of the companys operations (these will be the ones to challenge your model)Be honest about your model strengths and weaknesses in the long run, an inappropriate sale is worse than no sale

  • Consultant preparation continuedIdentify appropriate contact persons for the client company these should encompass all the areas of expertise upon which the client will need helpBe realistic as to what the client can learn on his/her own and be patient you are the ones who know the model inside and out and the client is going up a steep learning curve

  • Once you are under wayCompanyread the user manualsinvest the time to learn the softwaretell your vendor about software problems you encounterConsultant / Software Vendorcheck in frequently with the clientrein in scope creep

  • Small Company PerspectiveElizabeth Wiesner

  • Business NeedsEverything driven by business needsComplete assessmentCompare alternatives for fit with companyprioritize needscostsdetermine and include other interested parties

  • ExamplesStrategic planningFinancial projectionsCoordination among company operationsLoss reservingManagement trainingCostTime of deliveryComplexity of useThoroughness of model

  • Implementation IssuesBuy/buildTransfer knowledge to internal staffSpread of knowledge internallyKeeping all interested departments involvedKeep it simple

  • Issues with Day to Day UseDesire to use as crystal ballAcceptance throughout companyUnderstandable communicationChanging resultsFollowing a project process and using DFA as a tool, not the project itself

    Very brief comments as most companies have already dealt with this

    How your company is organized vs the lines that you would like to analyze?LatentBalance with auditors once you actually measure it, does it need to be booked/

    Unknown Latent vs latent we already haveLong tail vs Short tail. By ignoring reserves, a long tail line of business looks good.

    Sensitivity issues with AuditorsMaybe not a big deal if this is your standard deviationBut can start to really make a differenceAYSL Example If using this information to help decide where a stop loss should attach or how much premium you should pay for it starts to become significant.

    Probability of beating plan large difference!May find easier to cap the ability to do better than the plan i.e., only 5% better than plan loss ratio

    Negative numbersRelationships that cant understand give business logic toEven have had people say that if you first use years 1 10, then years 2 11, signs changeA positively correlated with B, B positively correlated with C, C negatively correlated with A. Impossible.Probably fairly strong signal that we should be using something besides correlationFor Some analyses the inter-line correlation will be vital for others, it may not be.People will get hung up on every assumption important to know which assumptions matterInstant loss of credibility if you make a silly mistake.It will be hard enough to get buy in even when youve been very thorough with analysis let alone with a crazy result that makes it appear that you dont have a sense of the company or the business.Still a lack of understanding as to what DFA is and can doCorrelations, Amount of Capital needed, What lines to Write

    Initial attempts represented a lot of good work but blew board away and besides planting some seeds NO impact to the companys operations

    Management used to being told that they cant get good info. And now you present them the world. May not be sure where to start.

    Even with small group may need to repeat several timesImportant to get buy in from other areas & have tiesCant be viewed as back-room typesInteraction is key

    Reserving Area working very hard to produce real reserve rangeTied into Cat Management, Operating Areas, Planning