ebook img

Modern Actuarial Risk Theory: Using R PDF

393 Pages·2008·2 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Modern Actuarial Risk Theory: Using R

Modern Actuarial Risk Theory Rob Kaas • Marc Goovaerts Jan Dhaene • Michel Denuit Modern Actuarial Risk Theory Using R Second Edition Professor Rob Kaas Professor Marc Goovaerts UvA / KE UvA / KE Roetersstraat 11 Roetersstraat 11 1018 WB Amsterdam 1018 WB Amsterdam The Netherlands The Netherlands [email protected] K.U. Leuven University of Amsterdam Naamsestraat 69 Professor Jan Dhaene 3000 Leuven AFI (Accountancy, Finance, Insurance) Belgium Research Group [email protected] Faculteit Economie en Bedrijfswetenschappen K.U. Leuven University of Amsterdam Professor Michel Denuit Naamsestraat 69 Institut de Statistique 3000 Leuven Université Catholique de Louvain Belgium Voie du Roman Pays 20 [email protected] 1348 Louvain-la-Neuve Belgium [email protected] ISBN: 978-3-540-70992-3 e-ISBN: 978-3-540-70998-5 Library of Congress Control Number: 2008931511 © 2008 Springer-Verlag Berlin Heidelberg This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifi cally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfi lm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag. Violations are liable for prosecution under the German Copyright Law. The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of a specifi c statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: WMXDesign GmbH, Heidelberg Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com Foreword to the First Edition Studythepastifyouwoulddefinethefuture— Confucius,551BC-479BC RiskTheoryhasbeenidentifiedandrecognizedasanimportantpartofactuarialed- ucation;thisisforexampledocumentedbytheSyllabusoftheSocietyofActuaries andbytherecommendationsoftheGroupeConsultatif.Henceitisdesirabletohave adiversityoftextbooksinthisarea. This text in risk theory is original in several respects. In the language of figure skating or gymnastics, the text has two parts, the compulsory part and the free- stylepart.ThecompulsorypartincludesChapters1–4,whicharecompatiblewith officialmaterialoftheSocietyofActuaries.Thisfeaturemakesthetextalsouseful tostudentswhopreparethemselvesfortheactuarialexams.Otherchaptersaremore of a free-style nature, for example Chapter 7 (Ordering of Risks, a speciality of the authors). And I would like to mention Chapters 9 and 11 in particular. To my knowledge, this is the first text in risk theory with an introduction to Generalized LinearModels. Specialpedagogicaleffortshavebeenmadethroughoutthebook.Theclearlan- guage and the numerous exercises are an example for this. Thus the book can be highlyrecommendedasatextbook. Icongratulatetheauthorstotheirtext,andIwouldliketothankthemalsointhe name of students and teachers that they undertook the effort to translate their text intoEnglish.Iamsurethatthetextwillbesuccessfullyusedinmanyclassrooms. Lausanne,2001 HansGerber v Preface to the Second Edition WhenItookoffice,onlyhighenergyphysicistshadeverheardof whatiscalledtheWorldwideWeb...Nowevenmycathasits ownpage—BillClinton,1996 Thisbookgivesacomprehensivesurveyofnon-lifeinsurancemathematics.Origi- nallywrittenforusewiththeactuarialscienceprogramsattheUniversitiesofAm- sterdam and Leuven, it is now inuse at many other universities, as well as for the non-academic actuarial education program organized by the Dutch Actuarial So- ciety. It provides a link to the further theoretical study of actuarial science. The methodspresentedcannotonlybeusedinnon-lifeinsurance,butarealsoeffective inotherbranchesofactuarialscience,aswellas,ofcourse,inactuarialpractice. Apart from the standard theory, this text contains methods directly relevant for actuarialpractice,forexampletheratingofautomobileinsurancepolicies,premium principlesandriskmeasures,andIBNRmodels.Also,theimportantactuarialstatis- ticaltooloftheGeneralizedLinearModelsisstudied.Thesemodelsprovideextra possibilities beyond ordinary linear models and regression that are the statistical tools of choice for econometricians. Furthermore, a short introduction is given to credibilitytheory.Anothertopicwhichalwayshasenjoyedtheattentionofriskthe- oreticiansisthestudyoforderingofrisks.Thebookreflectsthestateoftheartin actuarialrisktheory;manyresultspresentedwerepublishedintheactuariallitera- tureonlyrecently. Inthissecondeditionofthebook,wehaveaimedtomakethetheoryevenmore directly applicable by using the software R. It provides an implementation of the language S, not unlike S-Plus. It is not just a set of statistical routines but a full- fledgedobjectorientedprogramminglanguage.Othersoftwaremayprovidesimilar capabilities, but the great advantage of R is that it is open source, hence available toeveryonefreeofcharge.Thisiswhywefeeljustifiedinimposingitontheusers ofthisbookasadefactostandard.Ontheinternet,alotofdocumentationaboutR canbefound.InanAppendix,wegivesomeexamplesofuseofR.Afterageneral introduction,explaininghowitworks,westudyaproblemfromriskmanagement, tryingtoforecastthefuturebehaviorofstockpriceswithasimplemodel,basedon stockpricesofthreerecentyears.Next,weshowhowtouseRtogeneratepseudo- randomdatasetsthatresemblewhatmightbeencounteredinactuarialpractice. vii viii Preface Modelsandparadigmsstudied Thetimeaspectisessentialinmanymodelsoflifeinsurance.Betweenpayingpre- miumsandcollectingtheresultingpension,decadesmayelapse.Thistimeelement is less prominent in non-life insurance. Here, however, the statistical models are generally more involved. The topics in the first five chapters of this textbook are basicfornon-lifeactuarialscience.Theremainingchapterscontainshortintroduc- tionstoothertopicstraditionallyregardedasnon-lifeactuarialscience. 1.Theexpectedutilitymodel The very existence of insurers can be explained by the expected utility model. In thismodel,aninsuredisariskaverseandrationaldecisionmaker,whobyvirtueof Jensen’sinequalityisreadytopaymorethantheexpectedvalueofhisclaimsjustto beinasecurefinancialposition.Themechanismthroughwhichdecisionsaretaken underuncertaintyisnotbydirectcomparisonoftheexpectedpayoffsofdecisions, butratheroftheexpectedutilitiesassociatedwiththesepayoffs. 2.Theindividualriskmodel In the individual risk model, as well as in the collective risk model below, the to- tal claims on a portfolio of insurance contracts is the random variable of interest. Wewanttocompute,forexample,theprobabilitythatacertaincapitalwillbesuf- ficient to pay these claims, or the value-at-risk at level 99.5% associated with the portfolio,beingthe99.5%quantileofitscumulativedistributionfunction(cdf).The totalclaimsismodeledasthesumofallclaimsonthepolicies,whichareassumed independent.Suchclaimscannotalwaysbemodeledaspurelydiscreterandomvari- ables,noraspurelycontinuousones,andweuseanotation,involvingStieltjesinte- gralsanddifferentials,encompassingboththeseasspecialcases. Theindividualmodel,thoughthemostrealisticpossible,isnotalwaysverycon- venient, because the available dataset is not in any way condensed. The obvious technique to use in this model is convolution, but it is generally quite awkward. Using transforms like the moment generating function sometimes helps. The Fast FourierTransform(FFT)techniquegivesafastwaytocomputeadistributionfrom itscharacteristicfunction.ItcaneasilybeimplementedinR. Wealsopresentapproximationsbasedonfittingmomentsofthedistribution.The CentralLimitTheorem,fittingtwomoments,isnotsufficientlyaccurateintheim- portant right-hand tail of the distribution. So we also look at some methods using threemoments:thetranslatedgammaandthenormalpowerapproximation. 3.Collectiveriskmodels Amodelthatisoftenusedtoapproximatetheindividualmodelisthecollectiverisk model.Inthismodel,aninsuranceportfolioisregardedasaprocessthatproduces claimsovertime.Thesizesoftheseclaimsaretakentobeindependent,identically distributedrandomvariables,independentalsoofthenumberofclaimsgenerated. This makes the total claims the sum of a random number of iid individual claim amounts.UsuallyoneassumesadditionallythatthenumberofclaimsisaPoisson variatewiththerightmean,orallowsforsomeoverdispersionbytakinganegative Preface ix binomial claim number. For the cdf of the individual claims,one takes an average ofthecdfsoftheindividualpolicies.Thisleadstoaclosefittingandcomputation- ally tractable model. Several techniques, including Panjer’s recursion formula, to computethecdfofthetotalclaimsmodeledthiswayarepresented. For some purposes it is convenient to replace the observed claim severity dis- tributionbyaparametriclossdistribution.Familiesthatmaybeconsideredarefor examplethegammaandthelognormaldistributions.Wepresentanumberofsuch distributions,andalsodemonstratehowtoestimatetheparametersfromdata.Fur- ther, we show how to generate pseudo-random samples from these distributions, beyondthestandardfacilitiesofferedbyR. 4.Theruinmodel Theruinmodeldescribesthestabilityofaninsurer.Startingfromcapitaluattime t=0,hiscapitalisassumedtoincreaselinearlyintimebyfixedannualpremiums, butitdecreaseswithajumpwheneveraclaimoccurs.Ruinoccurswhenthecapital isnegativeatsomepointintime.Theprobabilitythatthiseverhappens,underthe assumptionthattheannualpremiumaswellastheclaimgeneratingprocessremain unchanged, is a good indication of whether the insurer’s assets match his liabili- ties sufficiently. If not, one may take out more reinsurance, raise the premiums or increasetheinitialcapital. Analyticalmethodstocomputeruinprobabilitiesexistonlyforclaimsdistribu- tions that are mixtures and combinations of exponential distributions. Algorithms exist for discrete distributions with not too many mass points. Also, tight upper and lower bounds can be derived. Instead of looking at the ruin probability ψ(u) with initial capital u, often one just considers an upper bound e−Ru for it (Lund- berg), where the number R is the so-called adjustment coefficient and depends on theclaimsizedistributionandthesafetyloadingcontainedinthepremium. Computing a ruin probability assumes the portfolio to be unchanged eternally. Moreover, it considers just the insurance risk, not the financial risk. Therefore not much weight should be attached to its precise value beyond, say, the first relevant decimal.Thoughsomeclaimthatsurvivalprobabilitiesare‘thegoalofrisktheory’, manyactuarialpractitionersareoftheopinionthatruintheory,howevertopicalstill in academic circles, is of no significance to them. Nonetheless, we recommend to study at least the first three sections of Chapter 4, which contain the description ofthePoissonprocessaswellassomekeyresults.Asimpleproofisprovidedfor Lundberg’sexponentialupperbound,aswellasaderivationoftheruinprobability incaseofexponentialclaimsizes. 5.Premiumprinciplesandriskmeasures Assumingthatthecdfofariskisknown,oratleastsomecharacteristicsofitlike meanandvariance,apremiumprincipleassignstotheriskarealnumberusedasa financialcompensationfortheonewhotakesoverthisrisk.Notethatwestudyonly riskpremiums,disregardingsurchargesforcostsincurredbytheinsurancecompany. Bythelawoflargenumbers,toavoideventualruinthetotalpremiumshouldbeat leastequaltotheexpectedtotalclaims,butadditionally,therehastobealoadingin x Preface thepremiumtocompensatetheinsurerformakingavailablehisriskcarryingcapac- ity.Fromthisloading,theinsurerhastobuildareservoirtodrawuponinadverse times, so as to avoid getting in ruin. We present a number of premium principles, together with the most important properties that characterize premium principles. The choice of a premium principle depends heavily on the importance attached to suchproperties.Thereisnopremiumprinciplethatisuniformlybest. Risk measures also attach a real number to some risky situation. Examples are premiums, infinite ruin probabilities, one-year probabilities of insolvency, the re- quiredcapitaltobeabletopayallclaimswithaprescribedprobability,theexpected valueoftheshortfallofclaimsoveravailablecapital,andmore. 6.Bonus-malussystems Withsometypesofinsurance,notablycarinsurance,chargingapremiumbasedex- clusivelyonfactorsknownaprioriisinsufficient.Toincorporatetheeffectofrisk factorsofwhichtheuseasratingfactorsisinappropriate,suchasraceorquiteoften sexofthepolicyholder,andalsoofnon-observablefactors,suchasstateofhealth, reflexesandaccidentproneness,manycountriesapplyanexperienceratingsystem. Suchsystemsontheonehandusepremiumsbasedonapriorifactorssuchastype of coverage and list-price or weight of a car, on the other hand they adjust these premiums by using a bonus-malus system, where one gets more discount after a claim-freeyear,butpaysmoreafterfilingoneormoreclaims.Inthisway,premi- ums are charged that reflect the exact driving capabilities of the driver better. The situationcanbemodeledasaMarkovchain. The quality of a bonus-malus system is determined by the degree in which the premium paid is in proportion to the risk. The Loimaranta efficiency equals the elasticity of the mean premium against the expected number of claims. Finding it involvescomputingeigenvectorsoftheMarkovmatrixoftransitionprobabilities.R providestoolstodothis. 7.Orderingofrisks It is the very essence of the actuary’s profession to be able to express preferences betweenrandomfuturegainsorlosses.Therefore,stochasticorderingisavitalpart ofhiseducationandofhistoolbox.SometimesithappensthatfortwolossesX and Y,itisknownthateverysensibledecisionmakerpreferslosingX,becauseY isin a sense ‘larger’ than X. It may also happen that only the smaller group of all risk averse decision makers agree about which risk to prefer. In this case, riskY may belargerthanX,ormerelymore‘spread’,whichalsomakesarisklessattractive. Whenweinterpret‘morespread’ashavingthickertailsofthecumulativedistribu- tionfunction,wegetamethodoforderingrisksthathasmanyappealingproperties. Forexample,thepreferredlossalsooutperformstheotheroneasregardszeroutility premiums, ruin probabilities, and stop-loss premiums for compound distributions with these risks as individual terms. It can be shown that the collective model of Chapter 3 is more spread than the individual model it approximates, hence using thecollectivemodel,inmostcases,leadstomoreconservativedecisionsregarding premiums to be asked, reserves to be held, and values-at-risk. Also, we can prove Preface xi thatthestop-lossinsurance,demonstratedtobeoptimalasregardsthevarianceof theretainedriskinChapter1,isalsopreferable,otherthingsbeingequal,intheeyes ofallriskaversedecisionmakers. Sometimes,stop-losspremiumshavetobesetunderincompleteinformation.We give a method to compute the maximal possible stop-loss premium assuming that themean,thevarianceandanupperboundforariskareknown. Intheindividualandthecollectivemodel,aswellasinruinmodels,weassume that the claim sizes are stochastically independent non-negative random variables. Sometimesthisassumptionisnotfulfilled,forexamplethereisanobviousdepen- dencebetweenthemortalityrisksofamarriedcouple,betweentheearthquakerisks ofneighboringhouses,andbetweenconsecutivepaymentsresultingfromalifein- surancepolicy,notonlyifthepaymentsstoporstartincaseofdeath,butalsoincase ofarandomforceofinterest.Wegiveashortintroductiontotheriskorderingthat appliesforthiscase.Itturnsoutthatstop-losspremiumsforasumofrandomvari- ables with an unknown joint distribution but fixed marginals are maximal if these variablesareasdependentasthemarginaldistributionsallow,makingitimpossible thattheoutcomeofoneis‘hedged’byanother. In finance, frequently one has to determine the distribution of the sum of de- pendentlognormalrandomvariables.Weapplythetheoryoforderingofrisksand comonotonicitytogiveboundsforthatdistribution. Wealsogiveashortintroductioninthetheoryoforderingofmultivariaterisks. Onemightsaythattworandomsvariablesaremorerelatedthananotherpairwith the same marginals if their correlation is higher. But a more robust criterion is to restrict this to the case that their joint cdf is uniformly larger. In that case it can beprovedthatthesumoftheserandomvariablesislargerinstop-lossorder.There areboundsforjointscdfsdatingbacktoFre´chetinthe1950’sandHo¨ffdinginthe 1940’s.Forarandompair(X,Y),thecopulaisthejointcdfoftheranksF (X)and X F (Y).Usingthesmallestandthelargestcopula,itispossibletoconstructrandom Y pairswitharbitraryprescribedmarginalsand(rank)correlations. 8.Credibilitytheory The claims experience on a policy may vary by two different causes. The first is thequalityoftherisk,expressedthroughariskparameter.Thisrepresentstheaver- ageannualclaimsinthehypotheticalsituationthatthepolicyismonitoredwithout change over a very long period of time. The other is the purely random good and bad luck of the policyholder that results in yearly deviations from the risk para- meter. Credibility theory assumes that the risk quality is a drawing from a certain structuredistribution,andthatconditionallygiventheriskquality,theactualclaims experienceisasamplefromadistributionhavingtheriskqualityasitsmeanvalue. Thepredictorfornextyear’sexperiencethatislinearintheclaimsexperienceand optimalinthesenseofleastsquaresturnsouttobeaweightedaverageoftheclaims experience of the individual contract and the experience for the whole portfolio. Theweightfactoristhecredibilityattachedtotheindividualexperience,henceitis called the credibility factor, and the resulting premiums are called credibility pre- xii Preface miums.Asaspecialcase,westudyabonus-malussystemforcarinsurancebased onaPoisson-gammamixturemodel. CredibilitytheoryisactuallyaBayesianinferencemethod.Bothcredibilityand generalizedlinearmodels(seebelow)areinfactspecialcasesofso-calledGeneral- izedLinearMixedModels(GLMM),andtheRfunctionglmmisabletodealwith boththerandomandthefixedparametersinthesemodels. 9.Generalizedlinearmodels Many problems in actuarial statistics are Generalized Linear Models (GLM). In- steadofassuminganormallydistributederrorterm,othertypesofrandomnessare allowedaswell,suchasPoisson,gammaandbinomial.Also,theexpectedvaluesof thedependentvariablesneednotbelinearintheregressors.Theymayalsobesome functionofalinearformofthecovariates,forexamplethelogarithmleadingtothe multiplicativemodelsthatareappropriateinmanyinsurancesituations. Thisway,onecanforexampletackletheproblemofestimatingthereservetobe keptforIBNRclaims,seebelow.Butonecanalsoeasilyestimatethepremiumsto bechargedfordriversfromregioniinbonusclass jwithcarweightw. In credibility models, there are random group effects, but in GLMs the effects arefixed,thoughunknown.TheglmmfunctioninRcanhandleamultitudeofmod- els,includingthosewithbothrandomandfixedeffects. 10.IBNRtechniques Animportantstatisticalproblemforthepracticingactuaryistheforecastingofthe total of the claims that are Incurred, But Not Reported, hence the acronym IBNR, ornotfullysettled.Mosttechniquestodetermineestimatesforthistotalarebased on so-called run-off triangles, in which claim totals are grouped by year of origin anddevelopmentyear.Manytraditionalactuarialreservingmethodsturnouttobe maximumlikelihoodestimationsinspecialcasesofGLMs. Wedescribetheworkingsoftheubiquitouschainladdermethodtopredictfuture losses,aswellas,briefly,theBornhuetter-Fergusonmethod,whichaimstoincorpo- rateactuarialknowledgeabouttheportfolio.Wealsoshowhowthesemethodscan beimplementedinR,usingtheglmfunction.Inthissameframework,manyexten- sionsandvariantsofthechainladdermethodcaneasilybeintroduced.Englandand Verrallhaveproposedmethodstodescribethepredictionerrorwiththechainladder method,bothananalyticalestimateofthevarianceandabootstrappingmethodto obtainanestimateforthepredictivedistribution.WedescribeanRimplementation ofthesemethods. 11.MoreonGLMs For the second edition, we extended the material in virtually all chapters, mostly involvingtheuseofR,butwealsoaddsomemorematerialonGLMs.Webriefly recapitulatetheGauss-Markovtheoryofordinarylinearmodelsfoundinmanyother textsonstatisticsandeconometrics, andexplain howthealgorithmbyNelderand Wedderburn works, showing how it can be implemented in R. We also study the stochasticcomponentofaGLM,statingthattheobservationsareindependentran-

Description:
Modern Actuarial Risk Theory contains what every actuary needs to know about non-life insurance mathematics. It starts with the standard material like utility theory, individual and collective model and basic ruin theory. Other topics are risk measures and premium principles, bonus-malus systems, or
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.