Transcription
13/01/2011Open Forum – 21st January 2011Marcus Bowser (Chair)Neil CantleNeil AllanUnravelling thecomplexity of risk 2010 The Actuarial Profession www.actuaries.org.ukPlan for this session Why is a new approach needed and why now? Overview of complex adaptive systems (CAS)– What are they?– Why should you be interested?– Basis of the science behind CAS Are companies and organisations CAS? Can risks be modelled as a CAS? Examples of applications for the profession Open discussion and questions 2010 The Actuarial Profession www.actuaries.org.uk11
13/01/2011Unravelling the complexity of riskOverview of complexadaptive systems 2010 The Actuarial Profession www.actuaries.org.ukPath to our understanding 2010 The Actuarial Profession www.actuaries.org.uk32
13/01/2011Roadmap of the development of complexity science 2010 The Actuarial Profession www.actuaries.org.uk4Systems Thinking Systems thinking is both a worldview that:– Problems cannot be addressed by reduction of the system– System behaviour is about interactions and relationships and– Emergent behaviour is a result of those interactions And a process or methodology– To understanding complex system behaviour– To see both the “forest and the trees”– Identify possible solutions and system learning– Utilises complexity science and other desciplines 2010 The Actuarial Profession www.actuaries.org.uk53
13/01/2011Complexity and complex systemsThe development of complexity science is a shift inscientific approach towards an interdisciplinaryparadigm with the potential to profoundlyaffect business, organisations and government.The goal of complexity science is to understandcomplex systems: what ”rules” govern theirbehaviour, how they manage change, learn efficientlyand optimise their own behaviour.6 2010 The Actuarial Profession www.actuaries.org.ukWhat is a system ?OutputInput“a set of components interconnected for a purpose.” 2010 The Actuarial Profession www.actuaries.org.uk74
13/01/2011What is a complex system ?InputInputOutputFeedbackOutput8 2010 The Actuarial Profession www.actuaries.org.ukWhat is a complex adaptive system ?OutputInputInputElements can change 2010 The Actuarial Profession www.actuaries.org.ukOutputFeedback95
13/01/2011Complex Adaptive Systems Examples:– Sand Pile– Immune system– Weather system– Forests– Birds flocking– Organisations– Supply chains– ERM– Fish stocks10 2010 The Actuarial Profession www.actuaries.org.ukA System and a HeapA systemA heapInterconnecting parts functioning as a wholeA collection of partsChanged if you take away pieces or add morepieces.Essential properties are unchanged whetheryou add or take away pieces.The arrangement of the pieces is crucialThe arrangement of the pieces is irrelevantThe parts are connected and work togetherThe parts are not connected and can functionseparatelyIts behaviour depends on the total structure.Its behaviour (if any) depends on its size or onthe number of pieces in the heap.(O'Connor and McDermott, 1997) 2010 The Actuarial Profession www.actuaries.org.uk116
13/01/2011Complex Adaptive System Characteristics– Has a purpose– Emergence – the whole has properties not held by subcomponents– Self Organisation – structure and hierarchy but few leveragepoints– Interacting feedback loops – causing highly non-linearbehaviour– Counter-intuitive and non-intended consequences– Has tipping point or critical complexity limit before collapse– Evolves and history is important– Cause and symptom separated in time and space 2010 The Actuarial Profession www.actuaries.org.uk12Emergence – E.g. MusicYou can explore the characteristicsof individual notes.but you cannot know the tunewithout knowing the interactions(score) 2010 The Actuarial Profession www.actuaries.org.uk137
13/01/2011Self-Organisation and emergence14 2010 The Actuarial Profession www.actuaries.org.ukNon-linearity The Beer Game (MIT, 1960’s)Source: www.beergame.org 2010 The Actuarial Profession www.actuaries.org.uk158
13/01/2011Unintended consequences People “understand” bits of risk, not the whole thing 2010 The Actuarial Profession www.actuaries.org.uk16Examples of tipping point collapse Liquidity crunch Cutting down too manytrees Relying on debt Building too many offices orresidences Ozone layer, ground water,agricultural soils etc Sand pile 2010 The Actuarial Profession www.actuaries.org.uk179
13/01/2011Causes of tipping point & overshoot Poor understanding of the level and the causes of the limit––––Humans are not perfectly rationalThey suffer bounded rationalityMental models incomplete / insufficient time to consider themCan’t even mentally simulate a first order linear feedback loop Tendency to be focused on indirect, delayed indicator for healthof system Momentum in the system and positive reinforcing loops Long delays in deciding, responding & affecting change Competition and focus on short term measures (long-termresponse may be different to short-term one) 2010 The Actuarial Profession www.actuaries.org.uk18Evolution – path dependency and history 2010 The Actuarial Profession www.actuaries.org.uk1910
13/01/2011Entropic cycling (after Hitchens) ivorsemergeDominanceemergesDecay orCollapseVariety issuppressedMoribundsystem20 2010 The Actuarial Profession www.actuaries.org.ukIdealised heating systemEnergy 2010 The Actuarial Profession emHeat2111
13/01/2011Real world heating systemCost of nimiseenergy useComfortLevelEnvironment22 2010 The Actuarial Profession www.actuaries.org.ukBusiness as a heating systemCost of financial ptimise financialresourcesComfortLevelEnvironment 2010 The Actuarial Profession www.actuaries.org.uk2312
13/01/2011Recap – Complex Adaptive Systems Systems theory is a structured way to describe a set ofinteracting components which have a purpose Complex adaptive systems (CAS) have defined properties The study of CAS is interdisciplinary – so are applicable tools Complex behaviour can arise from simple rules Emergence requires a holistic approach before studying parts Important to know a systems critical complexity trajectory24 2010 The Actuarial Profession www.actuaries.org.ukUnravelling the complexity of riskAre companies complexadaptive systems? 2010 The Actuarial Profession www.actuaries.org.uk13
13/01/2011Relevance to Companies Companies are CAS because they consist of people who are:– Adhering to cultural norms, beliefs, principals– Following processes, learning, adapting, interdependent– Communicate, use initiative, often irrational, interact The industry and related companies are self-organising External environment is changing and impacts companies Emergent behaviour brings significant new systemic risks Evolution and history is important26 2010 The Actuarial Profession www.actuaries.org.ukThe human factor“There can no longer be any doubt that the micro assumptions of [economic] theory– the assumptions of perfect rationality – are contrary to fact. It is not a question ofapproximation; they do not even remotely describe the process that human beingsuse for making decisions in complex situations.”Herbert Simon 1979“How do humans reason in situations that are complicated or ill-defined? Modernpsychology tells that as humans we are only moderately good at deductive logic,and we make only moderate use of it. But we are superb at seeing or recognisingor matching patterns – behaviours that confer obvious evolutionary benefits. Inproblems of complication, then, we look for patterns.”Brian Arthur “Inductive reasoning and bounded rationality” American Economic Review 84 #2 (1994) 2010 The Actuarial Profession www.actuaries.org.uk2714
13/01/2011Organisations as complex systems Organisations are segmented rather than monolithic Stable segments within organisations are quite small Connections between segments are of varying strength, andthey produce ambiguity The way information flows and interactions occur mattersbecause:– Influences culture, hierarchy and structure– Impacts on speed of communication28 2010 The Actuarial Profession www.actuaries.org.ukUnravelling the complexity of riskCan risks be modelled as acomplex adaptive system? 2010 The Actuarial Profession www.actuaries.org.uk15
13/01/2011Applied to risk Risk is the unintended emergent property of a CAS Risk is a process which emerges over time from the complexinteractions of many factors Risk has multiple-characteristics Risk has structure and hierarchy Human bias is highly prevalent in assessing risk Emerging risk is a function of the past system performance 2010 The Actuarial Profession www.actuaries.org.uk30Motivation For New Approach Conceptual framework typically used for risk is flawed– Risk is an emergent property so aggregating the behaviour ofcomponents cannot tell you about the whole Risk assessment nearly always relies upon human judgement– Humans are not good at assessing risk Frameworks provide limited predictive capability– Models focus on outcomes not real drivers Business has become increasingly complex and techniques are still aboutlinear behaviours and “normal” distributions Time to evolve 2010 The Actuarial Profession www.actuaries.org.uk3116
13/01/2011Mis-Framing Risk Traditional risk management approaches oversimplify– Makes modelling more tractable– Makes data easier to organise– But does not describe how risks really behave Risks are– Treated as events which happen at a point in time– Characterised by a single “dominant” feature– Considered as being homogeneous according to that label In slow-moving benign conditions these assumptions work OK In a complex environment they really don’t32 2010 The Actuarial Profession www.actuaries.org.ukUnravelling the complexity of riskExamples of applicationsfor the Profession 2010 The Actuarial Profession www.actuaries.org.uk17
13/01/2011But first.don’t oversimplify Looking for patterns needs information Many attempts to monitor risk throw that away at outset Don’t guess in advance what you expect to see Need a “model-free” approach to see emergenceAllow people to “mix” colours34 2010 The Actuarial Profession www.actuaries.org.ukUnderstanding The SystemKey NodesKey Drivers 2010 The Actuarial Profession www.actuaries.org.ukGaps3518
13/01/2011Cognitive Mapping The theories we use here are based around:– Personal Construct Theory (George Kelly 1955) – you knowyour environment– Grounded Theory (Glaser and Strauss 1967) – generate atheory from the research– Cognitive mapping (Colin Eden, Fran Akermann and SteveCropper 1990) – combine multiple “theories” to form singleperspective of a problem 2010 The Actuarial Profession www.actuaries.org.uk36A Dynamic Loop From Cognitive Map 2010 The Actuarial Profession www.actuaries.org.uk3719
13/01/2011Can Be Modelled With Systems DynamicsA small change in trainingtime and experiencedleaver rate has dramaticimpact on service quality. 2010 The Actuarial Profession www.actuaries.org.uk38Applications Rapidly elicit highly detailed description of risk profile andimplicit dynamics Feeds into:– Business planning– ORSA– Scenario development (and hence modelling)– Risk appetite framework– Emerging risk identification 2010 The Actuarial Profession www.actuaries.org.uk3920
13/01/2011Influence Modelling Lower frequency events tend to be quite heterogeneous Statistical models therefore problematic from outset More “correct” to model according to underlying cause Bayesian Networks can be used to capture expert knowledge ofrisk behaviour No need to correlate events, simply link by common cause Wide range of sophistication possible Good way to integrate expert knowledge with observedoutcomes40 2010 The Actuarial Profession www.actuaries.org.ukBayesian Networks Permits more transparencyand better engagementfrom businessSIMPLE EXAMPLE Combinations of earliertools can help to determinerelevant key drivers of riskoutcomeVaR 1.40m (99.5%) 2010 The Actuarial Profession www.actuaries.org.uk4121
13/01/2011Bayesian Networks (2)What if workvolumes arehigh?VaR 1.44m (99.5%)“Evidence” ofoutcome propagatesback up the modelWhat would need tohappen for worstoutcome to occur?Experts initially guess training is 60-80%effective and 25-35% ineffectiveParameterise model at 70%/30%Whichnodes/connectionsare most significantto outcome? 2010 The Actuarial Profession www.actuaries.org.ukObserve 80% effective over past yearUpdate distribution to 71%/29%(assumes expert guess follows Dirichlet distribution)42Bayesian Networks (3) Advantages:– Easier to test sensitivities/what-if analysis– Combines hard and soft data– Incorporate hard and soft evidence– Fast – no simulation– Can be projected sensibly through trends in drivers– Easy to communicate– Can combine with statistical models– Easy to establish risk monitoring linked to model components 2010 The Actuarial Profession www.actuaries.org.uk4322
13/01/2011Applications Operational risk modelling Scenario modelling for extreme risks Risk appetite and limit setting 2010 The Actuarial Profession www.actuaries.org.uk44Evolution is a signature of complex systems 2010 The Actuarial Profession www.actuaries.org.uk23
13/01/2011Typical Approach To Risk Identification People confuse “characteristics” with the risk itself Natural tendency to look at risk by “summing the parts” isencouraged by approaches to modelling and regulation ofsolvency capitalEnterprise RiskMarketCreditInsuranceOperational.46 2010 The Actuarial Profession www.actuaries.org.ukWhat Risk Really Looks LikeEnterprise Risk Looking at real risks we seethey have multiplecharacteristicsCredit They combine to produce“new” outcomes By seeking to understand theforces driving these dynamicsOperational .we can spot risks early .and make better judgementabout what matters 2010 The Actuarial Profession www.actuaries.org.uk4724
13/01/2011Risk evolution Produces reliable evolutionary information such as classification, direction,connection It demonstrates how a risk reached a certain state and how it might evolve Understanding the risk “DNA” enables modelling to show which risk areasare currently most prone to emerge as the new strain or breed of risks48 2010 The Actuarial Profession www.actuaries.org.ukComparison of Biological, Linguistic, Enterprise RiskBiological EvolutionLinguistic EvolutionEnterprise Risk EvolutionDiscrete charactersVocabulary, syntax, soundsCauses, loses, risk registersCommon ancestorsWords with common originRisks from common originMutationInnovationInnovation, regulationNatural selectionSocial selectionManagement selectionHorizontal gene transferBorrowing from otherlanguagesTransfer of info betweenbusinesses and industriesFossilsAncient textsHistoric case studiesSpecies splitting into othersLanguage Lineage SplitsRisk categories (strategic,operational, financial etc)ExtinctionLanguage deathRisk eradication 2010 The Actuarial Profession www.actuaries.org.ukAfter Pagel (2009) Nature4925
13/01/2011Cladistics a simple example(a) paired fins, (b) jaws, (c) large dermal bones, (d) fin rays, (e) lungs, and(f) rasping tongue 2010 The Actuarial Profession www.actuaries.org.uk50Analysing Risks Using Multiple Characteristics Determine risk characteristics (example) 2010 The Actuarial Profession www.actuaries.org.uk5126
13/01/2011Analysing Risks Using Multiple CharacteristicsNow categoriserisks accordingto “all” thecharacteristicsthey haveRisk ScenarioCharacteristic Number1. Liquidity challenge2. Regulation changes 1251, 15, 16, 17, 18, 19, 26, 33, 363. Violation of Privacy Protection9, 10, 12, 14, 17, 20, 21, 31, 344. Trusted Insider Technology Risks10, 31, 345. Business Continuity12, 30, 356. Technology development10, 31, 34, 357. Product26, 368. Geographical1, 2, 8, 18, 19, 26, 369. Regulation changes 217, 19, 3610. Succession Planning3311. Model complexity21, 22, 3212. Convergence of Products1, 26, 3613. Regulation changes 39, 10, 34, 3614. Poor decision making1, 35, 3715. Misunderstanding of risks2, 3, 1216. HR policies9, 10, 12, 3717. Long-term planning1, 32, 33, 3618. Tech infrastructure30, 35, 3719. Tax rules16, 26, 3620. Regulation differences18, 26, 3621. Tax management2622. Infrastructure30, 35, 37 2010 The Actuarial Profession www.actuaries.org.uk52Compute cladistic tree 2010 The Actuarial Profession www.actuaries.org.uk5327
13/01/2011Interpretation knowing path dependancy We can label thebranches to show“ancestor”development The coloured regionshelp to highlightgroups according to“early” genes 2010 The Actuarial Profession www.actuaries.org.uk54Emerging issues Risk profile continues to evolve (lots of short branches),potentially indicating risks not being controlled We can see risk #13 is actually not much like the otherregulatory risks and seems to be more like “control” failure risks Technology seems to be linked to control failure in this firm Much of the regulatory risk is to do with products 2010 The Actuarial Profession www.actuaries.org.uk5528
13/01/2011Evolutionary connectivity measure Some risks are quite close to many others. Likely to findemerging risks including traits of these risksKey: Red indicates risks are closely related, Green indicates risks are widely separated in evolutionary terms 2010 The Actuarial Profession www.actuaries.org.uk56Recap Risks have a unique sequence, very much like a DNA Collective risk systems evolve and co-evolve The path-dependency is an important aspect of a risk A risk’s evolutionary progression can be analysed Predictions made about how risks might develop It is a efficient way to classify and manage risks 2010 The Actuarial Profession www.actuaries.org.uk5729
13/01/2011Applications Any data with a large number of characteristics:– Classification of risk information– Anticipation of emerging risk possibilities– Analysis of organisation (e.g. Personnel skills, affinities)– Analysis of customer data– Analysis of business pipeline58 2010 The Actuarial Profession www.actuaries.org.ukUncertainty & EntropyFinding the tipping point 2010 The Actuarial Profession www.actuaries.org.uk30
13/01/2011Entropy/Uncertainty Measuring the information content (entropy) of system tells uswhether performance is making sense Information I(x) - log p(x) Entropy average information - p(x) log p(x) Intuition – high entropy high uncertainty:– Impossible event (p(x) 0) is surprising (I(x) )– Certain event (p(x) 1) is not interesting (I(x) 0) Through understanding your “system”, identify relevantvariables to monitor If their information content is high/volatile you need to know why 2010 The Actuarial Profession www.actuaries.org.ukNon-linear relationships Are we still talking?Different levels of correlationExample U[0,2 ]R U[4, 5]X R cosY R sinSample of 1000Correlation 0.0Mutual Info 1.0 2010 The Actuarial Profession www.actuaries.org.uk6131
13/01/2011Looking beneath the surfaceSameoutcomebutdifferentdriversProduced byMilliman using:62 2010 The Actuarial Profession www.actuaries.org.ukUncovering hidden changesTrend of SCR components looks stable over time.but uncertainty differs between components.and overall uncertainty changes over time.Produced byMilliman using:OpRiskgets moreimportantEqRiskgets lessimportant 2010 The Actuarial Profession www.actuaries.org.uk6332
13/01/2011Applications Model-free complexity analysis can be applied to:– Risk monitoring – spotting early emerging risk signals– Business performance – signs of sluggish/out-performance– Business intelligence – factors affecting customers, markets– Understanding non-linear model outputs– Determining rating factors for risks 2010 The Actuarial Profession www.actuaries.org.uk64Summary We can frame companies/industries as complex adaptivesystems Complex adaptive systems give out signals Using the right scientific tools you can spot them Interactions are the important part Early warnings are possible Don’t throw away information – look for patterns Try not to guess what is going on before you look at the data Evolution is informative about possible future trends Improved understanding facilitates better models/management 2010 The Actuarial Profession www.actuaries.org.uk6533
13/01/2011Questions or comments?Expressions of individual views bymembers of The Actuarial Professionand its staff are encouraged.The views expressed in this presentationare those of the presenters.66 2010 The Actuarial Profession www.actuaries.org.ukCharacteristics of simple, complex systems andcomplex adaptive systemsSimple systemsComplex systemsComplex adaptive systemsHave predictable behaviour; e.g.Generate counter-intuitive behaviour that is fullThe elements of a system can change themselves (thisa fixed interest bank account.of surprises; e.g. lower taxes and interest ratesrelates to notions of autonomy).leading to higher unemployment.Few interactions and feedback orA large array of variables with manyComplex outcomes can emerge from a few simple rulesfeed forward loops; e.g. a simpleinteractions, lags, feedback loops and feed(this relates to initial starting conditions and the idea thatbarter economy with few goodsforward loops, which create the possibility thatcomplicated targets and plans may stifle creative andand services.new, self-organizing behaviours will emerge;adaptive ability).e.g. most large organizations, life itself.Centralized decision making; e.g.Decentralized decision making – because power Small changes can have big effects and large changes maypower is concentrated among ais more diffuse, the numerous componentshave no effect – i.e. non-linearity operates (e.g. in the UKfew decision makers.generate the actual system behaviour.a small band of lorry drivers interconnected by mobilephones almost brought the country to a standstill byblocking petrol deliveries to service stations).Are decomposable because ofAre irreducible – neglecting any part of theThrive on tension and paradox. (It is argued that healthyweak interactions; i.e. it isprocess or severing any of the connectionsorganizations exist on the edge of chaos – a region ofpossible to look at componentslinking its parts usually destroys essentialmoderate certainty and agreement).without losing properties of theaspects of the system behaviour or structure.whole.There are dynamic changes in the system andthe environment.(After Casti, 1994, pp.271–273 and Plsek 2001) 2010 The Actuarial Profession www.actuaries.org.uk6734
Marcus Bowser (Chair) Neil Cantle Neil Allan Unravelling the complexity of risk Plan for this session Why is a new approach needed and why now? Overview of complex adaptive systems (CAS) -What are they? -Why should you be interested? -Basis of the science behind CAS Are companies and organisations CAS?
