Risk Dashboards with R Shiny: a C-ROSS case study. Phil Joubert, Oliver Wyman

Transcription

1 Risk Dashboards with R Shiny: a C-ROSS case study Phil Joubert, Oliver Wyman phil.joubert@oliverwyman.com 06 May 2015

2 Agenda Setting the scene: What is C-ROSS? What is a risk dashboard? What is R Shiny? Putting it together: Designing the system Designing the dashboard Developing code in RStudio The payoff: What s wrong with our current set up? Delivering insights (or, what s the point of all this?) Summary and conclusions Q&A 06 May

3 Setting the scene 06 May 2015

4 Setting the scene 2. Risk Dashboards Trendy and useful presentation technique 1. C-ROSS New Chinese regulatory framework 3. R Shiny Trendy web-interface extension to a popular statistical programming language

5 What is C-ROSS? Overall framework of C-ROSS Unified supervision Emerging market Risk-oriented and Value-based 1 System features Pillar 1 - Quantitative capital requirements Quantitative capital requirements Actual capital assessment standards & capital classifications Supervision measures Pillar 2 - Qualitative supervision requirements Integrated Risk Rating Solvency Aligned Risk Management Requirements and Assessment Supervision check and measurements Pillar 3 - Market discipline mechanism Improve risk disclosure Transparency Develop market disciplinary mechanism and optimize the market environment 2 Supervision elements Company s solvency management 3 Supervision basis Source: Overall Framework of the Second-Generation Solvency Supervision System of China by CIRC

6 C-ROSS: Pillar One Total Required Capital Market Credit Life Non-Life Interest Rates Spreads Mortality & Morbidity Surrenders Expense Catastrophe Premium & Reserve Equity Defaults Projected Mortality Projected Surrender Typhoon Property Premium Reserve Pandemic Mortality Mass Surrender Earthquake Auto Premium Reserve Real Estate Agricultural Premium Reserve Morbidity 06 May

7 What is a Dashboard? Communicates information succinctly Delivers pertinent information Uses knowledge of human perception Displays deeper information on demand make it simple but not too simple make it visual drill-down capability 06 May

8 (Risk) Dashboards examples All investments / cash in one place Bright colours, big fonts Relative importance? Pie charts? Why this position? 06 May

9 (Risk) Dashboards examples Too much information Not visual enough 06 May

10 Managers of differing seniority have different needs Risk reporting hierarchy model Reporting hierarchy Board Executive management level BU management level Enterprise overview Report Executive summary Risk type and BU overview Appendix Report Executive summary Risk type and BU specifics Appendix Report Position and exposure drill-down Appendix

11 What is R Shiny? Web interface for R Designed to deliver small apps Simple rules based layout structures Built as part of the RStudio suite Uses CSS / Node.js under the hood 06 May

12 Putting it together 06 May 2015

13 Schematic of a management information system Data Parameters Calculation engine Results Report generation / presentation tool 06 May

14 Schematic of a management information system demo implementation Data Parameters Calculation engine Results Report generation / presentation tool Files R Package Memory R Shiny 06 May

15 Designing your C-ROSS calculations Risk A Risk B Risk C Risk D 1 ρ AB ρ AC ρ AD Risk X = E x RF Table lookups = E x RF 0 ( 1 + k 1 k n ) 1 ρ BC ρ BD 1 ρ CD 1 Aggregation = VCV Matrix Multiplication 1 1 Other required features Hierarchy data structure Node walking I / O 06 May

16 Designing your dashboard: Brainstorm some ideas Solvency margin (current, recent) Integrated Risk Rating Required capital vs Available capital Available capital by tier Required capital by risk Required capital by BU Asset mix Exposures by BU / geography Situation under stress scenarios Market indicators Narrative / action log May

17 Developing code in RStudio Issue with R alone No IDE No version control Difficult to check / test code and difficult to apply the encapsulation principle Solution with RStudio Visual debugger (break, watch, step over, etc) Integrated version control with git and github (and SVN) Accessible development of packages, including testing Some issues remain: Obscure syntax Run times Open source 06 May

18 R Shiny the basics 1 Web server ShinyApp server Web browser UI.r Server.r Any R code 2 Reactivity Input X Output Y Y = f(x) When user changes X, Shiny recalculates f() and updates Y 06 May

19 Package shinydashboard UI.r library(shiny) library(shinydashboard) Load up the new library dashboardpage( dashboardheader(title = "Old Faithful Geyser Data"), dashboardsidebar(), dashboardbody( fluidrow( box( sliderinput("bins", "Number of bins:", min = 1, max = 50, value = 30) ), box( plotoutput("distplot") )))) Slightly different to the standard Shiny declarations Every element needs to be in a box (unless it is already a box!) 06 May

20 Package shinydashboard Look familiar? 06 May

21 Implementing your C-ROSS dashboard dashboardpage( dashboardheader( ), dashboardsidebar( ), dashboardbody( ) ) A Title List tabs Lay out each tab s contents A B C B C 06 May

22 A C UI.r tabitems( tabitem( tabname = "Summary", fluidrow( valueboxoutput( "vbxsummarycalculationdate" ), valueboxoutput( "vbxsummarysolvenymargin" ) )) B Server.r output$vbxsummarysolvenymargin <- rendervaluebox({ valuebox( printpercent( LatestSolvencyMargin ), "Solvency Margin", icon= icon("money"), B C color = ragcolour(latestsolvencymargin) ) A }) 06 May

23 Available and Required Capital Drill down 06 May

24 Exposures by BU 06 May

25 The payoff 06 May 2015

26 What s wrong with our current set-up? 1. Out-dated techniques 2. Bad engineering: failing to minimise operational risks and optimise rewarded risks 3. Poor use of resources 06 May

27 Delivering insights (or, what s the point?) To make good business decisions managers need information which is: presented clearly accurate timely A well designed dashboard will address the presentational issue Accuracy is improved by using code which is tested and version controlled Timeliness is a function of people and process the people issue is often easier to optimise than process (e.g. data delivery times may be outside of your control) 06 May

28 Summary and conclusions C-ROSS Dashboards R Shiny A quantum leap for the Chinese insurance industry, which should improve management and customer outcomes, but which presents short-term operational, systems and education challenges A neat way of presenting pertinent information visually, and an opportunity for actuaries and insurance risk people to take advantage of insights developed in the software industry A quick, powerful way of creating and delivering management information, and a welcome user interface to R 06 May

29 Questions? Comments? Expressions of individual views by members of the Institute and Faculty of Actuaries and its staff are encouraged. The views expressed in this presentation are those of the presenter. 06 May

30 Speaker Bio Phil Joubert is a principal in the Financial Services practice of Oliver Wyman, based in Hong Kong Office. Phil has fifteen years of experience in the financial services industry, having worked in areas as diverse as actuarial consulting and derivatives trading. He has worked in both Europe and Asia-Pac for a variety of insurers, banks and software houses, and specialises in risk and capital modelling and systems design Recent experience Regulatory capital model design and implementation for several insurers in Europe Economic capital implementation for leading pan-asian insurance group Derivatives trading and market risk management Capital aggregation systems design for leading ESG provider Actuarial automation implementation project for specialist life insurer Phil holds an MSc in Finance & Mathematics from Imperial College and is a Fellow of the Faculty of Actuaries. He joined Oliver Wyman in 2014, having spent several years as an independent actuarial consultant. Previously he worked at Deutsche Bank and Natixis as a trader and he started his career with Deloitte Actuarial. 06 May