edp project report osp update
TRANSCRIPT
AGENDA
EDP • Introduction (Rob Stevenson) • Motivation • What was done • What was learnt • What has been produced • So What? • Implications • What Next?
Questions
OSP • Background (Trevor Maynard) • Oasis Vision and Lloyd's • OSP Timeline • Uncertainties • Model Validation • So What? • How to get involved Questions
EDP - AGENDA • Introduction (Rob Stevenson) • Motivation • What was done • What was learnt • What has been produced • So What? • Implications • What Next?
Questions
EDP Plan
Orientation Overview Requirements Design Evaluation Finalise Deliverables Communicate
Jun Aug Oct Task Jul Sep Nov
Orientation Meeting (19th June)
LMA EMWG Forum (10th July)
Requirements Meetings
Evaluation Workshop (9th Sept)
LMA
= Technical Steering Meetings = Other Meetings = EMWG Meetings
Outputs (30th June)
Plus monthly Progress Reports
EMWG Forum
(10th Dec)
Who’s Who Sponsor: LMA Board & David Gittings Overall Steering: EMWG, chair Rob Stevenson Technical Steering: TSG, chair Sian Fleming LMA: Gary Budinger Pat Hakong Margaret Fawke Project Team: Dickie Whitaker (Oasis Palm Tree) Peter Taylor (Oasis Palm Tree) Eliza Tadley (KCC) Ian Nicol (ATD) Phil Burgin (ATD) Richard Toothill (ATD)
Stakeholder Requirements Meetings
Who When No. of Attendees
No. of requirements
Software Houses 9th July 12 29 Brokers 15th July 7 12 Modelling Companies 16th July 6 10 Xchanging (XIS) 21st July 5 7 Carriers 22nd July 42 55 Transport providers (TMEL) 25th July 2 4
Summary of Requirements 1. Maintainability (change process)
2. Extensibility (applicability to new LOBs,
insurance types)
3. Data sharing (through a centralised model?)
4. Information management (exposure data, results, guidelines)
5. Scalability & performance (data access, storage, transport)
6. Interoperability (of new & existing formats)
7. Template amendments (spreadsheets)
• Commonality (over all data types, lines of business etc) • Flexibility (to handle new data formats) • Be controllable • Be quick & easy (not slow & bureaucratic!)
Lines of business: • Property • Offshore energy • Specialty Diverse data inputs into loss modelling formats
Types of insurance: • D & F • Treaty • Binders
Workstreams Ref Workstream Description Interest expressed by (Lead in bold) A1 Non-property Exposure Data Capture &
Database Talbot; Faraday; Navigators; Munich Re; Beazley; SCOR; Catlin; Atrium; AIR
A2 Binders Cleansing Kiln; Faraday; Navigators; Hardy; SCOR; Talbot; Catlin; OIM; ImageCat; Catex; LMA
A3 Treaty Exposure Database Talbot; Faraday; Hardy; SCOR; Catlin; Aspen Re; ImageCat
A4 Portfolio Data Lake Barbican; Mitsui; SCOR; Markel; ImageCat; IBM
A5 Results Data Lake SCOR; Navigators; Talbot; Markel; Aspen Re; IBM
B1 Business Implementation Guidelines (incorporating Data Quality & Project Information)
LMA; Aspen Re; Beazley; Faraday; Hardy; Navigators; SCOR; Talbot; Torus; Agencyport (Xuber); AIR; Catex; EQECAT; IBM; ImageCat; JBA
B2 Technology Oasis Palm Tree; Faraday; Navigators; SCOR; Talbot; Markel; AIR; Catex; EQECAT; IBM; JBA
C1 Hadoop ‘Show and Tell’ Oasis Palm Tree; IBM (BigInsights); NTT (Pivotal HD); Northdoor/Microsoft (HD Insight) …
C2 GIS ‘Show and Tell’ Oasis Palm Tree; Esri; ImageCat; RMA; Spatial Key …
EDP - What have we learnt? • Can automate formatted data
– Once formatted, data can be made interchangeable – Staging areas can be used to clean the data
• Can improve data quality and retain audit trail back to source
• Can reform data management + reduce unit costs – No longer constrained by rigidities and slowness of
traditional databases – Can use “Data Lakes” to review data and then transform
into databases for different user purposes – These are enterprise solutions not just for exposure and
cat modelling results and can be achieved in-house or using software houses
Improve Efficiencies
Formatted File Format
Clean
Possible Data Lake
Transform Load
Transformed and Cleansed
Canonical Data
Model
Review Source
Formatted File
Share
Analyse Source
Format Review Analyse Clean Load Share
Automate from format onwards
Validate
Validate
EDP - Management of Exposure Data
Calculation Engine
Input Exposures
and Policies
Outputs
User Interface
Input Models
Portfolio Results
Models Exposure Database Results Database
Workflow Reporting
Calculator
EDP - Filestream
SQLServer
Spreadsheet
e.g. Power Tools
Filestream Folder
view as source view transformed
Front-end Tools
EDP - Data Lakes
Idea of diagram thanks due to LOOM
Data Lake
Meta-data
Model
Mapforce
View Transform
XML, XSLT, SQL, MapReduce
EDP - Products
Business Implementation
Guidelines
Data and transformation
designs
Reference Computer Systems
Pilot Projects
LMA, LMA Website
ACORD, with link to LMA
LMA Market Processes (6 mths)
LMA Board sponsored
Non-property exposures Portfolio data lake Results data lake
Example BIGs
Information • Lloyd’s and LMA • Data Standards • Suppliers • Common Pitfalls • FAQs • Reference Implementations
Good Practice • Data Recording • Data Processing • Data Management • Catastrophe Models
US Property Binders
EDP - So what? • Efficiencies from Automation
– Redefine processes – Remove manual intervention – Re-use standard schemas and transformations
• Data Quality – Easier to attend to data if logistics sorted and tools provided – Can retain audit trail of versions and changes
• Data Management – Reduce reliance on any one external company – Scalable (can be done in volume), cheaper – NOT QUITE READY but … a fundamental change in enterprise
data management is underway • Options to do these in-house or use software houses • Reduce reliance on RMS
EDP - Implications • Automate data transformations to load exposure
and policy data • Provide full audit trail of data as they are changed
along the process • Manage very large datasets efficiently and flexibly • Create your own data stores for exposures and
cat modelling results • Use in-memory analytic tools now (many of these
features available within existing Microsoft products)
What Next? • EDP has shown there is potential for
– Streamlining data capture – Re-using data once captured automatically – Managing data with “Data Lakes”
• Market pilots to prove practical benefits A. Non-property exposures B. Property Exposures and Policies Data Lake C. Cat Modelling Results Data Lake
• Each to run for 6 months in 2014 if sponsoring Managing Agents volunteer
• Suggest (LMA produce) Invitation To Tender and get suppliers to quote
OSP - AGENDA • Background (Trevor Maynard) • Oasis Vision and Lloyd's • OSP Timeline • Uncertainties • Model Validation • So What? • How to get involved Questions
Goals of Oasis • Improve risk assessment through better
models, transparency, performance, and innovation
• Provide open source software and an open framework for model and software development
• Establish a commercially vibrant community of providers and users of Oasis software, models, data, and tools
Core functions of Oasis • Build and maintain robust software (official
model versions)…and front end tba? • Build develop and sustain the community • Develop and maintain web portal, working
parties and e-commerce platform • Stimulate education & innovation in
catastrophe loss modelling and associated data
Oasis LMF Members: 22 -> 40 • ClimateKIC • Lloyd’s • SCOR • Catlin • Validus • Ren Re • Hiscox • TigerRisk Partners • Cathedral • Novae • Zurich • Liberty • Aspen
• Aon Benfield • Guy Carpenter • Willis • Partner Re • Allianz • Axis • Amlin • Tokio Millennium Re/Kiln • Suncorp • JLTRe • GenRe • Swiss Re • Beazley
• Argo • Ark • Ascot/AIG • Barbican • Brit • Canopius • Chaucer • Hardy • Mitsui Sumitomo • QBE • R&Q • C V Starr • W R Berkley • XL
and many more in the pipeline …
2015 Strategically • Preparation of impact for OSP • Next phase of big model integration:
– US Quake – Euro wind & storm surge – Japan
• Service community • Education & outreach
Back
End
: Se
rver
s
Database Servers N
etez
za
Web Server
Mid
dle
Tier
: Dj
ango
WS
over
VM
in
a V
M
Data Storage
R FE
Javascript Front-End in browser
Fron
t End
: Re
fere
nce
web
inte
rfac
es
Django RESTFul WebServices
File System
MyS
QL
SQLS
erve
r
Python/SQL mid-tier
ODBC
Django Admin
Excel webservice FE
ODBC ODBC
MySQL - Django and ‘internal MySQL’
Grid & Cluster Other Technologies
External Web Services
3rd Party Apps – eg Windows C#, Android,
etc
Loading tools/ webservices
Linkages to Third Party Products
Dist
ribut
ed
SQLS
erve
r
Hado
op
Bare
-m
etal
(C)
html5
C++/C
Mod
el
xsd/xslt
Oasis Technical Architecture
OSP - Key Products Service OSP only
Available to market later
ValidationOasis Checklist (completed)MethodologyTools (e.g. reference vulnerability curves)Expert opinion on the sciencePalm Tree supportEvaluationTest planBenchmark exposure datasets for validationRun time environmentPalm Tree supportAdvice on hardware needsFront ends Extensible business front endLicensingShared service
OSP – Model Validation • The Science • Justification • Historical Event Results • Comparisons
– Event sets and footprints – Vulnerability Functions – Benchmark portfolios
• Fitness for Purpose • User Evaluation
Uncertainties in a Cat Loss Model
Damage Event
Ground-up Loss
Insured Loss
Loss to insurer
Interest
Policy
Primary – Event
Frequency
Clustering
Secondary - Event Footprint
Choice of peril intensity
Peril intensity
Secondary - Vulnerability
Vulnerability
Damageability
Exposure Data
Data quality
Location correlation
Stochastic Modelling
Discretisation error
Sampling error Socio-economic
“Loss amplification”
Legal judgements
ARA HurLoss focuses on
Occupancy - Industrial L.A., 1991, 2-story, Level 0 for all subtypes; 50 simulations,with 50,000-yr walkthrough each
TU: 40%BSW_C: 20%BSW_M: 20%W_E: 10%CBF: 7%LS: 3%
1000 Yr RP
Loss Distribution
Ensemble – Poor Data
Source: ImageCat, private communication
Better Data
CODA TypeL.A., 1991, 2-story, Level 2
TU: 100%
1000 Yr RP
Loss Distribution
Source: ImageCat, private communication
Example Oasis Output
15% share of $1m xs $1m location deductible of $500
Ground-up Loss AEP
Layer AEP
Event Loss Distribution
OSP - So What? • The deep view of risk shows the full range of
potential outcomes • This in turn affects decisions and numbers for:
– Risk Selection – Pricing – Exposure Management – Capital
• And raises the bar for many existing models
OSP – How to get involved • Q1 2015
– Front-end Design – Model Validation – Benchmark Data Creation
• Q2 2015 – (Core User) Testing
• Q3 2015 – Market Evaluation