1. Equity-Based Insurance Guarantees Conference
May 31 – June 1, 2010
Tokyo, Japan
Risk Managing Living Benefits
Andy Rallis, David Schrager, Denys Semagin
2. 1
VA Risk Management
David Schrager Denys Semagin
Head of Financial Engineering Head of Quant Development
ING Life Japan, Annuity Risk Management division
June 1st 2010 (13.45 – 15.45 hours)
www.ing.com
Agenda
David Schrager
• ING Experience in Financial Crisis
• Lapse modeling in reserving / pricing
• Risk management
• Different approaches
• Suitability for financial management
Denys Semagin
• ING Insurance Library
2
3. 2
What Worked in the Crisis
• Hedge program never stopped
• Goal: Replicate Guarantee Value movements using financial derivatives to reduce P&L volatility
Hedge programe Hedge target Risk factors Hedge instrument Hedge coverage Target hedge ratio
EQ delta hedge First order delta claims TOPIXX, EUROS EQ index futures First order items 100%
IR delta hedge First order delta claims JPY, EUR, USD IRS First and higherorder items 80-90%
FX delta hedge g First order delta claims EUR/JPY, ,
USD/JPY FX forwards First order items 100%
Rho hedge First order rho claims and premium JPY, EUR, USD IRS First and higherorder items 100%
• Enough liquidity for all hedge instrument
• FX forwards, EQ futures, IRS markets are very liquid
• Infrastructure puts asset / liability risk management a ”risk platform”.
• Address time zone differences
• Providing liability model(s) to project the market risk (“Greeks”) using Monte Carlo Simulation
• Computer system platform cooperating with other ING Insurance Entities
DataMart
Trading
System DB
ING
Insurance
Library
ING F FO
- Trading (EQ, FX,IRS)
ING F BO
- Recon
ING Insurance
-Monitoring / maintenance
-reporting / setup
Accounting
system
Broker BBG
SPVA site
Trading
application
server
Problems Faced in the Crisis
• Basis risk for Active fund
• Non disclosed position: key rate duration (interest rate sensitivity) was not available for several
bond fund (they do not provide) which were unhedged. From 2009, after negotiation for key rate
duration, we reduce the unhedged exposure from 10% to 1%.
• Active fund performed worse than the index during the crisis.
• Delta hedging P&L results are driven by assumptions
• P&L at issue is driven by pricing vs. reserving
• P&L over time is driven by economic assumptions vs. realization
• Hedge instruments mismatch
• Mismatch of hedge instruments and assets.
• More pronounced in crisis.
• Interest rate spread on cash bonds compared to swaps reversed.
• Accounting challenges
• Often the liability is measured at historic or locked-in assumptions while derivatives used to
hedge the liability are fair valued
• Timing differences when different markets close
• Lapse model (Customer behaviour)
• Although lapse leads a lower account value, customers chose lapse due to the lack of money in
the crisis.
• The current lapse rate function is jumping around even under low volatility environment.
• As a result, Greeks become more random than necessary, which worsens hedge performance.
4. 3
Risk Management at ING
Summary:
1. Product design
• Transfer or limit risk according to sensible risk appetite
2. Product Pricing
• What products? What markets? What price and when?
Then the rest follows…
3. Risk measurement
• Need to set limits and ensure limit monitoring in place
4. Risk mitigation
• Cannot prevent failure with good risk management after pricing stage,
that’s where all the damage is done
5
Agenda
David Schrager
• ING experience in financial crisis
• Lapse modeling in reserving / pricing
• Risk management
• Different approaches
• Suitability for financial management
Denys Semagin
• ING Insurance Library
6
5. 4
Modern Economic Valuation Principles
• Valuation of the embedded options in VA contracts requires
understanding of:
• Option theory: risk neutral valuation principles
• Policy holder behavior: clients right to lapse
• Different policy holder behavior modeling approaches can be used:
• Traditional actuarial
• Econometric
• Early exercise behavior
• Rational exercise of the right to lapse
• Eventually valuation and hence reserving principles should be judged
by ensuing effects on key financials and financial health of a company
7
Traditional Actuarial Approach
The following principles are at work
• Best estimate client behavior
• “Expert” opinion
• No / minimal interaction with underlying economics of the contract:
• SV / SP or SV / K
This has the following effects on the key financials of the company
• P&L necessary different from zero, both positive or negative because
experience will deviate from assumptions
• Capital is required for this risk
•• There’s no built in mechanism that guarantees beforehand whether the
product will be profitable
Conclusion
• Approach is limited and unsuitable for financial management in modern
financial industry, it’s great if you want to “support the business”
8
6. 5
Econometric Approach
The following principles are at work
• Client behavior is function of explanatory variables
• Empirical data based
• Interaction with underlying economics by using variables related to that like:
• Moneyness / Forward moneyness (i.e. IR and M&E + AMF corrected) Time to maturity
• Age / Gender Fund volatility
• Tax laws Distributor
This has the following effects on the key financials of the company
• P&L necessary different from zero (likely less than with pure deterministic
approach), both positive or negative because experience will deviate from
assumptions
• Capital is required for this risk
• There’s no built in mechanism that guarantees beforehand whether the product
will be profitable
Conclusion
• Approach is justifiable but unsuitable for financial management in modern financial
industry where capital is expensive and investors/governments demand solid
returns and capital positions
9
Early Exercise Approach
The following principles are at work
• Client behavior is worst case from financial economic perspective of company
• Rational behavior based
• Interaction with underlying economics by using variables related to that like:
• Moneyness / Forward moneyness (i.e. IR and M&E + AMF corrected) Time to maturity
• Age / Gender Fund volatility
• Tax laws
This has the following effects on the key financials of the company
• P&L from this source is necessarily positive because experience will deviate from
assumptions
• No capital is required for this risk
• There’s built in mechanism that guarantees beforehand that the product will be
profitable
Conclusion
• Approach is only possible one in modern financial industry where capital is
expensive and investors/governments demand solid returns and capital positions
10
8. 7
Example of Bad Product Design
• Consider standard MGDB+MGSB w/o ratchet or clicks
• Strike = SP
• AV(0) = SP
• No surrender charge
• M&E fee is fixed
13
Risk Management by Product Design
Alternatively policyholder behavior risk can be mitigated by product design
• More sophisticated surrender charge mechanism
• Time to maturity
• Moneyness
• VIX level
• …
• Adjustable M&E fee
• Up-front fee payment
14
9. 8
Agenda
David Schrager
• ING experience in financial crisis
• Lapse modeling in reserving / pricing
• Risk management
• Different approaches
• Suitability for financial management
Denys Semagin
• ING Insurance Library
15
ING Insurance Library: What is it?
IIL is a powerful in-house software library that could be combined
with any type of user interface and has unlimited potential to
support valuation for any type of products in ING Group.
• A calculation engine that provides risk and other metrics for ING insurance products
• A software framework using re-usable code that will facilitate rapid delivery of new
products and enhancements moving forward
• A replacement for old liability models that are not flexibly callable, not easily extendable,
and operationally insecure.
• Passed major audits in Q2 2009 and in Q1 2010
• Includes database for secure and auditable storage of inputs and outputs
• Other business units within ING Group consider IIL for risk management related tasks
10. 9
ING Insurance Library: Business Impact Summary
• Basis Risk better addressed in IIL => reducing P&L volatility of Basis Risk by 20%
• Intra-day hedging targeted through IIL => reducing P&L volatility
• Hedging and control of P&L volatility are more fair and efficient with tunable risk factor
modeling.
• All currently hedged products (9) are in IIL (modeled and approximated)
• Around 2 months for full development and testing cycle targeted for new VA products
• Cooperation within ING group for hedging, stochastic models/calibration development
ING Insurance Library: Business Alignment
Financial:
• Better attribution and control of P&L volatility
• Improved risk management to protect solvency and ratings
• Reduced hedging costs
Operational:
• Faster time to new products
• Fewer operational errors
Compliance:
• Single repository of data for risk regulatory and reporting purposes
• Tight controls on processes, separation of duties
• Secure data storage
11. 10
ING Insurance Library: General Functional Diagram
IIL Request Service
AFRM DataMart
Enterprise Data
Inforce
Market
Fund Assumptions
Product Assumptions
Actuary Assumptions
Mainframe
FRM
Server Server
Server Server
Server Server
Server Server
Server Server
IIL Client
Batch
IRM ALM
Server Server
Server Server
Server Server
Server Server
Server Server
IIL Request Service
Pluto
IIL Service Data
Sophis
Hedging Model
Market Trades
Moses
Reporting
ING Insurance Library: General Architecture Overview
ING insurance library framework
Enterprise IT
External systems Market Trades, IFRS, Special Reporting, Research and Development, Prototyping, Etc.
Sophi
s
Excel Othe
rs
Message interface Message interface Message interface
Library request message interface (xml)
Enterprise Data Base:
Market data
In-force data
Assumptions data
Etc.
Message interface
MoSe
s
Message Transport
Profitability
Analysis
Reporting
Valuation
Components
IIL Data
Base
Data
interface
Sensitivity
Analysis
Queries
12. 11
ING Insurance Library: Facts Sheet
Item Legacy Approach ING Insurance Library
Product flow development 4 months (1 product) 6 weeks (2 products)
Example of atomic module development
Add multi-multi threshold lapse model 3 weeks (3 products) 1 week (all products)
Daily hedge runtime
500,000 contracts 3h05’ 2h35’
(valuation + hedge sensitivities)
Post processing 1h50’ 0h20’
Total ~5 hours ~3 hours
Scalability and configuration
Risk types supported 0 14 (EQ+IR+FX, configurable)
Fund groups supported 8 (fixed) 46 (configurable)
Add risk factor N/A No development required
Add fund group Development required No development required
Data output formats Fixed Configurable
Load Balancing/Resources None/Fixed Configurable/Scalable
ING Insurance Library: Architecture And Road Map
Architecture type: Multi-tier client-server
Component communication:
• Messages: standardized XML
• Message transport channel: message queue (activated on server side)
Components:
• Service clients application (manual control on user side or automated on server side)
• Service application (calculation service, server side only)
• Resource manager / load-balancing applications (server side only)
• Configuration and logs: flat files (server side only)
•• Databases and GUI: Oracle and web service (server side only)
13. 12
ING Insurance Library: Current System Architecture
Client:
Enterprise
Data
IIL DB
Internet
Queries
Data Storage
Scheduled
Automated
Client
Web DB GUI
Queries
Queries
Server
Server-side applications:
Resource manager / Load balancer / Logging
Message Broker
Request /
Response
Messages/Responses
network / Server Farm
intranet
Restricted PRODUCTION Environment
Client:
IIL Run GUI
ING Insurance Library: Remote Access to IIL/DB Operations
Corporat
e
I t t
Local Intranet Area
Remote
User
Intranet
Area
Web DB Client
URL
Remote Desktop IIL Service / DB / Sophis
Dedicated PC for
remote access
14. 13
ING Insurance Library
DB Web GUI and data control
Summary panel in DB Web GUI readable by human user
• Each set of data tables will be grouped by a Run ID and referred as a
component in a IIL request message
• Data Settings
• Data comparison
• Data validation
• Data approval
• Audit trail of user and modified data
• User rights
• Grouping data for Runs
• Running for a set of data grouped by Run ID
ING Insurance Library: Process Flow Diagram
Data up loader IIL
Inforce
M k t
IIL
Service
Mainframe
Enterprise
Data
Market
data
Assumptions
IIL DB
Assumptions
GUI
Load
Balancing
BBG ftp
User
Output
PV/Greeks
Automated
Client
Run
Manager
Automation Process (inforce, client-service, operator emails, etc)
15. 14
ING Insurance Library: Data Flow in Full Trading System Set-up
Trading
Enterprise
Acc.
Data
Broker BBG
system
DB
Data
ING
Insurance
Library
IIL
DB User group B
Trading
system
application
server
-Monitoring - Trading
- Reconciliation - monitoring
-maintenance
(EQ, FX,IRS)
-reporting
-setup
User Group A
ING Insurance Library: Model Flow in IIL Run
External processes and Memory Cache
Model Calibration Fund Regression
Result storage Inputs
Risk Factor
Projection
Risk Factor
Mapping to Funds
SPVA Liability
Projection
Generation of
Results
IIL Process Engines and Object Cache
16. 15
ING Insurance Library
DB Web GUI and XML request message - (illustration)
Summary panel in DB Web GUI readable by human user
• Each set of data tables will be grouped by a Run ID and referred as a
component in a IIL request message
IIL request message template: run specs readable by service
• A IIL request message is a root component with optional sub-components
• <Request>
• <Component>
• <Name>FRMProcessFlow</Name>
• <Type>ProcessFlow</Type>
• <Parameters></Parameters>
• <Outputs></Outputs>
• <Component><-- RF projections -- >
• <Component><-- HW Model -- ></Component>
• <Component><-- BS Model -- ></Component>
• <Component><-- FX Model -- ></Component>
• </Component>
• <Component> <-- SPVA evaluation -- >
• <Component><-- Lapse -- ></Component>
• <Component><-- Mortality -- ></Component>
• <Component><-- AV Model -- ></Component>
• <Component><-- Click Model -- ></Component>
• <Component><-- SP Model -- ></Component>
• </Component>
• </Component>< -- Data component -- ></Component>
• </Component>< -- Etc… -- ></Component>
• <Reply>
• <Status></Status>
• </Reply>
• </Component>
• </Request>
ING Insurance Library: Functional Stratification of IIL Service
Utilities
Components
Data Cache Set Up
IIL Objects Set Up
Process Flow Set Up
Flow Engines
Individual Models
Message
MESSAGE
Evaluation
Layer
Internal
Set Up
Layer
External
Set Up
Layer
IIL SERVICE AREA
TRANSPORT
AREA
17. 16
ING Insurance Library: Stochastic Modeling Framework for VA
Financial Market Model Actuarial Models
Calibration based
on Market Data
Calibration based
on risk and return
assumptions
Risk Neutral
Simulation
Real World
Simulation
Scenarios for risk factors
Mapping of risk factors to
investment funds
Scenarios for investment
f d
Variable Annuity Liability
Models
Dynamic Policy
holder behavior
models
Mortality models
(static / trend / …)
Surrender Charge
model
funds
Time Steps
Standard DB+SB / Click / Ratchet / Best-Of / IB / WB
Risk Reporting
PV DB+SB / PV Fee / Deltas / Gammas / Cross-Gammas / Rho / KRD
ING Insurance Library: Flexibility of Evaluation Flow in IIL Service
Run same evaluation flow with a
different interest rate model, reporting
t t i lb ifi th
Results
Calculation
SPVA Evaluator
Hull-White IR Model
Results
Calculation
SPVA Evaluator
BDT IR Model
type, etc, simply by specifying the
new models in the evaluation request.
SPVA Evaluator
Replace this component
Zero Rate Curve
Bootstrapping
Swap
Curve
Libor
Curve
Foundation Layer
Zero Rate Curve
Bootstrapping
Swap
Curve
Libor
Curve
Foundation Layer
p p
Zero Rate Curve
Bootstrapping
Swap
Curve
Libor
Curve
Foundation Layer
18. 17
ING Insurance Library: flexible model building blocks and flow invariance
Lapse rate: Account Value model
l = F(moneyness,
lapse assumptions)
Surrender
Charge
Assumprtions
Strike model
Moneyness Model
Surrender Charge
Payoff
lapse_NSC SV (left):
moneyness = AV_1 / Strike_1
Guarantee model
SC SS (right):
moneyness = Lapse Rate Model
Lapse
Assumptions
AV_2 / Guarantee
ING Insurance Library: Development Summary
• Includes state of the art stochastic models
• Database and load balancing are parts of IIL framework
• Supports rapid development through code re-use
• Software can be developed in and shared among different locations
• Service can be deployed to other locations on Windows (or Linux)
• Fully supports hedge program by providing ALL hedge sensitivities as well as
Performance Attribution