Title: 2006 Seminar for the Appointed Actuary
1Canadian Institute of Actuaries
LInstitut canadien des actuaires
- 2006 Seminar for the Appointed Actuary
- Colloque pour lactuaire désigné 2006
2Stochastic Equity Modeling
- Dr. Julia Lynn Wirch-Viinikka
- AVP Investment Products Pricing
3Agenda
- Equity Risk where is it?
- Stochastic Modeling what is it?
- What options do we have for modeling equity risk?
- How do we start?
- How do we improve our model?
- How do we illustrate our results?
- How complicated can it get?
4Where is Your Equity Risk?
- Assets backing Liabilities (LTC)
- Surplus
- Liabilities
- Seg Fund, VA and UL Guarantees
- Equity Linked Products
- Fee Income based on Fund Value
- Hedging Mismatch
- Tracking error, basis risk
5Why Manage Equity Risk?
- Regulatory requirement
- Equity Limits, MCCSR requirements
- DCAT testing
- Valuation
- Reserves and Capital Requirements
- CGAAP, IFRS, US GAAP results
- Income and Surplus volatility
- Risk Management objectives
6Market risk vs. Insurance risk
- Traditional insurance risks, such as mortality
and longevity are less risky when pooled
together each individual follows their own
scenario and the insurance company pays off on
the average - Capital market risks dont diversify every
policyholder follows the same market scenario at
the same time
7Which Risks can be Managed?
- Risks that are identifiable and well understood
- Risks that are monitored and controlled
- Risks where there is the knowledge and expertise
to effectively manage them. - Where the reward is sufficient for the remaining
risk - Where financial instruments and methods are
available to hedge or control risk
8What is a Model
- Imitation/simplification of a real world system
- Tool that provides statistical estimates and not
exact results - Computational, statistical or judgment-based
- Helpful for product design and pricing,
valuation, forecasting, risk management,
financial reporting, and performance management. - Understand how your liability value changes over
time, when your liability value needs to be
calculated stochastically
9What is a Stochastic Model?
- A model that involves probability or randomness
- Random inputs (Normal, Lognormal, Uniform)
- Generally run many times (1000, 10000)
- Representative sampling (Yvonne Chueh)
- Distribution of outputs
- Estimates of statistics (mean, ile, std.dev)
- Error estimates (direct or bootstrapping)
10What is Model Risk?
- Model risk the possibility of loss or error
resulting from the use of models. - Model misspecification
- Assumption misspecification
- Inappropriate use or application
- Inadequate testing, validation, and documentation
- Lack of knowledge or understanding, user and/or
management - Error and negligence
11How do you Model Equity Risk?
- Flat Return(8) with an Extreme MfAD(-30)
- Set of deterministic scenarios (stress tests)
- Purchase sets of stochastic scenarios
- Stochastic Scenarios
- Normal/Lognormal Returns
- Autocorrelated Returns (time series)
- Regime Switching LogNormal (RSLN)
- One correlation matrix
- Different correlation matrices for each regime
- Other stochastic model (Wilkie, Smith, Lognormal,
Stoch Volatility, empirical) - Risk Neutral or Real World
12Yield Curve vs. Equity
- Are they related?
- Direct relation shows zero correlation
- However
- Bond Funds and Equity Indices show 30-60
correlation - Duration analysis can explain 90 of bond fund
returns - an( int int-1) Bond Fund Return (t-1,t)
- One way to connect Yield Curves with Equity
Returns - Leads to interest rates driving equity returns
13What Equity Risk do you model?
- Indices
- Stock Market Indices
- North America SP500, TSX, NASDAQ
- Europe FTSE, DAX
- Industry specific? Company specific?
- Public Equity / Private Equity
- Do you model
- Hedge Funds? Pass-through products?
- Real Estate? REITs?
- Credit Spreads/Counterparty Risk?
- Currency Risk?
14Is Equity Related to other Returns?
- NO ?? Independent
- Correlation Matrix (Normal/Lognormal)
- Regime Switching Assumptions
- Time Series, Volatility Jumps
- Macro-Economic Drivers (Wilkie Model)
- Does it matter?
- It depends on what you are trying to do
15Scenario Generators
- Issues
- Is the focus on the mean, median, or tail events?
- Economic vs. Risk Neutral model
- Calibration (current/historical data)
- Numerous Scenario Generators to choose from
- Desirable Characteristics to check for
- Integrated model
- Incorporates the principle of parsimony
- Flexible. A component approach.
- Beware Often there is a false sense of precision
16Why risk-neutral?
- Financial derivatives value depends on the value
of another financial instrument - Their prices do not depend on the particular
risk-preferences of the purchaser - so we can assume any risk-preferences
- Mathematically convenient to assume purchaser is
risk-neutral - If you project market movements along a
risk-neutral random walk and discount asset
payoffs at the risk-free rate, you will obtain
the fair value of that asset
17Fair Value
- Two portfolios with identical payoffs must have
the same price - ?Arbitrage - opportunity for profit buy the
less expensive portfolio and sell the more
expensive portfolio - FOR INSURANCE Liabilities
- No Arbitrage doesnt work perfectly the market
cannot freely buy and sell the insurance
liability - Risk-neutral pricing tells you what it would cost
to buy the same payoffs in the market. (not
necessarily a good estimate of the expected cost
of the guarantee if left unhedged)
18Risk-Neutral Valuation
- Brownian Motion
-
- µ expected risk free forward rate
- s implied volatility, e random error
- Does Risk-Neutral Market-Consistent?
- If µ and s are market-consistent, the prices that
the model produces are market consistent - Both µ and s can be functions of time
- s is often considered to be a function of market
levels (market volatility increases when market
levels fall)
19Real World Model
- Brownian Motion for the stochastic model
- Drift rate historical long-term avg returns
(not the risk-free forward curve) - Volatility long term average or stochastic
(GARCH, jump diffusion, RSLN) - Goal to reflect a reasonable distribution of
potential future returns - Fewer expected payoffs of the embedded option
than under risk-neutral valuation (on average,
the stock market has a better return than
risk-free investments) - Higher variability of profit by scenario
- The bad tail can be very bad
20Rule of Thumb
- Tail risk
- Use real-world valuation to measure tail risk
- Average cost
- Use real world inputs when you are willing to
accept the average result with a high amount of
variability - Use risk neutral when you want results (e.g. a
price or a profit measure) which you can be very
confident can be realized (through hedging)
21Who uses your Equity Models?
- Hedging (Financial Engineering)
- Market-consistent pricing - RN
- Risk Management, Valuation and Pricing (Actuarial
Modeling) - Tail exposures RW
- Volatility - RW
- Averages RW/RN
- Static Hedging - RN
- Dynamic Hedging RW/RN
22Regime Switching Models
- Discrete time (e.g. daily, monthly)
- Any model with different parameters in each
regime (Normal, AR(1), ARCH.) - 2-Regime Lognormal Monthly estimation software
free from SOA website - Very simple stoch vol model
- Tractable, intuitive, 2 Regimes are usually
enough for monthly data - 6 parameters m1, m2,
s1, s2, p12, p21 - Regime 1 Low Vol, High Mean, High Persistence
(small p12) - Regime 2 High Vol, Low Mean, Low Persistence
(large p21)
232-Regime LogNormal
- REGIME 1 r1
- Low Volatility s1
- High Mean m1
REGIME 2 r2 High Volatility s2 Low Mean m2
24Simple Stochastic Model
- 3-year 100 Seg Fund Maturity Guarantee
- MER 3
Regime 1 Regime 2
Fund LN1(11,16) LN2(-8, 20)
P(Switch) p124 p2122
Time in Regime Time in Regime
Regime 1 84.6
Regime 2 15.4
Mean Std.Dev
8.4 18.1
25Simple Stochastic Model Scen 1
- 3-yr Maturity Guarantee No death / lapse
- Initial Deposit 1 Top-up 0
Time 0 1 2 3
Uniform RAND()0.934 0.641 0.135 0.053
Regime 2 2 1 1
Normal NORMSINV(RAND()) -0.1635 0.7642 0.9195
Return (1it) expmut sigmasqrt(t)RN 0.89 1.26 1.29
Fund Fund(t-1)Return 0.89 1.13 1.46
Fund Less MER FundLessMER(t-1)Return (1-MER) 0.87 1.06 1.33
26Simple Stochastic Model Scen 2
- 3-yr Maturity Guarantee No death / lapse
- Initial Deposit 1 Top-up 0.19
Time 0 1 2 3
Uniform RAND()0.649 0.039 0.827 0.154
Regime 1 2 2 1
Normal NORMSINV(RAND()) 0.1635 -0.7642 0.3195
Return (1it) expmut sigmasqrt(t)RN 0.95 0.79 1.17
Fund Fund(t-1)Return 0.95 0.76 0.88
Fund Less MER FundLessMER(t-1)Return (1-MER) 0.93 0.71 0.81
27More Advanced Stochastic Models
- Other Modeling Considerations
- Death and Lapse (dynamic lapse?)
- Death Benefits and Living Benefits
- Ratchets and Resets
- Policyholder Behaviour
- Commissions / Surrender Charges / DAC
- Reserves / Capital
- Net Income / Tax / Distributable Earnings
- Discount Rates for Present Values
- Illustrating Results
- Hedging Strategies
28Summary Statistics
- Mean, Standard Deviation, Skewness, Kurtosis,
- Percentiles (Quantiles)
- Confidence intervals http//www.fenews.com/fen47/
topics_act_analysis/topics-act-analysis.htm - CTE 95 Mean of worst 5 of results
- Variance Estimate Hancock and Manistre NAAJ
9(2) 129-156
29Box Plots
30Histograms and CTEs
- Histogram of scenario outcomes
31How Many Scenarios are Enough?
- Convergence / Sampling error
- Variance Reduction Techniques may help
- Many techniques work for averages not tails
32Are you taking a Holistic Approach?
- ERM Approach takes advantage of synergies across
products - Consistent set of RW and/or RN scenarios used for
all lines of business - Projections aggregated by scenario across lines
of business - Yield curve and equity return assumptions must be
consistent - More difficult if two Tier Stochastic simulation
is required
331-Tier Stochastic Simulation
- Projected Liability Payouts
- Can determine t0 reserve(CTE70-80 and TBSR
(CTE95) - Can determine liability payout projections
- Can not accurately determine future reserve and
capital projections (approximations NPATH,
Black-Scholes)
V0
0
T
Time
342-Tier Stochastic Simulation
- Projected Liability Payouts, Reserves, Capital,
Net Income .
- Can determine t3 reserve for each stochastic
scenario (CTE70-80) - Can determine future capital needs and net income
projections - Much more time consuming
V0
0
T
Time
352-Tier Stochastic Simulation
- Projected Liability Payouts, Reserves, Capital,
Net Income .
- Much more time consuming
- 1000 Tier 1 Scenarios
- 10 time steps each
- 100010 points to perform a second tier
simulation - 500 scenarios at each point 5,000,000 Tier 2
scenarios
V0
0
T
Time
36Insurance Options
- Embedded options in insurance liabilities are
different from financial options - Sub-optimal exercise behavior
- FPDA can pay surrender charges and get a new
contract if new money rates rise - Evidence PHs are inefficient in using this
option - Some PH will not surrender their contracts no
matter how uncompetitive their renewal rate - Segregated Funds (VA/VL) GMAB should invest in
the most aggressive funds available - CAPM more risk implies more return
- Evidence PHs invest in conservative and balanced
funds
37Stochastic Modeling Challenges
- Option payoffs that depend on policyholder
behavior will reflect - Historical behavior patterns
- Actuarial judgment
- Path-dependent behavior (ie. lower lapses for in
the money guarantees) can be modeled - Introduces uncertainty to valuation results
- Practitioners have argued about the proper way
to model behavior in a risk-neutral framework - (library.soa.org/library-pdf/RRN0608.pdf by M.
Evans)
38Stochastic Modeling Challenges (Continued)
- Long-term nature of liabilities
- Expected market forward rates past 30 years is
needed for valuation - Instruments that will hedge the yield curve past
30 years or equity risks past 10 years are
illiquid or unavailable - Computational Requirements
- Distributed processing (AXIS, MatLab, .)
- 2-Tier Stochastic Analysis (Stochastic-in-Stochast
ic)
39Conclusions
- Equity risk is not like traditional insurance
risk. - Stochastic Modeling is a tool that can help us
understand complex dynamic processes. - Start simple and build.
- Test uncertain assumptions.
- Develop expertise.