identify and create value through data analytics across the credit cycle in consumer credit. Presentation at EFMA consumer credit conference by george georgakopoulos
identify and create value through data analytics across the credit cycle in consumer credit. Presentation at EFMA consumer credit conference by george georgakopoulos
Presentation for 2014 Valuation Actuary Symposium (New York).
After an introduction to the history of variable annuity financial modeling and current modeling paradigms, this presentation covers the unique modeling considerations related to variable annuities.
"The Future of Financial Services”, organized by Capco and NYU-Poly
Presenter: Allen Ferrell
Greenfield Professor of Securities Law
Harvard Law School
June 16, 2011
Demystifying Flexible Staffing's Role in Today's Labor Market & EconomyBeeline
Andrew Steinerman, JP Morgan
Business & Education Services
Mr. Steinerman explored how flexible labor is a concurrent indicator of the economy and a leading read on labor. According to Mr. Steinerman's research and relative to past recoveries, flexible labor has seen a faster lift off the bottom in the current cycle, as companies have recognized the value of labor flexibility in the uncertain economic environment following the great recession.
This compelling look into the economies of scale for flexible labor also showed how:
The professional segment drives flexible staffing growth and profitability
Flexible staffing is still a secular growth industry
Global trends in flexible staffing penetration
Jessie, PhD, PE, City of Sugar Land, TX and Sunil Kommineni, PhD, BCEE, Malcolm Pirnie, Houston, TX discuss the City's plans to meet the Fort Bend Subsidene District's mandated reduction of groundwater use by treating water from Oyster Creek.
Presentation for 2014 Valuation Actuary Symposium (New York).
After an introduction to the history of variable annuity financial modeling and current modeling paradigms, this presentation covers the unique modeling considerations related to variable annuities.
"The Future of Financial Services”, organized by Capco and NYU-Poly
Presenter: Allen Ferrell
Greenfield Professor of Securities Law
Harvard Law School
June 16, 2011
Demystifying Flexible Staffing's Role in Today's Labor Market & EconomyBeeline
Andrew Steinerman, JP Morgan
Business & Education Services
Mr. Steinerman explored how flexible labor is a concurrent indicator of the economy and a leading read on labor. According to Mr. Steinerman's research and relative to past recoveries, flexible labor has seen a faster lift off the bottom in the current cycle, as companies have recognized the value of labor flexibility in the uncertain economic environment following the great recession.
This compelling look into the economies of scale for flexible labor also showed how:
The professional segment drives flexible staffing growth and profitability
Flexible staffing is still a secular growth industry
Global trends in flexible staffing penetration
Jessie, PhD, PE, City of Sugar Land, TX and Sunil Kommineni, PhD, BCEE, Malcolm Pirnie, Houston, TX discuss the City's plans to meet the Fort Bend Subsidene District's mandated reduction of groundwater use by treating water from Oyster Creek.
Weather Outlook - Dr. Elwynn Taylor, Climatologist, Ag Meteorologist, Iowa State University, from the 2012 World Pork Expo, June 6-8, Des Moines, Iowa, USA.
1. Modeling and Managing
Basis Risk
Society of Actuaries
Equity-
Equity-Based Insurance Guarantees Conference
Boston MA, Oct 12th 2009
Dr. Pin Chung, Vice President, Allianz Investment Management
Dr. Thiemo Krink, Director, Allianz Investment Management
1
2. Agenda
§ Overview
§ What is Basis Risk
§ Sources of Basis Risk
§ Fund Mapping
§ Example
§ Basis Risk Management
§ Conclusions
2
7. What is Basis Risk?
Basis risk is the deviation between
expected vs. actual funds performance
Under-
Under- or over-performance
over-
relative to benchmark portfolio
Correlation risk
7
8. Tracking Error
n
Tracking Error = ∑(R
i =1
p.i − RB,i )2 /(n − 1)
Measure the severity of deviation
Annualized volatility of the difference
between return of fund and benchmark
5% tracking error commonly allowed
for actively managed funds
8
9. Possible sources of Basis Risk
Possible Sources
Changes of fund manager’style
s
Deviation of benchmark & surrogate portfolio
Unanticipated expenses (fund, tax, etc.)
Changes of policyholder’investment strategy
s
Changes of funds’return profiles
Changes of indices correlation relationships
9
10. Fund Mapping
Fund mapping goal
To reflect the systematic components
(beta coefficients) of the underlying funds
across selected hedging indices
Considerations
How many indices to use?
How long period of data to use?
How frequent to update model?
10
11. Fund Mapping Methods
Fund mapping methods
Seriatim on seriatim: Fund Level
Forced asset allocation: Investment Style
Others
11
12. Seriatim on seriatim method
Collect historical daily or weekly returns of each
fund and candidate hedging indices return
profile for 3 to 7 years
Perform OLS regression on each fund to
indices to obtain beta coefficients for each
fund
One could apply different weighting
scheme to return data; e.g., equal
weighted or exponential decay
12
13. Forced asset allocation method
Collect historical daily or weekly returns profile
of each asset allocation and candidate hedging
indices return profile for 3 to 7 years
Perform OLS regression on each asset
allocation to indices to obtain beta
coefficients for each asset allocation
One could apply different weighting
scheme to return data; e.g., equal
weighted or exponential decay
13
14. Seriatim on seriatim
Mutual Fund 1 Replicating Portfolio 1
Name 1, 1 Index 1
Name 1, 2
Index 2
Name 1, 20
Index 3
Index 1
Mutual Fund 2 Replicating Portfolio 2
Name 2, 1 Index 1
Name Index 2
Index 2
2, 2
Name 2, 15 Index 3
Index 3
Mutual Fund 3
Replicating Portfolio 3
Name 3, 1 Index 1
Name 3, 2 Index 2
Name Index 3
3, 25 14
15. Forced asset allocation
Aggressive Replicating Portfolio 1
Mutual Fund A, 1 Index 1
Mutual Fund A, 2
Index 2
Mutual Fund A, 30
Index 3
Index 1
Moderate Replicating Portfolio 2
Mutual Fund M, 1 Index 1
Mutual Fund Index 2
Index 2
M, 2
Mutual Fund M, 20 Index 3
Index 3
Conservative
Replicating Portfolio 3
Mutual Fund C, 1 Index 1
Mutual Fund C, 2 Index 2
Mutual Fund Index 3
C, 10 15
16. Example (forced allocation)
Investment style Underlying fund
Equity fund
Aggressive
Balance fund
Moderate
Bond fund
Conservative
Real Estate fund
16
17. Example (continued)
§ In matrix form:
Equity Fund
Aggressive Allocation 80 10 7 3
Balance Fund
Moderate Allocation = 15 70 10 5
Conservative Allocation 5 15 70 10 Bond Fund
RealEstate Fund
Investment Styles Policy Weights Underlying Funds
n
Note that: ExpReturn(Equ ityFund) = ∑ wi ⋅ ExpReturn( Namei )
i =1
17
18. Example (continued)
§ Regress return of each investment style to return
of three liquid and tradable indices, e.g., SPTR,
AGG, and EAFE to obtain the following:
Aggressive AssetAllocation 0.72 0.13 0.15 SPTR
ModerateAssetAllocation = 0.55 0.12 0.33 AGG
ConservativeassetAllocation 0.14 0.75 0.11 EAFE
0.83
2
With R =0.76 For given
Σ3×3
0.82
18
19. Example (continued)
§ Map each investment style to four
hedging indices, SPTR, AGG, EAFE, RUT
SPTR
AggressiveAssetAllocation 0.63 0.08 0.15 0.14
AGG
ModerateAssetAllocation = 0.45 0.22 0.13 0.20
ConservativeassetAllocation 0.08 0.72 0.05 0.15 EAFE
RUT
0.87
2
With R =0.79
For given
Σ4×4
0.84
19
20. Deficiencies of procedure
What went wrong?
Less quick access to fund composition
Fund manager deviates from policy mandate
Policyholders exercise leeway too frequently
Linear relationship breaks down
Correlation relationship breaks down
20
21. Basis Risk Management (continued)
Fund Managers
Limit mutual funds choice
Establish direct relationship
Provide “
real time”allocation
21
22. Basis Risk Management
hedge develop price in
more indices, contingency basis risk
currencies plan
set review fund
aside capital performance
VA writers
overlay qualitative lower number
considerations of funds
frequent
direct contact use index, ETFs,
monitor fund
with managers passive funds
mapping
22
23. Conclusions
§ Basis risk is a risk with great potential to cause
severe financial loss.
§ A few ways to reduce basis risk:
Ø Moving away from actively managed funds into
indices, ETFs or passive funds.
Ø Simplify product design and price in the basis risk
component.
Ø Increase frequency of monitoring fund mapping
procedure.
Ø Have a back-up plan to react appropriately.
back-
23
24. Q&A
§ A journey of a thousand miles begins
with a single step. Lao-tzu
Lao-
§ Thanks for your attention. Questions?
Dr. Pin Chung, Vice President, Allianz Investment Management
pin.chung@allianzlife.com,
pin.chung@allianzlife.com, (763) 765-7647
765-
Dr. Thiemo Krink, Director, Allianz Investment Management
thiemo.krink@allianzlife.com,
thiemo.krink@allianzlife.com, (763) 765-7979
765-
24
26. Ordinary Least Squares (OLS)
General linear model: yi=β0+β1xi,1+β2xi,2+…+βp-1xi,p-1+εI
+…+β i,p-
yi is the ith value of the dependent variable
[Fund returns]
β0,β1,…,βp-1 are the regression parameters
,…,β
[Beta coefficients]
xi,1,xi,2,…,xi,p-1 are known values of independent variables
i,p-
[Indices returns]
εi is the independent random error, with N(0,σ2)
N(0,σ
[Residuals]
26
27. Review of OLS (continued)
To estimate β , we minimize the total sum of squares S,
where:
S=Σεi2=Σ(yi-β0-β1xi,1-β2xi,2-…-βp-1xi,p-1-εi)2, where i=1,2,…,n
S=Σε i,p-
Next, simultaneously solving the normal equations:
∂S/∂β0=0, ∂S/∂β1=0,…,∂S/∂βp-1=0
∂S/∂β ∂S/∂β =0,…,∂S/∂β
In matrix form: minimizing S=(Y-Xβ)’ -Xβ)
S=(Y (Y
By solving: ∂[(Y-Xβ)’ -Xβ)]/∂β=0
∂[(Y (Y )]/∂β
With the resulting estimator for β expressed as:
b=(X’ -1X’
X) Y
27
28. Review of OLS (continued)
Let mean(Y) be the mean of the observed values;
mean(Y
fit(Y
fit(Y) be the vector of the predicted values;
mean[fit(Y
mean[fit(Y)] be the mean of the predicted values;
e denote the vector of residuals from the model fit:
e(nx1) =Y-Xb=Y-fit(Y)
Xb= fit(Y
Let SStotal = SSreg+SSerr
SStotal = Total sum of squares
= [Y-mean(Y)]’ -mean(Y)]
[Y mean(Y [Y mean(Y
[Y
SSreg = Regression sum of squares
= {fit(Y)-mean[fit(Y)]}’ Y)-mean[fit(Y)]}
{fit(Y mean[fit(Y {fit( mean[fit(Y
{fit(Y
SSerr = Residual sum of squares
= {Y-mean[fit(Y)]}’ -mean[fit(Y)]}
{Y mean[fit(Y {Y mean[fit(Y
{Y
28
29. Review of OLS (continued)
§ The coefficient of determination is:
R2=SSreg/SStotal
§ R2 compares explained variance with total variance
§ Higher R2 indicates regression fits data better
§ Be cautious not to over fitting the model
§ Use modified R2 to avoid spurious regression
29