1 the arbitrage-free equilibrium pricing of liabilities in an incomplete market: application to a...
Post on 22-Dec-2015
215 views
TRANSCRIPT
1
The arbitrage-free equilibrium pricing of liabilities in an incomplete market:application to a South African retirement fund
Hacettepe University 27/6/2011
Rob Thomson
2
Agenda
1. Introduction
2. Liabilities specification & model
3. Pricing method
4. Results and sensitivity
5. Conclusions
3
Introduction: aim
Apply the pricing method of Thomson (2005) to:• market-portfolio model (Thomson, unpublished);• equilibrium asset-category model (Thomson & Gott, 2009);• a DB retirement-fund model;with a view to operationalising the pricing of such a fund and
quantifying the effects of:• non-additivity due to incompleteness;• guarantees implicit in reasonable expectations of pension
increases; and• the sensitivity of the price of illustrative liabilities to
sources of risk and parameters of the model.
4
Introduction: method
Market-portfolio model
Return on market
portfolio
Mortality model
Liabilities specification
Risk-free return
Inflation rate
Salary model
Equilibrium asset-categories
model
Returns on asset
categories
Pension-increases
model
Asset-liability model
Price of liabilities
5
Introduction: method
Market-portfolio model
Return on market
portfolio
Mortality model
Liabilities specification
Risk-free return
Inflation rate
Salary model
Equilibrium asset-categories
model
Returns on asset
categories
Pension-increases
model
Asset-liability model
Price of liabilities
6
Introduction: method
Market-portfolio model
Return on market
portfolio
Mortality model
Liabilities specification
Risk-free return
Inflation rate
Salary model
Equilibrium asset-categories
model
Returns on asset
categories
Pension-increases
model
Asset-liability model
Price of liabilities
7
Liabilities specification & model
• no exits before retirement
• mortality only after retirement
• projected unit method
• salaries and pensions expressed in real terms
8
Liabilities specification & model: salaries model
, 1 expmt m t t tS S -
1 3 2 7t t t tb b
t x x t
expx x -
expx x -
22
, 1
x
xx tM
-
0,01
1 0,005b -
2 0,005b 0,03
0,016 0,5 0,1
0,042 0,5
0,08
10
Liabilities specification & model: pensioner mortality
00{ }98 98
{ } { }00{ }
PNLxSAP SAIL
x xILx
{ } { } 1 expSAP SAPx t x t t -
98{ } { } exp 10SAP SAP
x x
where:
, 1 7t t t tb - 0,004 -
0,001b -
0,005
11
Pricing method: primary & secondary simulations
primary simulation secondary simulation
Time T -1 Time T
primary simulation node
Time 0 Time 1 Time t -1 Time t
12
Pricing method: state-space vector
1
1
1
N
It
It u
Ct
tCt u
t
x t
x t
P s
P s
P s
P s
P
P
x
where:
exp ( )It ItP s Y s -
exp ( )Ct CtP s Y s -
expt t
13
Pricing method: primary simulations of the state-space vector
time 0 time 1 time 1 time T - 1
primary simulation 1 x 11 x 21 x T - 1,1
primary simulation 2 x 12 x 22 x T - 1,2
x 0
primary simulation I x 1I x 2I x T - 1,I
14
Pricing method: secondary simulations
time T - 1 time T time T - 1 time T
x* T 11, p* T 11 x* T 11, p* T 11
x T - 1,1 x* T 12, p* T 12 x T - 1,1, p T - 1,1 x* T 12, p* T 12
x* T 1J, p* T 1J x* T 1J, p* T 1J
x* T 21, p* T 21 x* T 21, p* T 21
x T - 1,2 x* T 22, p* T 22 x T - 1,2, p T - 1,2 x* T 22, p* T 22
x* T 2J , p* T 2J x* T 2J , p* T 2J
x* TI 1, p* TI 1 x* TI 1, p* TI 1
x T - 1,I x* TI 2, p* TI 2 x T - 1,I , p T - 1,I x* TI 2, p* TI 2
x* TIJ , p* TIJ x* TIJ , p* TIJ
Final year
15
Pricing method: secondary simulations
time t - 1 time t time t time t time t - 1 time t
x* t 11 x* t 11, p* t 11 x t 1, p t 1 x* t 11, p* t 11
x t - 1,1 x* t 12 x* t 12, p* t 12 x t 2, p t 2 x t - 1,1, p t - 1,1 x* t 12, p* t 12
x* t 1J x* t 1J, p* t 1J x* t 1J, p* t 1J
x* t 21 x* t 21, p* t 21 x* t 21, p* t 21
x t - 1,2 x* t 22 x* t 22, p* t 22 x t - 1,2, p t - 1,2 x* t 22, p* t 22
x* t 2J x* t 2J , p* t 2J x* t 2J , p* t 2J
x* tI 1 x* tI 1, p* tI 1 x* tI 1, p* tI 1
x t - 1,I x* tI 2 x* tI 2, p* tI 2 x t - 1,I , p t - 1,I x* tI 2, p* tI 2
x* tIJ x* tIJ , p* tIJ x tI , p tI x* tIJ , p* tIJ
year t
16
Pricing method: secondary simulations
time 0 time 1 time 1 time 1 time 0 time 1
x* 11 x* 11, p* 11 x 11, p 11 x* 11, p* 11
x 0 x* 12 x* 12, p* 12 x 12, p t 12 x 0, p 0 x* 12, p* 12
x* 1J x* 1J , p* 1J x 1I , p 1I x* 1J , p* 1J
year 1
17
Pricing method
2 2 1ˆˆ ˆ ˆ ˆt Ft FVt Vt FVt Σσ σ- -
1ˆ ˆt Vt tfΣ 1Vtz μ- -1
t tt
m zz 1
VttMt μm ˆˆ tVttMt mΣm ˆˆ 2 ˆ ˆHMt t FVt m σ
*2
ˆ ˆ ˆˆˆ
HMt t MtFt
Mt
*, 1
1 ˆˆ ˆL t Ft Ft Mt tt
P ff
- - -
18
Results and sensitivity
Sex Age Value per unit accrued pension Aggregate value deterministic
valuation stochastic
price deterministic
valuation stochastic price
1 member entire cohort %
incr %
incr R’million %
incr (1) (2) (3) (4) (5) (6) (7) (8) Female
25 14,11 16,00 13,4 15,84 -1,0 17 19 12,2 35 11,94 13,46 12,8 13,36 -0,8 172 193 11,8 45 12,00 13,49 12,4 13,39 -0,7 311 347 11,6 55 12,67 13,70 8,1 13,62 -0,6 347 372 7,5 65 13,36 13,78 3,1 13,78 0,0 434 447 3,1 75 9,53 9,69 1,7 9,69 0,0 236 240 1,7 85 5,96 6,01 0,9 6,01 0,0 62 63 0,9 total 1 579 1 681 6,5
19
Results and sensitivity
deterministic valuation
stochastic price
R’million % incr Female 1 579 1 681 6,5 Male 1 351 1 427 5,6 Total 2 930 3 109 6,1 Aggregate 2 930 3 094 5,6 Adjusted to fund data 2 912 3 074 5,6
20
Results and sensitivity
Deterministic valuation of model-point data 2 930difference due to risk-free stochastic pricing 11Risk-free stochastic price 2 941hedge-portfolio risks –19Stochastic price with hedge-portfolio risks 2 922residual risks –20Stochastic price: hedge-portfolio & residual risks 2 902cost of guarantee 192Stochastic price based on model-point data 3 094adjustment to fund data –20Stochastic price based on fund data 3 074
21
Results and sensitivity: major effects
Parameter Test result name description standard
value test
value price
(R’million) change in price
(%) standard values 3 094 N/A
1b general salary increase: sensitivity to inflation
–0,005 0 3 098 0,14
g return on market portfolio: sensitivity to risk-free rate
1,39 1,2 3 082 -0,37
M return on market portfolio: residual volatility
0,159 0,1 3 108 0,45
b force of inflation: residual volatility
–0,01379 0 3 061 –1,07
23
Conclusions• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.
24
Conclusions• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.• Stochastic price 5,6% higher than deterministic: because of pension
guarantee (may be overstated).
25
Conclusions• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.• Stochastic price 5,6% higher than deterministic.• Without guarantee, stochastic price only 1% less than deterministic:
If the valuation of the liabilities should allow for a risk premium only to the extent that the trustees are unable to avoid risk, then the valuation basis must be much closer to a risk-free basis than that produced by the risk premiums typically used.
26
Conclusions
• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.• Stochastic price 5,6% higher than deterministic: because of
pension guarantee.• Without guarantee, stochastic price only 1% less than
deterministic.• Effects of non-additivity: intra-cohort 0-1% plus inter-cohort
0,5%.
27
Conclusions
• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.• Stochastic price 5,6% higher than deterministic: because of
pension guarantee.• Without guarantee, stochastic price only 1% less than
deterministic.• Effects of non-additivity: intra-cohort 0-1% plus inter-cohort
0,5%.• Major sensitivities:
volatility of force of inflation in excess of conditional ex-ante expected inflation;sensitivity of ex-ante expected returns on the market portfolio to risk-free returns;residual volatility of the return on the market portfolio.
28
Conclusions
• Method computationally demanding, but not impossible.• Convergence complicated, but Sobol numbers expedite it.• Stochastic price 5,6% higher than deterministic: because of
pension guarantee.• Without guarantee, stochastic price only 1% less than
deterministic.• Effects of non-additivity: intra-cohort 0-1% plus inter-cohort
0,5%.• Major sensitivities.• Overall effect: Excluding uncertainties common to deterministic
and stochastic valuations, an error of about 5,6% is reduced to uncertainty of about 1%.