Beruflich Dokumente
Kultur Dokumente
Summary
1 INTRODUCTION
3 ESTIMATION
4 PREDICTION
5 POSSIBLE EXTENSIONS
6 APPLICATION
1. INTRODUCTION
• To provide :
- precision of the estimation of the parameters
- not only point predictions but also distributions of predictions
• To obtain :
- explicit or quasi explicit prediction formulas
- a framework easily extended to a joint modelling of an
historical and a risk-neutral dynamics
• key tools :
- affine stochastic processes, in particular the pair (Zt , Zt2 )
when Zt is a gaussian AR(1)
- recursive formulas for multihorizon Laplace transforms of
affine processes
- cross sectional asymptotic estimators of mortality intensities
- extended Kalman filter
= exp[−λ(x; t + 1)]
key idea
S(x; t + h) = P(Y > t + h|Y > t, age at t = x, zt )
= Et P(Y > t + h|Y > t, age at t = x, zt+h ),
= Et Πhu=1 P(Y > t + u|Y > t + u − 1, age at t = x, zt+h )
X
h
= Et exp[− λ(x + u − 1; t + u)]
u=1
2 ) : multihorizon
Since λ(x + u − 1; t + u) linear in (Zt+u , Zt+u
2
Laplace transform of (Zt , Zt )
=⇒ Simple recursive formula
3. ESTIMATION
Assumption :
The lifetime variables Yi,to ,(individual i of cohort to ), are
independent conditionally to the factor Zt and are driven by the
mortality intensity specified above.
The survivor probability exp[−λ(x, Zt+1 , t + 1)] = π(Zt+1 , t + 1)
is estimated by the observed survivor rates π̂(x, t + 1) and,
conditional on Zt , the π̂(x, t + 1) are asymptotically
independent, gaussian
√ D
nx,t [π̂(x, t+1)−π(x, Zt+1 , t+1)] −→ N[{0, π(x, Zt+1 , t+1)[1−π(x, Zt+1 , t+1)]}
Delta method
√ D
nxt [λ̂(x, t + 1) − λ(x, Zt+1 , t + 1)] −→ N{0, exp[λ(x, Zt+1 , t + 1)] − 1}
4. PREDICTION
s s
Zt+n = ρZt+n−1 + σεst+n , n = 1, . . .
with initial condition Zts = Zt
iii) Deduce the simulated spot intensities
λs (x + n − 1, Zt+n
s , t + n)
S(x; t, t + h)
exp[−λP (x, t, t + h)] =
S(x; t, t + h − 1)
Et {exp(− ∧ (x, t, t + h − 1)) exp[−λ(x + h − 1; t + h)]
=
S(x, t, t + h − 1)
5. POSSIBLE EXTENSIONS
• Several factors
• Pricing methodology :
- (Zt , Zt2 ) extended to a linear process Ft = (Xt0 , Zt , Zt2 )
- SDF Mt,t+1 exponential affine in Ft , Ft+1
- RN dynamics also affine
- pricing at t of payoffs at T > t of the following types :
1lYi >T ) (recursive formulas)
or exp(w10 FT )1l(w20 FT <k ) 1l(Yi >T ) (unidimensional integrals)
6. APPLICATION
Data
Data drawn from the Berkeley human mortality database ,
For France,
disaggregated by gender
year 1899-2004
age 0-110 year
λ(x, t) = α1 exp(a1 x+b1 t)+α2 Zt exp(a2 x+b2 t)+α3 Zt2 exp(a3 x+b3 t)
Indeed,
there is a change of regime after the second world war, which is
easily detected by considering the average mortality intensities
per year
x̄
1 X
λ̄(t) = λ(x; t)
x̄ + 1
x=0
α + β exp(ax) is sufficient
F̂t = γ0 + γ1 exp(γ2 t) + Zt
Provides a residual plot, which is useful to analyse the pure
stochastic factor dynamic.
Final model :
S(x, t, t + h)
exp[−λP (x, t, t + h)] =
S(x; t, t + h − 1)
is the prospective (forward) survivor rate at t, between t + h − 1
and t + h
λP (x, t, t + h) : prospective (forward) mortality intensity at t,
between t + h − 1 and t + h.