Beruflich Dokumente
Kultur Dokumente
Distribution function of X: FX (x) = Pr(X x) Survival function s(x): s(x) = 1 FX (x) Probability of death between age x and age y: Pr(x < X z) = FX (z) FX (x) = s(x) s(z) Probability of death between age x and age y given survival to age x: Pr(x < X z|X > x) = = Notations:
t qx
(y)dy
0 x+n
= exp
x
(y)dy
Derivatives: d t qx dt d t px dt d Tx dt d Lx dt d x e dt =
t px
(x + t) = fT (x) (t)
= t px (x + t) = lx = dx = (x)x 1 e
= Pr[T (x) t] = prob. (x) dies within t years = distribution function of T (x)
t px
t|u qx
t qx
t+u px
x = e
0
t px dt
u qx+t
ex =
k=1
k px
V ar[T (x)] = 2
0
t t px dt 2 ex (2k 1) k px e2 x
Curtate future lifetime (K(x) greatest integer in T (x)): Pr[K(x) = k] = Pr[k T (x) < k + 1] = = =
k px k+1 px k px qx+k k| qx
V ar[K(x)] =
k=1
Tx =
0
lx+t dt
=
0
lx+t dt =
0
lx t px dt
n Lx
n mx
Lx+k a(x) =
t t px (x + t) dt
0 1 t px 0
= (x + t) dt
=
0
lx+t dt
Lx lx+1 lx lx+1
Recursion formulas: E[K] = ex = px (1 + ex+1 ) E[T ] = x = px (1 + x+1 ) + qx a(x) e e ex = ex: n + n px ex+n x = x: n + n px x+n e e e E[K (m + n)] = ex: m+n = ex: m + E[T (m + n)] = x: m+n e = x: m + e ex+m: n m px
m px ex+m: n
Median future lifetime of (x): m(x) P r[T (x) > m(x)] = 1 s(x + m(x)) = s(x) 2
E[Z] = Ax =
0
v t t px x (t)dt
E[Z] = Ax =
k=0
v k+1 k px qx+k
k| qx
V ar[Z] =
Ax (Ax )2 =
v k+1
k=0
2
V ar[Z] =
Ax (Ax )2
E[Z] = Ax: n =
1
v t t px x (t)dt
0
V ar[Z] =
m| Ax
V ar[Z] =
Ax: n (A1 n )2 x:
1
E[Z] = m| Ax =
m
1 m| Ax = Ax Ax: m
E[Z] = Ax: n =
k=0
2
v k+1 k px qx+k + v n n px
E[Z] = Ax: n =
0
v t t px x (t)dt + v n n px
Ax = v 2 qx + v 2 px 2Ax
1 = vqx + vpx Ax+1: n1
A1 n x:
m|n Ax
m Ex
= A1 m+n x: = m| Ax
1 Ax+m: n A1
= vpx = v
(m1)|n Ax+1
x: m
Ax: 1 Ax: 2
m+n| Ax
= vqx + v 2 px
(I A)x =
0 n
t + 1 v t t px x (t)dt
(I A)1 n x:
=
0
t + 1 v t t px x (t)dt
a (Ia) n
(IA)x =
0 n
t v t px x (t)dt
(IA)x: n
=
0 n
t v t t px x (t)dt
(DA)1 n x:
=
0 n
(n t )v t t px x (t)dt
d = =
(DA)1 n x:
=
0
(n t)v t t px x (t)dt
(IA)x = Ax + vpx (IA)x+1 = vqx + vpx [(IA)x+1 + Ax+1 ] (DA)1 n x: (IA)1 n x: (I A)1
x: n
1 = nvqx + vpx (DA)x+1: n1
(IA)1 n x:
n qx n px
Ax: n
m|n Ax
1
i=0 m|n qx
(IA)x 1 + ex (IA)x x e
ax = E[ T ] = a
n| ax
ax = 1 + vpx ax+1
t Ex dt
=
0
2
v t px dt =
0
ax = 1 + v 2 px 2ax+1 = 1 + vpx ax+1: n1 = a(m) n Ex ax+n x = vpx + vpx ax+1: n1 = ax + vpx (I)x+1 a
(m)
V ar[ T ] = a
Ax (Ax )2 2
ax: n
=
0
2
v t t px dt = Ax: n (Ax: n )2 2 n| ax
V ar[Y ] =
ax = E[ K+1 ] = a
2
v k k px
V ar[ K+1 ] = a
t Ex dt
Ax (Ax d2
k=0 2 )
n| ax n| ax
=
n n
v t t px dt =
n
ax: n
= E[Y ] =
k=0
2
v k k px
V ar[Y ] = n-yr certain and life annuity: ax: n ax: n = ax + a n ax: n = an + n| ax = a n + n Ex ax+n
Ax: n (Ax: n )2 d2 n| ax
n| ax Most important identity 1 = x + Ax a 1 Ax ax = x: n = 1 x: n a A 2 Ax: n = 1 (2) 2ax: n 1 Ax d 1 Ax: n ax: n = d 1 = dx: n + Ax: n a ax =
= E[Y ] =
k=n
v k k px
= ax ax: n =
n Ex
ax+n
= an +
k=n
v k k px
= a n +n| ax
Accumulation function:
n
ax = E[ K ] = a
k=1
v k k px sx: n
ax =
1 (1 + i)Ax i
ax: n = = n Ex
0
1 nt Ex+t
i=0
h P (Ax )
= = = =
h P (Ax: n ) h Px h Px: n
1 Pure endowment annual premium Px: n : it is the reciprocal of the actuarial accumulated value sx: n because the share of the survivor who 1 has deposited Px: n at the beginning of each year for n years is the contractual $1 pure endowment, i.e. 1 Px: n sx: n = 1
Fully discrete equivalence premiums (whole life and endowment only): Ax = Px ax dAx 1 Ax 1 d ax P 2 2 1+ Ax (Ax )2 d 2 Ax (Ax )2 (dx )2 a 2 Ax (Ax )2 (1 Ax )2
(1)
P minus P over P problems: The dierence in magnitude of level benet premiums is solely attributable to the investment feature of the contract. Hence, comparisons of the policy values of survivors at age x + n may be done by analyzing future benets:
1 ( n Px Px: n )x: n s 1 Px: n 1 Px: n Px: n n Px = 1 Ax+n = 1 Px: n 1 Px: n Px: n = 1= 1 Px: n
= Ax+n =
n Px
(Px: n n Px )x: n s
1 (Px: n Px: n )x: n s
Miscellaneous identities: P (Ax: n ) Ax: n = P (Ax: n ) + Ax: n ax: n ax: n = = = Px: n Px: n + d 1 P (Ax: n ) + 1 Px: n + d
A# a#
(m)
Benet reserve t V : The expected value of the prospective loss at time t. Continuous reserve formulas: Prospective: Retrospective: Premium di.: Paid-up Ins.: Annuity res.: Death ben.: Premium res.: a t V (Ax ) = Ax+t P (Ax )x+t A1 s V (Ax ) = P (Ax )x: t x: n t t Ex (Ax ) = P (Ax+t ) P (Ax ) ax+t tV P (Ax ) Ax+t t V (Ax ) = 1 P (Ax+t ) ax+t t V (Ax ) = 1 ax Ax+t Ax t V (Ax ) = x 1A (Ax+t ) P (Ax ) P t V (Ax ) = P (Ax+t ) +
Variance of the loss function V ar[ t L] = V ar[ t L] = V ar[ t L] = V ar[ t L] = 2 P 2 Ax+t (Ax+t )2 2 Ax+t (Ax+t )2 assuming EP (1 Ax )2 2 P 2 1+ Ax+t: nt (Ax+t: nt )2 2 Ax+t: nt (Ax+t: nt )2 assuming EP (1 Ax: n )2 1+
Cost of insurance: funding of the accumulated costs of the death claims incurred between age x and x + t by the living at t, e.g.
4 kx
= =
1 kx
dx qx = lx+1 px
= = = =
n n Vx n n Vx
1 n Vx: n )
= Px ax+n
n Vx: n m m Vx: n
= 1 n Vx = Ax+m: nm Ax+m Relation between various terminal reserves (whole life/endowment only):
m+n+p Vx
h-payment reserves:
h tV h k Vx: n h kV
= 1 (1
m Vx )(1
=
j=0
bh+j+1 v
j+1
j px+h qx+h+j
j=0
h+j v j j px+h
tV
=
0
t+u v u u px+t du
Recursion relations:
hV
h+1 V
h+1 V h+1 V
Terminology: policy year h+1 the policy year from time t = h to time t = h + 1 h V + h initial benet reserve for policy year h + 1 h V terminal benet reserve for policy year h h+1 V terminal benet reserve for policy year h + 1 Net amount at Risk for policy year h + 1 Net Amount Risk bh+1
h+1 V
When the death benet is dened as a function of the reserve: For each premium P , the cost of providing the ensuing years death benet , based on the net amount at risk at age x + h, is : vqx+h (bh+1 h+1 V ). The leftover, P vqx+h (bh+1 h+1 V ) is the source of reserve creation. Accumulated to age x + n, we have:
n1 nV
=
h=0
[P vqx+h (bh+1
n1
h+1 V
= Psn
h=0
vqx+h (bh+1
h+1 V
If the death benet is equal to the benet reserve for the rst n policy years
nV
= Psn
If the death benet is equal to $1 plus the benet reserve for the rst n policy years
n1 nV
= Psn
h=0
vqx+h (1 + i)nh
If the death benet is equal to $1 plus the benet reserve for the rst n policy years and qx+h q constant
nV
= P s n vq n = (P vq) n s s
h+s V
)
h+1 V
1s
1s qx+h+s
1s px+h+s
UDD i.e.
h+s V h+s V
E[h ] = 0 V ar[h ] = v 2 (bh+1 The Hattendorf theorem V ar[ h L] = V ar[h ] + v 2 px+h V ar[ h+1 L] = v 2 (bh+1 V ar[ h L] = v 2 (bh+1
h+1 V h+1 V
)2 px+h qx+h
)2 px+h qx+h + v 2 px+h V ar[ h+1 L] )2 px+h qx+h )2 px+h px+h+1 qx+h+1 )2 px+h px+h+1 px+h+2 qx+h+2 +
+v 4 (bh+2 +v 6 (bh+3
h+2 V h+3 V
10
Last survivor status T (xy): T (xy) + T (xy) = T (x) + T (y) T (xy) T (xy) = T (x) T (y) fT (xy) + fT (xy) = fT (x) + fT (y) FT (xy) + FT (xy) = FT (x) + FT (y)
t pxy
n| qx
n| qy
n| qxy
= =
t px t qx
t py + t qy t qx t qy
t pxy dt
exy =
t pxy dt 1
k pxy
xy = e
0
Variances:
t t pu dt (u )2 e
t t pxy dt (xy )2 e
V ar[T (xy)] = 2
0
e t t pxy dt (xy )2
t py [(x + t) + (y + t)]
= n px n py
k px
k py [IL]
k+1 pxy
k qxy =
= n qx n qy
P r[K = k] = = =
qx+k:y+k qx+k:y+k =
k| qxy
= = =
n pxy n px n qx
n pxy + n py 2 n p x n p y
+ n qy 2 n qx n qy
[1] axy
= ax + ay 2xy a
11
Common shock model: sT (x) (t) = sT (x) (t) sz (t) = sT (x) (t) et sT (y) (t) = sT (y) (t) sz (t) = sT (y) (t) et sT (x)T (y) (t) = sT (x) (t) T (y) (t) sz (t) = sT (x) (t) sT (y) (t) et xy (t) = (x + t) + (y + t) + Insurance functions:
Insurances: Ax = 1 x a xy = 1 xy A a xy = 1 xy A a
Au =
k=0
= Axy =
k=0
Annuity functions:
au =
0
2
v t t pu dt Au (Au )2 2
Axy =
var[Y ] =
Au (Au )2
Reversionary annuities: A reversioanry annuity is payable during the existence of one status u only if another status v has failed. E.g. an annuity of 1 per year payable continuously to (y) after the death of (x). ax|y = ay axy
Covariance of T (xy) and T (xy): Cov [T (xy), T (xy)] = Cov [T (x), T (y)] + {E [T (x)] E [T (xy)]} {(E [T (y)] E [T (xy)]} = Cov [T (x), T (y)] + (x xy ) (y xy ) e e e e = (x xy ) (y xy ) [IL] e e e e
12
(j) t qx
( ) t qx
=
j=1
(j) t qx
Marginal PDF of T :
fT (t) = t p( ) ( ) (t) x x
m
(j) x
= = the force of decrement due only to decrement j Conditional PDF: fJ|T (j|t) =
j=1
( ) x
=
j=1 ( ) t px
(j) x
Survivorship group: ( ) Group of la people at some age a at time t = 0. Each member of the group has a joint pdf for time until decrement and cause of decrement.
(j) n dx ( = la ) ( = la ) xa m ( ) n dx ( ) xa pa xa+n (j) n qx
= e
x (s)ds
( )
( ) t pa
(j) (t)dt a
=
j=1 m
(j) n dx
Derivative:
( la ) =
(j) la j=1
d dt
( ) t px
d = dt
( ) t qx
t p( ) x
( ) x
(j) qx
dx lx
(j)
( )
Integral forms of t qx :
t (j) t qx
=
0 t
( ) s px
(j) (s)ds x
(j) (s)ds x
( ) t qx
=
0
( ) s px
( ) (s)ds x
= 1 t qx(j)
13
A=
j=1
(j)
Instead of summing the benets for each possible cause of death, it is often easier to write the benet as one benet given regardless of the cause of death and add/subtract other benets according to the cause of death. Premiums:
( Px ) = ( ) (j) Px = k=0
= =
(j) (t) x =
(j) qx qxt
( )
( )
( )
(j) (t) x
(j) t px
(j) qx
v k k px
k=0 (j) Bk+1 v k+1 k=0
( )
1 t qx
k px qx+k
( )
( )
(j)
( ) t px
v k k px
= t qx(j)
1 (1) qx = qx(1) 1 qx(2) 2 1 (2) qx = qx(2) 1 qx(1) 2 1 1 1 (1) qx = qx(1) 1 qx(2) qx(3) + qx(2) qx(3) 2 2 3
14
Expense policy fee: The portion of G that is independent of b. Asset shares notations: G level annual contract premium
k AS
asset share assigned to the policy at time t = k expenses paid per policy at time t = k probability of decrement by death probability of decrement by withdrawal cash amount due to the policy holder as a withdrawal benet
ck fraction of premium paid for expenses at k (i.e. cG is the expenxe premium) ek (d) qx+k (w) qx+k
k CV
bk death benet due at time t = k Recursion formula: [ k AS + G(1 ck ) ek ] (1 + i) = qx+k bk+1 + qx+k = Direct formula:
n1 n AS = h=0 k+1 AS (d) (w) k+1 CV
+ px+k +
( )
k+1 AS
k+1 AS)
k+1 AS)
(d)
(w)
15
lx = l0 ex = en = (px )n 1 x = e = E[T ] = E[X] x: n = x (1 n px ) e e 1 V ar[T ] = V ar[x] = 2 mx = ln2 Median[T ] = = Median[X] px ex = = E[K] qx px V ar[K] = (qx )2 Chapter 4 Ax =
2
(I)x = (x )2 = a a Chapter 6
For fully discrete whole life, w/ EP, V ar[Loss] = p 2Ax For fully continuous whole life, w/EP, V ar[Loss] = 2Ax Chapter 7
tV
Ax
A1 n x:
n Ex
+ = + 2 = Ax (1 n Ex ) = e
n(+)
(Ax ) = 0, t 0
k Vx
= 0, k = 0, 1, 2, . . .
For fully discrete whole life, assuming EP, V ar[ k Loss] = p 2Ax , k = 0, 1, 2, . . . For fully continuous whole life, assuming EP, V ar[ t Loss] = 2Ax , t 0 Chapter 9 For two constant forces, i.e. M acting on (x) and F acting on (y), we have: Axy = M + F M + F + 1 M + F + 1 M + F qxy qxy + i 1+i qxy + i pxy qxy
(IA)x = Ax =
2
Ax = =
Ax: n
( + )2 q q+i q q + 2i + i2 Ax (1 n Ex )
Chapter 5 ax =
2
ax = ax =
ax = =
ax: n ax: n
= ax (1 n Ex )
De Moivres Law
Chapter 3 s(x) = 1 lx qx
n|m qx n px t px (x
2
+ t) Lx x e ex
x x = l0 ( x) 1 = (x) = x m = x xn = x = qx = (x) = fT (x), 0 t < x 1 = (lx + lx+1 ) 2 x = = E[T ] = Median[T ] 2 x 1 = E[K] = 2 2 2 ( x) = 12 ( x)2 1 = 12 qx 2dx = = 1 lx + lx+1 1 2 qx 1 = E[S] = 2 n = n n px + n q x 2 n = ex: n + n qx 2 a x x an x
Ax = Ax =
A1 n x:
1Ax d
and the
xx = e xx = e xy = e
Ax = A1 n x: =
For two lives with dierent s, simply translate one of the age by the dierence in s. E.g. Age 30, = 100 Age 15, = 85
( x)2 c (c + 1)2 (c + 2)
lx = l0 (x) =
n px
x = e
c x xn c x x = E[T ] c+1 17
x 2c + 1 2c x
x with = e
Chapter 6 = t qx , 0 t 1 qx , 0t1 = 1 tqx = lx t dx , 0 t 1 sqx = , 0s+t1 1 tqx 1 = V ar[K] + 12 1 qx = (x + ) = 2 1 1 qx 2 1 = lx dx 2 1 = 2 1 = px + q x 2 = qx , 0 t 1 P (Ax ) = P (A1 n ) = x: P (Ax: n ) =
(m) Px =
(x + t) lx+t
s qx+t
V ar[T ] mx Lx a(x) x: 1 e
t px (x
Px: n
(m)
= =
(m) n Px hP (m)
(A1 n ) = x:
Chapter 7
h (m) k Vx: n h (m) (Ax: n ) kV
+ t)
= = =
h k Vx: n h kV
(m)
Chapter 4 Ax =
(m) Ax =
A1 n x: (I A)1 n x: Ax: n
n| Ax
2
= = = =
Ax =
h k V( Ax: n )
i h 1 1 V + h Vx: h k k x: h
Chapter 10 UDDMDT
(j) t qx (j) qx ( qx ) (j) t px (j) = t qx
(j) x ( ) x
qx qx
(j) ( )
1 t qx
( )
1 t qx
( )
UDDASDT
(j) t qx
ax: n
(m)
(m)x: n (m)(1 n Ex ) a id (m) = (m) (m) 1 i d i i(m) m1 and (m) = (m) (m) 2m i d (m) (m) = ax n Ex ax+n
= t qx(j)
1 (1) qx = qx(1) 1 qx(2) 2 1 (2) qx = qx(2) 1 qx(1) 2 1 1 1 (1) qx = qx(1) 1 qx(2) qx(3) + qx(2) qx(3) 2 2 3
18
(t)k k!
Sum of Poisson processes: If N1 , , Nk are independent Poisson processes with rates 1 , , k then, N = N1 + + Nk is a Poisson process with rate = 1 + + k . Special events in a Poisson process: Let N be a Poisson process with rate . Some events i are special with a probability P r[event is special] = i and Ni counts the spe cial events of kind i. Then, Ni is a Poisson process with rate i = i and the Ni are independent of one another. If the probability i (t) changes with time, then
t
E[Ni (t)] =
0
(s)ds
Non-homogeneous Poisson process: (t) intensity function m(t) mean value function
t
=
0
(y)dy [m(t)]k k!
1 ]
P r[Sn t] =
j=n
et
(t)j j!
(t)n1 (n i)!
X(t) =
i=1
Yi
Two competing Poisson processes: Probability that n events in the Poisson process (N1 ,1 ) occur before m events in the Poisson process (N2 ,2 ):
n+m1 1 2 P r[Sn < Sm ] = k=n 1 2 P r[Sn < S1 ] = 1 2 P r[S1 < S1 ] = n+m1
k 1 1 + 2 1 1 + 2
n
1 1 + 2
2 1 + 2
n+m1k
E[(X
d)k ] +
=
d
= e (d)[1 F (d)]
E[X u] =
u
F (x)dx +
0
S(x)dx
Left truncated and shifted variable (aka excess loss variable): Y P = X d|X > d Mean excess loss function: eX (d) e(d) = E[Y P ] = E[X d|X > d] S(x)dx = d 1 F (d) E[X] E[X d] = 1 F (d) Higher moments of the excess loss variable: ek (d) = E[(X d)k |X > d] x
=
0
E[(X u) ] =
Moment generating functions mX (t): mX (t) = E[etX ] Sum of random variables Sk = X1 + +Xk :
k
x>d
mXj (t)
FY (y) = a1 FX1 (y) + + ak FXk (y) fY (y) = a1 fX1 (y) + + ak fXk (y) E[Y ] = a1 E[X1 ] + + ak E[Xk ]
2 2 E[Y 2 ] = a1 E[X1 ] + + ak E[Xk ]
Tail weight: existence of moments light tail hazard rate increases light tail Multiplication by a constant , y > 0: y 1 y Y = X FY (y) = FX ( ) and fY (y) = fX ( ) Raising to a power: > 0 : FY (y) = FX (y ) 1 fY (y) = y 1 fX (y ) Y = X < 0 : FY (y) = 1 FX (y ) fY (y) = y 1 fX (y ) > 0: transformed distribution = 1: inverse distribution < 0: inverse transformed Exponentiation: Y = eX FY (y) = FX (ln(y)) 1 fY (y) = y fX (ln(y))
Probability generating function (pgf ): PN (z) = E[z N ] = p0 + p1 z + p2 z 2 + P (1) = E[N ] P (1) = E[N (N 1)] Poisson distribution: pk = e k k!
Mixing: The random variable X depends upon a parameter , itself a random variable . For a given value = , the individual pdf is fX| (x|). fX (x) = fX| (x|)f ()d
Negative Binomial distribution: Number of failures before success r with probability of success p = 1/(1 + ) pk = k+r1 k 1+
k
If X is a Poisson distribution with parameter , and follows a Gamma distribution with parameters (, ), then the resulting distribution is a Negative Binomial distribution with parameters (r = , = ) If X is an Exponential distribution with mean 1/ and is a Gamma distribution with parameters (, ), then the resulting distribution is a 2-parameter Pareto distribution with parameters ( = , = 1 )
Exam M - Loss Models - LGD c
1 1+
P (z) = [1 (z 1)]r E[N ] = r V ar[N ] = r(1 + ) If N1 and N2 are independent Negative Binomial distributions with parameters (r1 , ) and (r2 , ), then N1 + N2 is a Negative Binomial distribution with parameters (r = r1 + r2 , ) 3
If N is a Negative Binomial distribution with parameters (r, ) and some events j are special with probability j , then Nj , the count of special events j, is a Negative Binomial distribution with parame ters (j = r, j = j ) r Geometric distribution: Special case of the negative binomial distribution with r = 1 pk = k (1 + )k+1 1 1 (z 1)
Compound frequency and (a, b, 0) class: Let S be a compound distribution of primary and secondary distributions N and M with pn = P r[N = n] and fn = P r[M = n]. If N is a member of the (a, b, 0) class gk = 1 1 af0
k k
a+
j=1
jb k
fj gkj
Mixed Poisson models: If P (z) id the pgf of a Poisson distribution with rate drawn from a discrete random variable , then P (z|) = e(z1) P (z) = p (1 )e1 (z1) + + p (n )en (z1) Exposure modications on frequency: n policies in force. Nj claims produced by the j-th policy. N = N1 + + Nn . If the Nj are independent and identcally distributed, then the probability generating function for N is PN (z) = [PN1 (z)]n If the company exposure grows to n , let N represent the new total number of claims, PN (z) = [PN1 (z)]n = [PN (z)] Conditional expectations:
n n
Binomial distribution: Counting number of successes in m trials given a probability of sucess q m q k (1 q)mk k P (z) = [1 + q(z 1)]m pk = E[N ] = mq V ar[N ] = mq(1 q) The (a, b, 0) class: pk b =a+ pk1 k a > 0 negative binomial or geometric distribution a = 0 Poisson distribution a < 0 Binomial distribution Compound frequency models: P (z) = Pprim (Psec (z))
E[X| = ] =
xfX| (x|)dx
E[X] = E [E[X|]] V ar[X] = E [V ar[X|]] + V ar[E[X|]] The variance of the random variable X is the sum of two parts: the mean of the conditional variance plus the variance of the conditional mean.
Exposure modications on frequency for common distributions: N Parameters for N Parameters for N Poisson = n n Negative Binomial r, r = n r, = n
Exam M - Loss Models - LGD c
X : 2Pareto[, ] Y P : 2Pareto[ = , = + ] Per loss variable: Loss elimination ratio: fY L (0) = FX (d) fY L (y) = fx (y + d) FY L (y) = FX (y + d) Per payment variable: fY P (y) = FY L (y) = fx (y + d) 1 FX (d) FX (y + d) FX (d) 1 FX (d) LERX (d) = relations: c(X d) = (cX) (cd) c(X d)+ = (cX cd)+ A = (A b) + (A b)+ Ination on expected cost per loss: E[Y L ] = (1 + r) E(X) E X d 1+r E[X d] E[X]
Relations per loss / per payment variable: = Y L |Y L > 0 E[Y L ] E[Y P ] = P r[Y L > 0] E[(Y L )k ] E[(Y P )k ] = P r[Y L > 0] Simple (or ordinary) deductible d: X = X Y Y
L P L
YP
(1 + r) E(X) E X 1 FX
d 1+r
E[Y ] =
d 1+r
u-coverage limit: YL = X u YP
L
= X u|X > 0
u
E[Y ] = E[X u] =
0
E[Y ] = E[(X d)+ ] E[Y P ] = E[X d|X > d] E[X] E[X d] = 1 FX (d) Franchise deductible d: X = X Y Y
L P
[1 Fx (x)]dx
P P Yfranchise = Yordinary + d
E[Y ] = E[X] E[X d] + d[1 FX (d)] E[Y ] = E[X|X > d] E[X] E[X d] = +d 1 FX (d)
Exam M - Loss Models - LGD c
Probability of payment for a loss with deductible d: v = 1 FX (d) Unmodied frequency: The number of claims N (frequency) does not change, only the probabilities associated with the severity are modied to include the possibility of zero payment. Bernoulli random variable: N = 0 @ 1p 1 @ p
d u E X 1+r 1+r E X
d 1+r d 1+r
E X
E[Y ] = (1 + r)
u 1+r
1 FX
Coinsurance factor : First calculate Y L and Y P with deductible and limits. Then apply Y L = Y L Y P = Y P
Deductible, limit, ination and coinsurance: d 1+r u u = 1+r E[Y L ] = (1 + r) {E[X u ] E[X d ]} E[X u ] E[X d ] E[Y P ] = (1 + r) 1 FX (d ) d = E[Y L ] = 2 (1 + r)2 E (X u )2 E (X d )2 2d E[X u ] + 2d E[X d ] Modied frequency: N (the modied frequency) has a compound distribution: The primary distribution is N , and the secondary distribution is the Bernoulli random variable I. The probability generating function is PN (z) = PN [1 + v(z 1)] The frequency distribution is modied to include only non-zero claim amounts. Each claim amount probability is modied by dividing it by th eprobability of a non-zero claim. For common distributions, the frequency distribution changes as follows (the number of positive payments is nothing else than a special event with probability = v) N Parameters for N Parameters for N Poisson = v Binomial m, q m = m, q = vq Negative Binomial r, r = r, = v
2
S =
j=1
Xj
FX
(n)
(k) =
j=0
FX
(n1)
(k j)fX (j)
E[S] = E[N ] E[X] V ar[S] = E[N ] V ar[X] + V ar[N ] E[X]2 S (E[S], V ar[S]) if E[N ] is large
FX (k) =
(0)
0 , k<0 1 , k0
Probability generating function: PS (z) = PN [PX (z)] Notations: fk = P r[X = k] pk = P r[N = k] gk = P r[S = k] Compound probabilities: g0 = PN (f0 ) gk = if P r[X = 0] = f0 = 0
fS (k) =
n=0
pn fX (k)
(n)
FS (k) =
n=0
p n FX
(n)
(k)
E[S] =
k=0
k fS (k) =
k=0
[1 FS (k)]
E [(S d)+ ] =
(x d)fS (x)dx
d
E [(S d)+ ] =
d+1
(x d)fS (x)
d1
jb a+ k
fj gkj
(x d)fS (x)
E [(S d)+ ] =
d
[1 FS (d)]
d1
= P r[sum of n Xs = k]
k
fX (k) =
j=0 (1) (0)
fX
(n1)
(k j)fX (j)
= E[S]
x=0
[1 FS (d)]
Linear interpolation a < d < b: E [(S d)+ ] = da bd E [(S a)+ ]+ E [(S b)+ ] ba ba
Recursion relations for stop-loss insurance: E [(S d 1)+ ] = E [(S d)+ ] [1 FS (d)] E [(S sj )] = E [(S sj+1 )] + (sj+1 sj )P r[S sj+1 ] where the sj s are the possible values of S, S : s0 = 0 < s1 < s2 < and P r[S sj+1 ] = 1 P r[S = s0 or S = s1 or or S = sj ]
Exam M - Loss Models - LGD c
Theorem:
k Qn k Qn
Inequality:
(i) k Pn (i) = Pn Pn+1 Pn+k1 k Q(i,i) n (i) (i)
= cash ow at time l if subject in state i at time l = present value of 1 paid k years after time t = n made while in state i, given that the subject is in state s at t = n
k vn (i)
AP Vs@n (C (i) ) =
k=0
(s,i) k Qn
n+k C
(i)
k vn
= cash ow at time l + 1 if subject in state i at time l and state j at time l + 1 = probability of being in state i at time l and in state j at time l + 1 transition from state i to state j
AP Vs@n (C
AP Vs@n (C (i,j) ) =
k=0
(s,i) k Qn
Qn+k
(i,j)
n+k+1 C
(i,j)
k+1 vn