Professional Documents
Culture Documents
UTILITY THEORY
Maximum premium P policy holder for random loss X with initial wealth a E[U (a − X)] = U (a − P )
Minimum premium Q insurer for random loss Y with initial wealth a E[U (a + Q − Y )] = U (a)
Absolute dominance
RISK MEASURES
STOCHASTIC RETURNS
IID returns
ASSET VALUATIONS
Notations
Cij
Ei = E(Ri ) Vi = σi2 = Var(Ri ) Cij = Cov(Ri , Rj ) ρij = Corr(Ri , Rj ) =
σi σj
X
N X
N X
N X
Portfolio P of N Assets EP = x i Ei VP = x2i Vi + 2 xi xj Cij
i=1 i=1 i=1 j<i
STOCHASTIC CALCULUS
Disjoint increments are independent: i.e. Wt2 − Wt1 and Wt3 − Wt2 are independent
if t1 < t2 < t3
Z T X
n−1
Itô integral σ(Wt , t) dWt = lim σ(Wti , ti )(Wti+1 − Wti )
n→∞
0
"Z i=0 #
T
if σ(·, t) ∈ C 2 and E σ 2 (Wt , dt) dt < ∞
"Z # 0
T
- Itô isometry E σ(Wt , dt) dWt = 0
"0Z # Z
T T
Var σ(Wt , dt) dWt = E σ 2 (Wt , dt) dt
0 0
"Z # Z
T S
- Martingale E σ(Wt , t) dWt FS = σ(Wt , t) dWt if T > S
0 0
Z T Z T Z T
- Linearity (σ(Wt , t) + ν(Wt , t)) dWt = σ(Wt , t) dWt + ν(Wt , t) dWt
0 0 0
"Z # "Z # Z
T T T
- Deterministic integrand E σ(t) dWt = 0 Var σ(t) dWt = σ 2 (t) dt
0 0 0
Z T Z T
Itô Process XT = X0 + µ(Xt , t) dt + σ(Xt , t) dWt dXt = µ(Xt , t) dt + σ(Xt , t) dWt
0 0
Vasicek’s model drt = κ(θ − rt ) dt + σdŴt θ : mean level of short rate, κ: speed of reversion
∂g ∂g 1 ∂2g 2
g(t, rt ) = P (t, T ; rt ): g must satisfy g(T, r) = 1 and + κ(θ − rt ) + σ − rt g(t, rt ) = 0
" !# ∂t ∂r 2 ∂r2
Z T
P (t, T ) = E exp rs ds = exp[H(T − t) − G(T − t)rt ] where:
t
1 σ2 σ2 2
G(τ ) = 1 − e−κτ H(τ ) = θ − 2 [G(τ ) − τ ] − G (τ )
κ κ 4κ
√ ∂g ∂g 1 ∂2g 2
CIR model drt = κ(θ − rt ) dt + σ rt dŴt + κ(θ − rt ) + σ rt − rt g(t, rt ) = 0
∂t ∂r 2 ∂r2
√
P (t, T ) = exp [H(T − t) − G(T − t)rt ] where γ = κ2 + 2σ 2 and
!
2 (eγτ − 1) 2κθ 2γ exp 12 (γ + κ)τ
G(τ ) = H(τ ) = 2 ln
(γ + κ) (eγτ − 1) + 2γ σ (γ + κ) (eγτ − 1) + 2γ
Hull-White model drt = κ(t)(θ(t) − rt ) dt + σ dŴt Same as Vasicek except G(T − t), H(T − t) replaced
with:
Z T Z s Z T
1
G(t, T ) = exp − κ(u) du ds H(t, T ) = − κ(s)θ(s)G(s, T ) − σ 2 G2 (s, T ) ds
t t t 2
RISK MODELS
Credit risk Expected Credit Loss (ECL) = Exposure At Default (EAD) × Probability of Default (PD)
×Loss Given Default (LGD)
X
k−1
Q(t, T ) = ΣeD(T −t) Σ−1 qik (t, T ) = σij σ̂jk edj (T −t) − 1
j=1
" " Z ## !
X
k−1 T
−1
If Λ = Λ(t) = ΣDU (t)Σ : qik (t, T )| U (t) = σij σ̂jk E exp dj U (s) ds −1
j=1 t
LIABILITY VALUATIONS
General loss Cij = rj · si · xi+j + eij Cij : incremental claim in AYi -DYj
development model
rj : development factor (df) for DYj
1X
n
Development factors Arithmetic avg: fj = fij
n i=1
X
fij × Li,j−1
i Li,j
Weighted avg: fj = X fij = where Li,j is cumulative
Li,j−1 Li,j−1
i
Inflation adjusted CL Cij = rj · si · xi+j + eij To adjust for inflation, first convert cumulative figures
to incremental figures
Average cost per claim (1) Calculate past ACpC (2) Develop average ACpC (3) Develop claim count
(ACpC)
Estimated ultimate claimsi = Ultimate ACpCi × Ultimate claim counti
RUIN THEORY
X
N (t)
Aggregate claim S(t) = Xi N (t): number of claims, Xi : amount for i-th claim, S(t) : aggregate claim
i=1
Ruin probability Ψ(U ) = P [U (t) < 0 for some t : 0 < t < ∞], Ψ(U, t) = P [U (τ ) < 0 for some t : 0 < τ ≤ t]
Ψ(U ), Ψ(U, τ ) are decreasing in U , increasing in τ , Ψ(U, τ ) < Ψ(U ) lim Ψ(U, t) = Ψ(U )
t→∞
k
(λt)
Poisson process N (t): Poisson process with rate λ P [N (t) = k] = exp[−λt]
k!
Compound Poisson E[S(t)] = λtE[Xi ] Var[S(t)] = λtE[Xi2 ] MS (r) = exp[λt(MX (r) − 1)]
model
To make E[U (t)] > U : c = (1 + θ)λE[Xi ] θ: premium loading factor (θ > 0)
λ
R is the unique positive root of: λ(MX (r) − 1) − cr = 0 If Xi ∼ Exp(α): R = α −
c
2(c − λE[Xi ]) 1 c
Bounds for adjustment coefficient: R < If Xi ≤ M : R > ln
λE[Xi2 ] M λE[Xi ]
θ 1 erb − era
Moment generating Exp(θ) : MX (r) = Γ(k, θ) : MX (r) = Unif[a, b] : M X (r) =
θ−r (1 − rθ)k r(b − a)
functions
General aggregate c > E[Si ] Si : aggregate claim in year i γ > 0: lim− E[exp[r(Si − c)]] = ∞
r→γ
model
R is unique positive value satisfying: E[exp[R(Si − c)]] = 1
Ruin probability Ψ(U, τ ) decreases For compound Poisson Ψ(U, τ ) increases for larger λ
dependence for larger θ models:
1 θU
If Xi ∼ Exp(1) Ψ(U ) = exp −
1+θ 1+θ
Proportional c = [(1 + θ) − (1 + ξ)(1 − α)]λE[Xi ]
reinsurance
ξ: reinsurer premium loading factor α: retention level
M = retention level
OPTION THEORY
American put Pt ≤ K Pt ≥ K − S t
Forward price zero dividend: S0 erT fixed dividend: (S0 − I)erT continuous dividend:
S0 exp[(r − δ)T ]
BLACK-SCHOLES MODEL
Equivalent measure P ∼ Q are equivalent ⇐⇒ P (E) > 0 whenever Q(E) > 0 for any event E.
Z t
Cameron-Martin-Girsanov There exists an equivalent measure Q s.t. Wt + γs ds is a Wiener process,
0
where γs is a previsible process.
µ−r
Martingale discounted price To make e−rt St martingle, apply CMG with γt = ,
σ
σ2
where St = S0 exp µ − t + σWt
2
Martingale representation Xt , Yt : martingale. Then, dYt = ϕt dXt for some ϕt
⇐⇒ Xt is not martingale under any other equivalent measure.
Black-Scholes formula (δ = 0) ct = St N (d1 ) − Ke−r(T −t) N (d2 ) pt = Ke−r(T −t) N (−d2 ) − St N (−d1 )
∂f ∂f σ 2 St2 ∂ 2 f
Martingale =⇒ −rf (t, St ) + + rSt + =0
∂t ∂s 2 ∂s2
BSE derivation (δ > 0) Set : value of investment starting with S0 dSet = (µ + δ)Set dt + σ dZt
σ2
Set = Se0 exp µ + δ − t + σZt Change in derivation:
2
∂f
dVt = −df + (dSt + δSt dt)
∂s
∂f ∂f σ 2 St2 ∂ 2 f
+ (r − δ)St + = rf (t, St )
∂t ∂s 2 ∂s2