- Changing probibility measure
- Brownian Motion and its Property
- Stochastic Integrals and It o ^ \hat{o} o^ Formula
- Stochastic Differential Equations
- S5 Equivalent Martingale Measure (changing measure)
- S6 Volatility Smile
- Greeks and Dynamic Hedging
Changing probibility measure
Condition
both measure agree on what’s possible and impossible.
P
(
w
)
>
0
⇔
P
∗
(
w
)
>
0.
\mathbb{P}(w)>0 \Leftrightarrow \mathbb{P}^*(w)>0.
P(w)>0⇔P∗(w)>0.
How to do: Radon-Nikodym Derivative
what:
R
N
=
d
P
d
Q
RN=\frac{d\mathbb{P}}{d\mathbb{Q}}
RN=dQdP
how to work:
E
Q
[
x
]
=
E
P
[
x
d
Q
d
P
]
\mathbb{E}^{\mathbb{Q}}[x]=\mathbb{E}^{\mathbb{P}}[x\frac{d\mathbb{Q}}{d\mathbb{P}}]
EQ[x]=EP[xdPdQ]
meaning:
- able to get Q \mathbb{Q} Q measure valuation while using P \mathbb{P} P measure parameters.
- feels like minimum adjusting to existing results( P \mathbb{P} P results).
Relevant Info: Conditional Expectation (not prob)
Terminology:
- Conditional on no previous information:
E 0 ∗ [ X ] = E ∗ [ X ] \mathbb{E}^*_0[X]=\mathbb{E}^*[X] E0∗[X]=E∗[X] - Conditional on N time:
E N ∗ [ X ] = X \mathbb{E}^*_N[X]=X EN∗[X]=X
Property:
- Linearity
for all constant a a a, b b b
E n [ a X + b Y ] = a E n [ X ] + b E n [ b ] . \mathbb{E}_n[aX+bY]=a\mathbb{E}_n[X]+b\mathbb{E}_n[b]. En[aX+bY]=aEn[X]+bEn[b]. - Extracting known variable
if X X X only depends on the first n n n tosses, then
E n [ X Y ] = X E n [ Y ] . \mathbb{E}_n[XY]=X\mathbb{E}_n[Y]. En[XY]=XEn[Y]. - Iterated conditioning
if 0 ≤ n ≤ m ≤ N 0 \leq n \leq m \leq N 0≤n≤m≤N, then n n n cancel m m m away:
E n [ E m [ X ] ] = E n [ X ] \mathbb{E}_n[\mathbb{E}_m[X] ] =\mathbb{E}_n[X] En[Em[X]]=En[X] - Independence
if X X X only depends on the ( n + 1 ) t h (n+1)^{th} (n+1)th toss, then
E n [ X ] = E [ X ] \mathbb{E}_n[X]=\mathbb{E}[X] En[X]=E[X]
Relevant Info: Martingale
what:
M
n
=
E
n
∗
[
M
n
+
1
]
M_n=\mathbb{E}_n^*[M_{n+1}]
Mn=En∗[Mn+1]
E
s
[
W
n
]
=
W
s
E_s[W_n]=W_s
Es[Wn]=Ws
what’s important:
- no drift, expectation always goes back to M 0 M_0 M0
- risk-neutral process follows martingale
- thus we have no drift makes risk-neutral measure
Derive continuous binomial tree formula from discrete b-tree
normal (r=0)
σ
T
2
=
σ
S
0
2
∗
T
(T time)
σ
1
s
t
e
p
2
=
σ
S
0
2
∗
T
n
(total n steps)
σ
1
s
t
e
p
=
σ
2
T
n
\begin{aligned} \sigma_T^2&=\sigma_{S_0}^2*T\quad \textnormal{(T time)}\\ \sigma_{1\ step}^2&=\frac{\sigma_{S_0}^2*T}{n}\quad \textnormal{(total n steps)}\\ \sigma_{1\ step}&=\sqrt{\frac{\sigma^2 T}{n}}\\ \end{aligned}
σT2σ1 step2σ1 step=σS02∗T(T time)=nσS02∗T(total n steps)=nσ2T
X
X
X is 1 or -1(moving up or down), and
E
(
X
)
=
0
E(X)=0
E(X)=0,
V
(
X
)
=
1
V(X)=1
V(X)=1.
X
i
∼
N
(
0
,
1
n
)
X_i\sim N(0,\frac{1}{n})
Xi∼N(0,n1)
S
n
=
S
0
+
∑
1
+
∞
X
i
σ
n
=
S
0
+
T
n
σ
∑
X
i
apply CLT
l
e
t
Z
=
∑
X
i
n
∼
N
(
0
,
1
)
=
S
0
+
T
σ
Z
\begin{aligned} S_n&=S_0+\sum^{+\infty}_{1} X_i\sigma _n\\ &=S_0+\sqrt{\frac{T}{n}}\sigma \sum X_i \quad \textnormal{apply CLT}\\ &let Z = \frac{\sum X_i}{\sqrt{n}} \sim N(0,1)\\ &=S_0+\sqrt{T}\sigma Z \end{aligned}
Sn=S0+1∑+∞Xiσn=S0+nTσ∑Xiapply CLTletZ=n∑Xi∼N(0,1)=S0+TσZ
lognormal (r involved)
l o g ( S t ) = l o g ( S t − 1 ) + ( r − 1 2 σ 2 ) Δ t + σ Δ t X t sum all equations l o g ( S t ) = l o g ( S 0 ) + ( r − 1 2 σ 2 ) T + σ T n ∑ X i apply CLT likewise l o g ( S t S 0 = e x p [ ( r − σ 2 2 ) T + σ T Z ] \begin{aligned} log(S_t)&=log(S_{t-1})+(r-\frac{1}{2}\sigma^2)\Delta t+\sigma \sqrt{\Delta t} X_t\\ &\textnormal{sum all equations}\\ log(S_t)&=log(S_0)+(r-\frac{1}{2}\sigma^2)T+\sigma\sqrt{\frac{T}{n}}\sum X_i\\ &\textnormal{apply CLT likewise}\\ log(\frac{S_t}{S_0}&=exp[(r-\frac{\sigma^2}{2})T+\sigma\sqrt{T}Z]\\ \end{aligned} log(St)log(St)log(S0St=log(St−1)+(r−21σ2)Δt+σΔtXtsum all equations=log(S0)+(r−21σ2)T+σnT∑Xiapply CLT likewise=exp[(r−2σ2)T+σTZ]
Brownian Motion and its Property
Random Walk
random walk process:
S
n
=
∑
X
i
,
X
i
=
±
1
S_n=\sum X_i, \quad X_i=\pm 1
Sn=∑Xi,Xi=±1
Reflection Principle:
up and down have same prob.
some example here:
-
W
t
W_t
Wt is a Brownian process. Conditional on
W
1
>
0
W_1>0
W1>0, what is the probability that
W
2
<
0
W_2<0
W2<0?
P ( W 2 < 0 ∣ W 1 > 0 ) = P ( W 2 ↓ ) P ( ∣ W 2 − W 1 ∣ > ∣ W 1 − 0 ∣ ) = 0.5 × 0.5 = 0.25 \begin{aligned} \mathbb{P}(W_2<0|W_1>0)&=\mathbb{P}(W_2 \downarrow)\mathbb{P}(|W_2-W_1|>|W_1-0|)\\ &=0.5\times 0.5=0.25 \end{aligned} P(W2<0∣W1>0)=P(W2↓)P(∣W2−W1∣>∣W1−0∣)=0.5×0.5=0.25
Brownian Motion
Derive from Random Walk:
- W n ( 0 ) = 0 W_n(0)=0 Wn(0)=0, start from 0.
- time spacing is 1 n \frac{1}{n} n1.
- up and dowm jumps equal and of size 1 n \frac{1}{\sqrt{n}} n1.
- measure
P
\mathbb{P}
P, given by up and down probabilities everywhere equal to
1
2
\frac{1}{2}
21.
W n ( i n ) = W n ( i − 1 n ) + X i n , 1 ≤ i W n ( t ) = t ( ∑ i = 1 n t X i n t ) l e t t = i n \begin{aligned} W_n(\frac{i}{n})&=W_n(\frac{i-1}{n})+\frac{X_i}{\sqrt{n}},\quad 1\leq i\\ W_n(t)&=\sqrt{t}(\frac{\sum_{i=1}^{nt} X_i}{\sqrt{nt}})\quad let t=\frac{i}{n} \end{aligned} Wn(ni)Wn(t)=Wn(ni−1)+nXi,1≤i=t(nt∑i=1ntXi)lett=ni
put n → ∞ n \to \infty n→∞, we distribe :
W t = lim n → ∞ W n ( t ) W_t=\lim_{n\to \infty}W_n(t) Wt=n→∞limWn(t)
Defination (for checking if Brownian):
- start from 0
- increments follow N(0,time)
- increments iid
Feature:
E [ W t ] = 0 E [ W t 2 ] = t E [ W t − W s ] = 0 V [ W t − W s ] = E [ ( W t − W s ) 2 ] = t − s C o v ( W s , W t ) = s , s < t C o r r ( W s , W t ) = s t \begin{aligned} \mathbb{E}[W_t]&=0\\ \mathbb{E}[W_t^2]&=t\\ \mathbb{E}[W_t-W_s]&=0\\ \mathbb{V}[W_t-W_s]&=\mathbb{E}[(W_t-W_s)^2]=t-s\\ Cov(W_s,W_t)&=s,\quad s<t\\ Corr(W_s,W_t)&=\sqrt{\frac{s}{t}}\\ \end{aligned} E[Wt]E[Wt2]E[Wt−Ws]V[Wt−Ws]Cov(Ws,Wt)Corr(Ws,Wt)=0=t=0=E[(Wt−Ws)2]=t−s=s,s<t=ts
Property:
- W t W_t Wt is continuous everywhere, but differentiable nowhere
- Brownian motion will hit any real value eventually.
- Once hit a value, hit infinitely often.
- fractal.
Brownian Martingales:
- W t W_t Wt
- W t 2 − t W_t^2-t Wt2−t
- W t 3 W_t^3 Wt3 is not, prove E S [ W t 3 ] ≠ W s 3 E_S[W_t^3]\neq W_s^3 ES[Wt3]=Ws3
- e σ W t e^{\sigma W_t} eσWt is not
- exp ( σ W t − σ 2 t 2 ) \exp(\sigma W_t - \frac{\sigma^2t}{2}) exp(σWt−2σ2t)
- technique:
E s [ W t 3 ] = E s [ ( W t − W s + W s ) 3 ] E_s[W_t^3]=E_s[(W_t-W_s+W_s)^3] Es[Wt3]=Es[(Wt−Ws+Ws)3] both add and substract W s W_s Ws
Stochastic Integrals and It o ^ \hat{o} o^ Formula
Models
normal items and a
W
t
W_t
Wt which stands for a stochastic process.
S
t
=
S
0
+
μ
t
+
σ
W
t
S_t = S_0+\mu t+\sigma W_t
St=S0+μt+σWt
intergral form
S
t
=
S
0
+
∫
0
t
μ
d
u
+
∫
0
t
σ
d
W
u
S_t = S_0+\int _0^t\mu du+\int_0^t \sigma dW_u
St=S0+∫0tμdu+∫0tσdWu
differential form
d
S
t
=
μ
(
t
,
S
t
)
d
t
+
σ
(
t
,
S
t
)
d
W
t
dS_t = \mu(t,S_t) dt+\sigma(t,S_t) dW_t
dSt=μ(t,St)dt+σ(t,St)dWt
Stochastic Integral
core: Taylor expansion.
If
S
t
S_t
St includes stochastic process
W
t
W_t
Wt, then it needs to expand to second moment since
V
a
r
(
W
t
)
=
t
Var(W_t)=t
Var(Wt)=t,
d
W
t
2
=
d
t
dW_t^2 = dt
dWt2=dt.
Property of I = ∫ 0 T f ( u ) d W t I=\int^T_0f(u)dW_t I=∫0Tf(u)dWt:
- E [ I ] = 0 \mathbb{E}[I]=0 E[I]=0
- E [ I 2 ] = E [ ( ∫ 0 T f ( u ) d W u ) 2 ] = E [ ∫ 0 T f ( u ) 2 d u ] \mathbb{E}[I^2]=\mathbb{E}[(\int_0^Tf(u)dW_u)^2]=\mathbb{E}[\int_0^Tf(u)^2du] E[I2]=E[(∫0Tf(u)dWu)2]=E[∫0Tf(u)2du], Ito’s Isometry
- E [ ∫ 0 T f ( u ) d W u × ∫ 0 T g ( u ) d W u ] = E [ ∫ 0 T f ( u ) g ( u ) d W u ] \mathbb{E}[\int_0^Tf(u)dW_u\times \int_0^Tg(u)dW_u]=\mathbb{E}[\int_0^Tf(u)g(u)dW_u] E[∫0Tf(u)dWu×∫0Tg(u)dWu]=E[∫0Tf(u)g(u)dWu]
- If
f
f
f is a deterministic function, then
I ∼ N ( 0 , ∫ 0 T f ( u ) 2 d u ) I\sim N(0,\int^T_0f(u)^2du) I∼N(0,∫0Tf(u)2du)
It o ^ \hat{o} o^ Formula
d X t = μ d t + σ d W t dX_t = \mu dt + \sigma dW_t dXt=μdt+σdWt
Y t = f ( X t ) Y_t=f(X_t) Yt=f(Xt) function of a stochastic process
derive X twice
d
Y
t
=
(
μ
t
f
′
+
1
2
σ
t
2
f
′
′
)
d
t
+
σ
t
f
′
d
W
t
dY_t = (\mu_tf'+\frac{1}{2}\sigma_t^2f'')dt+\sigma_tf'dW_t
dYt=(μtf′+21σt2f′′)dt+σtf′dWt
Y t = g ( X t , t ) Y_t=g(X_t,t) Yt=g(Xt,t)function of a stochastic process and time
d Y t = g t d t + ( μ t g x + 1 2 σ t 2 g x x ) d t + σ t g x d W t dY_t = g_tdt+(\mu_tg_x+\frac{1}{2}\sigma_t^2g_{xx})dt+\sigma_tg_xdW_t dYt=gtdt+(μtgx+21σt2gxx)dt+σtgxdWt
Applications
- Solve intergrals
Cal result for
∫ 0 T W t d W t \int_0^TW_tdW_t ∫0TWtdWt
A:
let X t = W t 2 X_t=W_t^2 Xt=Wt2, f x = 2 W t , f x x = 2 f_x=2W_t,f_{xx}=2 fx=2Wt,fxx=2
d X t = f x d W t + 1 2 f x x d t = 2 W t d W t + d t X T − X 0 = ∫ 0 T 2 W t d W t + T ∫ 0 T W t d W t = W T 2 2 − T 2 \begin{aligned} dX_t&=f_xdW_t+\frac{1}{2}f_{xx}dt\\ &=2W_tdW_t+dt\\ X_T-X_0&=\int_0^T 2W_tdW_t+T\\ \int_0^T W_tdW_t&=\frac{W_T^2}{2}-\frac{T}{2} \end{aligned} dXtXT−X0∫0TWtdWt=fxdWt+21fxxdt=2WtdWt+dt=∫0T2WtdWt+T=2WT2−2T
3 ways to cook e θ W t e^{\theta W_t} eθWt
Solve 1:
W
t
=
x
,
x
∼
N
(
0
,
t
)
W_t=x, x\sim N(0,\sqrt{t})
Wt=x,x∼N(0,t)
E
[
e
θ
W
t
]
=
1
2
π
t
∫
e
θ
x
e
−
x
2
2
t
d
x
\mathbb{E}[e^{\theta W_t}]=\frac{1}{\sqrt{2\pi t}}\int e^{\theta x}e^{-\frac{x^2}{2t}}dx
E[eθWt]=2πt1∫eθxe−2tx2dx
Solve 2:
W
t
∼
t
x
,
x
∼
N
(
0
,
1
)
W_t \sim \sqrt{t}x, x\sim N(0,1)
Wt∼tx,x∼N(0,1)
E
[
e
θ
t
X
]
=
1
2
π
∫
e
θ
t
x
e
−
x
2
2
d
x
\mathbb{E}[e^{\theta \sqrt{t}X}]=\frac{1}{\sqrt{2\pi}}\int e^{\theta \sqrt{t}x}e^{-\frac{x^2}{2}}dx
E[eθtX]=2π1∫eθtxe−2x2dx
Solve 3: Leibnits
Let
X
t
=
e
θ
W
t
X_t=e^{\theta W_t}
Xt=eθWt, by Ito’s formula we have
d
X
t
=
θ
X
t
d
W
t
+
1
2
θ
2
X
t
d
t
X
t
=
X
0
+
θ
∫
0
t
X
u
d
W
u
+
θ
2
2
∫
0
t
X
u
d
u
E
[
X
t
]
=
E
[
X
0
]
+
1
2
θ
2
∫
0
t
E
[
X
t
]
d
t
d
E
[
X
t
]
d
t
=
θ
2
2
E
[
X
t
]
E
[
X
t
]
=
E
[
e
θ
W
t
]
=
e
θ
2
2
t
\begin{aligned} dX_t &= \theta X_t dW_t+\frac{1}{2}\theta ^2 X_tdt\\ X_t&=X_0+\theta \int_0^t X_udW_u+\frac{\theta^2}{2}\int_0^tX_udu\\ \mathbb{E}[X_t]&=\mathbb{E}[X_0]+\frac{1}{2}\theta ^2 \int_0^t \mathbb{E}[X_t]dt\\ \frac{d\mathbb{E}[X_t]}{dt}&=\frac{\theta^2}{2}\mathbb{E}[X_t]\\ \mathbb{E}[X_t]&=\mathbb{E}[e^{\theta W_t}]=e^{\frac{\theta ^2}{2}t} \end{aligned}
dXtXtE[Xt]dtdE[Xt]E[Xt]=θXtdWt+21θ2Xtdt=X0+θ∫0tXudWu+2θ2∫0tXudu=E[X0]+21θ2∫0tE[Xt]dt=2θ2E[Xt]=E[eθWt]=e2θ2t
Conclusion: find Tylor expansion for general format(since can not find solution for exponential
W
t
W_t
Wt), take expectation then and use ODE to solve for results.
Rules/ Techniques
Tools:
- use log to separate X t X_t Xt from d x dx dx terms.
- Chain rule.
- Product rule(keep d W t × d W t dW_t\times dW_t dWt×dWt terms).
- Two different Brownian Motions
- ρ = 1 : d W t × d W t ′ = d t \rho = 1: dW_t\times dW'_t=dt ρ=1:dWt×dWt′=dt
- ρ = 0 : d W t × d W t ′ = 0 \rho = 0: dW_t\times dW'_t=0 ρ=0:dWt×dWt′=0
- Box calculus rules:
- only keep d W t dW_t dWt relevant terms.
Stochastic Differential Equations
Bachelier Model
d S t = σ S 0 d W t dS_t = \sigma S_0dW_t dSt=σS0dWt
Black-Scholes Model
d S t = r S t d t + σ S t d W t dS_t=rS_tdt+\sigma S_tdW_t dSt=rStdt+σStdWt
C ( S t , t ) = N ( d 1 ) S t − N ( d 2 ) P V ( K ) d 1 = 1 σ t [ ln ( S t K ) + ( r + σ 2 2 ) t ] d 2 = d 1 − σ t \begin{aligned} C(S_{t},t)&=N(d_{1})S_{t}-N(d_{2})PV(K)\\ d_{1}&={\frac {1}{\sigma {\sqrt {t}}}}\left[\ln \left({\frac {S_{t}}{K}}\right)+\left(r+{\frac {\sigma ^{2}}{2}}\right)t\right]\\ d_{2}&=d_{1}-\sigma {\sqrt {t}}\\ \end{aligned} C(St,t)d1d2=N(d1)St−N(d2)PV(K)=σt1[ln(KSt)+(r+2σ2)t]=d1−σt
deriving steps:
- S t S_t St formula
- find payment thershold x ∗ x^* x∗
- V t V_t Vt formula through intergration
- Discount to V 0 V_0 V0
Relevant Questions
- What’s the
K
K
K if call and put have equal price?
According to put call parity: 0 = c − p = S 0 − K e − r T 0=c-p=S_0-Ke^{-rT} 0=c−p=S0−Ke−rT
K = S 0 e r T K=S_0e^{rT} K=S0erT - Which is more expensive according to BSM?
K
K
K is equal price from last Q.
- call @ $K+$10$
- put @ $K-$10$
c>p
Since stock price is lognormal distribution, thus, moving down would result in a higher area of probability and moving up would cover a smaller areas of probability. - call gives unlimited potential upside, put gives limited potentials downside, why same worth at K K K?
BSM normal return vs log return
Normal | Lognormal | |
---|---|---|
Mean | log ( S 0 ) + ( r − σ 2 2 ) T \log (S_0)+(r-\frac{\sigma^2}{2})T log(S0)+(r−2σ2)T | S 0 e r T S_0e^{rT} S0erT |
Median | log ( S 0 ) + ( r − σ 2 2 ) T \log (S_0)+(r-\frac{\sigma^2}{2})T log(S0)+(r−2σ2)T | S 0 e ( r − σ 2 2 ) T S_0e^{(r-\frac{\sigma^2}{2})T} S0e(r−2σ2)T |
Mode | log ( S 0 ) + ( r − σ 2 2 ) T \log (S_0)+(r-\frac{\sigma^2}{2})T log(S0)+(r−2σ2)T | S 0 e ( r − 3 σ 2 2 ) T S_0e^{(r-\frac{3\sigma^2}{2})T} S0e(r−23σ2)T |
Variance | σ 2 T \sigma^2T σ2T | S 0 2 e 2 r T ( e σ 2 T − 1 ) S_0^2e^{2rT}(e^{\sigma^2T}-1) S02e2rT(eσ2T−1) |
Range | − ∞ < log ( S T ) < + ∞ -\infty<\log (S_T)<+\infty −∞<log(ST)<+∞ | 0 < S T < + ∞ 0<S_T<+\infty 0<ST<+∞ |
Approx Price
V 0 c = σ S 0 T 2 π V_0^c=\sigma S_0\sqrt{\frac{T}{2\pi}} V0c=σS02πT
Black 76, Forward Pricing Process
let
F
t
=
e
r
(
T
−
t
)
S
t
F_t=e^{r(T-t)}S_t
Ft=er(T−t)St
d
F
t
=
σ
F
t
d
W
t
dF_t=\sigma F_tdW_t
dFt=σFtdWt
Displaced-Diffusion Model
between lognormal and normal
d
F
t
=
σ
[
β
F
t
+
(
1
−
β
)
F
0
]
d
W
t
,
β
∈
[
0
,
1
]
dF_t = \sigma[\beta F_t+(1-\beta)F_0]dW_t, \quad \beta \in [0,1]
dFt=σ[βFt+(1−β)F0]dWt,β∈[0,1]
Solving Key:
rewrite formula as
d
[
F
t
+
(
1
−
β
)
β
F
0
]
=
σ
β
[
F
t
+
(
1
−
β
)
β
F
0
]
d
W
t
d[F_t +\frac{(1-\beta)}{\beta}F_0]= \sigma\beta [F_t+\frac{(1-\beta)}{\beta}F_0]dW_t
d[Ft+β(1−β)F0]=σβ[Ft+β(1−β)F0]dWt
forward rate process is allowed to be negative value as the lowerbound to the forward process.
Vasicek Model
an example discribing rates
d
r
t
=
κ
(
θ
−
r
t
)
d
t
+
σ
d
W
t
dr_t=\kappa (\theta-r_t)dt+\sigma dW_t
drt=κ(θ−rt)dt+σdWt
Solution Key:
let
X
t
=
e
κ
t
r
t
X_t = e^{\kappa t}r_t
Xt=eκtrt
Cox-Ingersoll-Ross
Quit??
S5 Equivalent Martingale Measure (changing measure)
Girsanov
we always want to change the mean of the distrubution to 0 so that it’s a martingale. So that we need to change measure.
d Q d P = exp ( − ∫ 0 T κ t W t − 1 2 ∫ 0 T κ t 2 t ) \frac{d\mathbb{Q}}{d\mathbb{P}}=\exp(-\int_0^T\kappa_t W_t-\frac{1}{2}\int_0^T\kappa_t ^2 t) dPdQ=exp(−∫0TκtWt−21∫0Tκt2t)
Example:
- Consider
X
t
X_t
Xt given by
d X t = μ d t + σ d W t dX_t = \mu dt + \sigma dW_t dXt=μdt+σdWt
where W W W is a P \mathbb{P} P-Brownian motion. Determine if there’s a measure Q \mathbb{Q} Q such that the drift of the prcess X t X_t Xt under Q \mathbb{Q} Q is v d t vdt vdt instead of μ d t \mu dt μdt?
d X t = v X t d t + σ X t ( d W t + μ − v σ d t ) = v X t d t + σ X t d W t ~ κ t = μ − v σ \begin{aligned} dX_t&=vX_t dt + \sigma X_t(dW_t + \frac{\mu - v}{\sigma}dt)\\ &= vX_t dt + \sigma X_t d\tilde{W_t}\\ \kappa _t &= \frac{\mu - v}{\sigma} \end{aligned} dXtκt=vXtdt+σXt(dWt+σμ−vdt)=vXtdt+σXtdWt~=σμ−v
Dynamic Hedging (deriving BSM)
d
S
t
=
μ
S
t
d
t
+
σ
S
t
d
W
t
d
B
t
=
r
B
t
d
t
let
Π
=
Δ
S
−
C
\begin{aligned} dS_t&=\mu S_t dt + \sigma S_t dW_t\\ dB_t&= rB_tdt\\ \textnormal{let }\Pi&=\Delta S-C\\ \end{aligned}
dStdBtlet Π=μStdt+σStdWt=rBtdt=ΔS−C
in order for dynamic hedging
d
Π
=
Δ
d
S
+
S
d
Δ
+
d
S
d
Δ
−
d
C
\begin{aligned} d\Pi&=\Delta dS+S d\Delta + dSd\Delta -dC\\ \end{aligned}
dΠ=ΔdS+SdΔ+dSdΔ−dC
considering self financing we have
S
d
Δ
+
d
S
d
Δ
=
0
S d\Delta + dSd\Delta=0
SdΔ+dSdΔ=0
d
Π
=
Δ
d
S
−
d
C
d
C
=
∂
C
∂
S
d
S
+
1
2
∂
2
C
∂
S
2
σ
2
S
2
d
t
+
∂
C
∂
t
d
t
d
Π
=
Δ
d
S
−
(
∂
C
∂
S
d
S
+
1
2
∂
2
C
∂
S
2
σ
2
S
2
d
t
+
∂
C
∂
t
d
t
)
Δ
=
∂
C
∂
S
d
Π
=
−
(
1
2
∂
2
C
∂
S
2
σ
2
S
2
+
∂
C
∂
t
)
d
t
\begin{aligned} d\Pi&=\Delta dS -dC\\ dC&=\frac{\partial C}{\partial S}dS+\frac{1}{2}\frac{\partial^2 C}{\partial S^2}\sigma^2 S^2 dt+\frac{\partial C}{\partial t}dt\\ d\Pi&=\Delta dS -(\frac{\partial C}{\partial S}dS+\frac{1}{2}\frac{\partial^2 C}{\partial S^2}\sigma^2 S^2 dt+\frac{\partial C}{\partial t}dt)\\ \Delta &= \frac{\partial C}{\partial S}\\ d\Pi &= -(\frac{1}{2}\frac{\partial^2 C}{\partial S^2}\sigma^2 S^2 +\frac{\partial C}{\partial t})dt\\ \end{aligned}
dΠdCdΠΔdΠ=ΔdS−dC=∂S∂CdS+21∂S2∂2Cσ2S2dt+∂t∂Cdt=ΔdS−(∂S∂CdS+21∂S2∂2Cσ2S2dt+∂t∂Cdt)=∂S∂C=−(21∂S2∂2Cσ2S2+∂t∂C)dt
Since
Π
\Pi
Π is tradable, and it’s return is determinstic, arbitrage argument tells that it should drift at risk free rate.
r
Π
d
t
=
−
(
1
2
∂
2
C
∂
S
2
σ
2
S
2
+
∂
C
∂
t
)
d
t
r
(
Δ
S
−
C
)
=
−
1
2
∂
2
C
∂
S
2
σ
2
S
2
+
∂
C
∂
t
r
C
=
∂
C
∂
S
r
S
t
+
∂
C
∂
t
+
1
2
∂
2
C
∂
S
2
σ
2
S
2
\begin{aligned} r\Pi dt &= -(\frac{1}{2}\frac{\partial^2 C}{\partial S^2}\sigma^2 S^2 +\frac{\partial C}{\partial t})dt\\ r(\Delta S-C)&=-\frac{1}{2}\frac{\partial^2 C}{\partial S^2}\sigma^2 S^2 +\frac{\partial C}{\partial t}\\ rC&= \frac{\partial C}{\partial S}rS_t + \frac{\partial C}{\partial t}+\frac{1}{2}\frac{\partial ^2 C}{\partial S^2}\sigma ^2 S^2 \end{aligned}
rΠdtr(ΔS−C)rC=−(21∂S2∂2Cσ2S2+∂t∂C)dt=−21∂S2∂2Cσ2S2+∂t∂C=∂S∂CrSt+∂t∂C+21∂S2∂2Cσ2S2
It’s known as Black-Scholes PDE
Numeraire
The only requirement for a particular security to qualify as a numeraire security is that it has to be strictly positive at all times.
Justification:
- In a complete market, any derivative security is attainable. Since we can hedge a derivative product perfectly, the derivative security loses its randomness and behaves like a risk-less bond.
- If we hedge according to our risk-neutral valuation framework, then all risk is eliminated, and the hedged portfolio grows at a risk-free rate.
- Consequently, the hedged portfolio divided by the risk-free rate is a martingale.
- In other words, if one portfolio is an arbitrage in one measure, then it is an arbitrage in all other equivalent measures.
- If the option price we determined under the risk-neutral measure is arbitrage-free, then it is arbitrage-free in the real world.
- If we can express security price processes discounted by a numeraire security as a martingale, then there can be no arbitrage opportunities.
- Under the risk-neutral probabilities associated to this numeraire security, the option price is also a martingale, and we can therefore determine it’s present value.
How to Do:
Example:
- Let X t = S t B t X_t = \frac{S_t}{B_t} Xt=BtSt
- Change X t X_t Xt to risk neutral
- record d W t n e w = d W t . . . dW_t^{new}=dW_t... dWtnew=dWt...
- place d W t n e w dW_t^{new} dWtnew into S t S_t St
- apply pricing
3 ways to Derive Blake-Shole Model
Solution 1
S6 Volatility Smile
Binary Options
Type | Numeraire | |
---|---|---|
Cash or nothing | I S t > K \mathbb{I}_{S_t>K} ISt>K | Bond |
Asset or nothing | S t I S t > K S_t\mathbb{I}_{S_t>K} StISt>K | Stock |
Radon-Nikodym as Numeraire Ratio
E
Q
N
[
G
t
]
=
E
Q
M
[
G
t
N
T
/
N
0
M
T
/
M
0
]
\mathbb{E}^{\mathbb{Q}^N}[G_t]=\mathbb{E}^{\mathbb{Q}^M}[G_t\frac{N_T/N_0}{M_T/M_0}]
EQN[Gt]=EQM[GtMT/M0NT/N0]
d
Q
N
d
Q
M
=
N
T
/
N
0
M
T
/
M
0
\frac{d\mathbb{Q}^N}{d\mathbb{Q}^M}=\frac{N_T/N_0}{M_T/M_0}
dQMdQN=MT/M0NT/N0
example:
- Bound Numeraire, basing on Lognormal stock price:
d Q N d Q M = N T / N 0 M T / M 0 = S T / S 0 B T / B 0 = e ( r − σ 2 2 ) t + σ W t ∗ e r t = e − σ 2 2 t + σ W t ∗ \begin{aligned} \frac{d\mathbb{Q}^N}{d\mathbb{Q}^M}&=\frac{N_T/N_0}{M_T/M_0}\\ &=\frac{S_T/S_0}{B_T/B_0}\\ &=\frac{e^{(r-\frac{\sigma^2}{2})t+\sigma W_t^*}}{e^{rt}}\\ &=e^{-\frac{\sigma^2}{2}t+\sigma W_t^*} \end{aligned} dQMdQN=MT/M0NT/N0=BT/B0ST/S0=erte(r−2σ2)t+σWt∗=e−2σ2t+σWt∗ - Stock Numeraire, Asset or nothing call:
V 0 = e − r t E Q ∗ [ S t I S t > K ] = e − r t E Q S [ S t I S t > K d Q ∗ Q S ] = e − r t E Q S [ S t I S t > K B t / B 0 S t / S 0 ] = S 0 E Q S [ I S t > K ] \begin{aligned} V_0&=e^{-rt}\mathbb{E}^{\mathbb{Q}^*}[S_t\mathbb{I}_{S_t>K}]\\ &=e^{-rt}\mathbb{E}^{\mathbb{Q}^S}[S_t\mathbb{I}_{S_t>K}\frac{d\mathbb{Q}^*}{\mathbb{Q}^S}]\\ &=e^{-rt}\mathbb{E}^{\mathbb{Q}^S}[S_t\mathbb{I}_{S_t>K}\frac{B_t/B_0}{S_t/S_0}]\\ &=S_0\mathbb{E}^{\mathbb{Q}^S}[\mathbb{I}_{S_t>K}]\\ \end{aligned} V0=e−rtEQ∗[StISt>K]=e−rtEQS[StISt>KQSdQ∗]=e−rtEQS[StISt>KSt/S0Bt/B0]=S0EQS[ISt>K]
Since d Q S d Q ∗ = e − σ 2 2 t + σ W t ∗ \frac{d\mathbb{Q}^S}{d\mathbb{Q}^*}=e^{-\frac{\sigma^2}{2}t+\sigma W_t^*} dQ∗dQS=e−2σ2t+σWt∗, then W t S = W t ∗ + κ t W_t^S=W_t^*+\kappa t WtS=Wt∗+κt, κ = − σ \kappa = -\sigma κ=−σ.
d W t S = d W t ∗ − σ d t d S t = ( r + σ 2 ) S t d t + σ S t d W t S , S t = S 0 exp [ ( r + 1 2 σ 2 ) T + σ W T S ] S t > K \begin{aligned} dW_t^S&=dW_t^*-\sigma dt\\ dS_t&=(r+\sigma ^2)S_tdt+\sigma S_tdW_t^S,\\ S_t&=S_0\exp{[(r+\frac{1}{2}\sigma ^2)T+\sigma W_T^S]}\\ S_t&>K\\ \end{aligned} dWtSdStStSt=dWt∗−σdt=(r+σ2)Stdt+σStdWtS,=S0exp[(r+21σ2)T+σWTS]>K
then find boundary and intergrat.
Volatility Smile
background: before 1987, the crash, vol is flat. After, people want more protective put, rising the put price and thus give rise to implied vol.
Constant Vol:
try to use origional models to derive σ \sigma σ.
Black 76 lognormal model:
d F t = σ F t d W t ∗ dF_t=\sigma F_tdW_t^* dFt=σFtdWt∗
Black 76 normal model:
d F t = σ F 0 d W t ∗ dF_t=\sigma F_0dW_t^* dFt=σF0dWt∗
Black 76 displaced-diffusion model:
d
F
t
=
[
β
F
t
+
(
1
−
β
)
F
0
]
σ
d
W
t
∗
dF_t=[\beta F_t + (1-\beta)F_0]\sigma dW_t^*
dFt=[βFt+(1−β)F0]σdWt∗
solving method refer to displaced-diffusion part.
Stochastic Vol
Idea is that volatility is also stochastic!
d
σ
t
=
v
σ
t
d
W
t
σ
,
d\sigma_t = v\sigma _tdW_t^{\sigma},
dσt=vσtdWtσ,
v
v
v is volatility of volatility(vol of vol)
Heston Model
somehow he uses
V
t
V_t
Vt stands for
σ
\sigma
σ
It’s popular among equity desks
d
S
t
=
r
S
t
d
t
+
V
t
S
t
d
W
t
S
d
V
t
=
κ
(
θ
−
V
t
)
d
t
+
v
V
t
d
W
t
V
d
W
t
S
d
W
t
V
=
ρ
d
t
\begin{aligned} dS_t&=rS_tdt+\sqrt{V_t}S_tdW_t^S\\ dV_t&=\kappa (\theta - V_t)dt + v\sqrt{V_t}dW_t^V\\ dW_t^SdW_t^V&=\rho dt \end{aligned}
dStdVtdWtSdWtV=rStdt+VtStdWtS=κ(θ−Vt)dt+vVtdWtV=ρdt
SABR (Stochastic Alpha-Beta-Rho)
d F t = α t F t β d W t F d α t = v α t d W t α d W t α d W t F = ρ d t \begin{aligned} dF_t&=\alpha_tF_t^{\beta}dW_t^F\\ d\alpha_t&=v\alpha_t dW_t^{\alpha}\\ dW_t^{\alpha}dW_t^F&=\rho dt \end{aligned} dFtdαtdWtαdWtF=αtFtβdWtF=vαtdWtα=ρdt
Analysis:
β
\beta
β
determin normal or lognormal the process is.
ρ → \rho \rightarrow ρ→ skeweness
Impact on stock return distribution:
- negative correlated ρ < 0 \rho < 0 ρ<0: S t ↓ ⇔ σ r ↑ S_t \downarrow \Leftrightarrow \sigma_r \uparrow St↓⇔σr↑
- fat left tail: as things go bad, sigma (as there’s no drift) gets worse.
Impact on price:
- Out-of-money P P P: ↑ \uparrow ↑
- Out-of-money C C C: ↓ \downarrow ↓
v v v vol of vol → \rightarrow → kurtosis
Impact on stock return distribution:
- v > 0 v > 0 v>0: More volatile than before (vol has vol)
- high kurtosis on stock return-- fat tails: extreme evnets occure more ofen, same impact on good and bad side
Impact on price:
- Near-money options: ↓ \downarrow ↓
- Far-from-money options: ↑ \uparrow ↑
Vol-smile Market Overview
Equity: Smerk
FX: Smile
Rates: Smerk with high vov
Commodity: Reverse Smerk, comsumers in market worried more about price rise. Most hold calls.
Greeks and Dynamic Hedging
normally ovserve Greeks with x-axis = stock price, as this is most volatile
Useful tools
K.
d
1
=
log
S
0
K
+
(
r
+
σ
2
2
)
T
σ
T
K
e
−
r
T
=
S
0
e
σ
2
T
2
−
σ
T
d
1
\begin{aligned} d_1&=\frac{\log \frac{S_0}{K}+(r+\frac{\sigma^2}{2})T}{\sigma\sqrt{T}}\\ Ke^{-rT}&=S_0e^{\frac{\sigma^2T}{2}-\sigma\sqrt{T}d_1}\\ \end{aligned}
d1Ke−rT=σTlogKS0+(r+2σ2)T=S0e2σ2T−σTd1
Since in doing
∂
Φ
(
d
1
)
∂
S
0
\frac{\partial \Phi(d_1)}{\partial S_0}
∂S0∂Φ(d1) need to substitute K with all residual stuff to cancel terms.
Delta Δ \Delta Δ
Normal CDF alike curve
Long maturity: slowly moving up, which also means that Gamma with low kurtosis
Gamma Γ \Gamma Γ
Bell curve
Angry Gamma traders, trade highly volatile short term options
Vega (Las-vegas)
Bell curve
Elegant Vega traders, trade 1yr+ options, close to maturity deliver to Gamma traders
Long maturity: high vega, smooth.
BS’s Dynamic Hedging
explaining self financing
V
t
=
ϕ
t
S
t
+
ψ
t
B
t
d
V
t
=
ϕ
t
d
S
t
+
S
t
d
ϕ
t
+
d
S
t
d
ϕ
t
+
ψ
t
d
B
t
+
B
t
d
ψ
t
+
d
ϕ
t
d
B
t
d
V
t
=
ϕ
t
d
S
t
+
S
t
d
ϕ
t
+
d
S
t
d
ϕ
t
+
ψ
t
d
B
t
+
B
t
d
ψ
t
\begin{aligned} V_t&=\phi_t S_t + \psi_t B_t\\ dV_t &= \phi_t dS_t + S_t d\phi _t + dS_t d\phi _t + \\ &\quad \psi_t dB_t + B_td\psi_t + d\phi_tdB_t\\ dV_t &= \phi_t dS_t + S_t d\phi _t + dS_t d\phi _t + \psi_t dB_t + B_td\psi_t \end{aligned}
VtdVtdVt=ϕtSt+ψtBt=ϕtdSt+Stdϕt+dStdϕt+ψtdBt+Btdψt+dϕtdBt=ϕtdSt+Stdϕt+dStdϕt+ψtdBt+Btdψt
drop one term as
d
B
t
dB_t
dBt does not have
d
W
t
dW_t
dWt
self-financing:
d
V
t
=
ϕ
t
d
S
t
+
ψ
t
d
B
t
dV_t=\phi_t dS_t+\psi_t dB_t
dVt=ϕtdSt+ψtdBt
BS Dynamic Hedging
Let
X
t
=
C
t
−
ϕ
t
S
t
−
ψ
t
B
t
X_t=C_t - \phi_t S_t - \psi_t B_t
Xt=Ct−ϕtSt−ψtBt and
X
0
=
0
X_0=0
X0=0, aka, a replicating portfolio
apply Ito formula
d
X
t
=
d
C
t
−
ϕ
t
d
S
t
−
ψ
t
d
B
t
d
C
t
=
∂
C
∂
S
d
S
t
+
(
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
)
d
t
d
B
t
=
r
B
t
d
t
\begin{aligned} dX_t&=dC_t-\phi_t dS_t-\psi_t dB_t\\ dC_t &= \frac{\partial C}{\partial S}dS_t + (\frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2})dt \\ dB_t&=rB_tdt \end{aligned}
dXtdCtdBt=dCt−ϕtdSt−ψtdBt=∂S∂CdSt+(∂t∂C+21σ2St2∂S2∂2C)dt=rBtdt
let
d
S
t
dS_t
dSt cancel out, we have
ϕ
t
=
∂
C
∂
S
d
X
t
=
(
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
−
ψ
t
r
B
t
)
d
t
\begin{aligned} \phi_t &= \frac{\partial C}{\partial S}\\ dX_t&=(\frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2}-\psi_trB_t)dt \\ \end{aligned}
ϕtdXt=∂S∂C=(∂t∂C+21σ2St2∂S2∂2C−ψtrBt)dt
Considering the portfolio is tradable, so it should be arbitrage free, thus should drift at riskless rate
d
X
t
=
r
X
t
d
t
=
0
dX_t = rX_tdt=0
dXt=rXtdt=0
ψ
t
=
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
r
B
t
X
t
=
C
t
−
ϕ
S
t
−
ψ
t
B
t
=
X
0
=
0
C
t
=
∂
C
∂
S
S
t
+
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
r
B
t
B
t
r
C
t
=
∂
C
∂
S
r
S
t
+
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
\begin{aligned} \psi_t&=\frac{\frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2}}{rB_t} \\ X_t &= C_t-\phi S_t - \psi_t B_t=X_0=0\\ C_t&=\frac{\partial C}{\partial S}S_t + \frac{\frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2}}{rB_t}B_t\\ rC_t&=\frac{\partial C}{\partial S}rS_t + \frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2}\\ \end{aligned}
ψtXtCtrCt=rBt∂t∂C+21σ2St2∂S2∂2C=Ct−ϕSt−ψtBt=X0=0=∂S∂CSt+rBt∂t∂C+21σ2St2∂S2∂2CBt=∂S∂CrSt+∂t∂C+21σ2St2∂S2∂2C
Which is Blake-Schole PDE
Interperatation
replication portfolio consists of Stock and Bond
- option price not dependent on drift μ \mu μ.
- dynamic hedging strategy enables the seller to hedge against unfavorable market moves by cancelling the amrket trend.(option price stay same in bull or bear market)
- dont need to estimate trend μ \mu μ.
- risk of price fluctuation exists as σ \sigma σ.
- knowing the position to hold for bond and stocks.
Understanding BSM:
Understanding 1: Hedging/Replicating
C
=
∂
C
∂
S
S
t
+
∂
C
∂
t
+
1
2
σ
2
S
t
2
∂
2
C
∂
S
2
r
\begin{aligned} C &= \frac{\partial C}{\partial S}S_t + \frac{\frac{\partial C}{\partial t}+\frac{1}{2}\sigma^2 S_t^2 \frac{\partial^2 C}{\partial S^2}}{r} \end{aligned}
C=∂S∂CSt+r∂t∂C+21σ2St2∂S2∂2C
Understanding 2: Digital calls
C
=
Asset Digital
C
−
K
×
Cash Digital
C
=
E
∗
[
S
t
I
S
t
>
K
]
−
K
e
−
r
t
E
∗
[
I
S
t
>
K
]
\begin{aligned} C &= \textnormal{Asset Digital }C - K \times \textnormal{Cash Digital }C\\ &=\mathbb{E}^*[S_t\mathbb{I}_{S_t>K}] - Ke^{-rt}\mathbb{E}^*[\mathbb{I}_{S_t>K}]\\ \end{aligned}
C=Asset Digital C−K×Cash Digital C=E∗[StISt>K]−Ke−rtE∗[ISt>K]
Understanding 3: Prob under different measure
C
=
E
∗
[
S
t
I
S
t
>
K
]
−
K
e
−
r
t
E
∗
[
I
S
t
>
K
]
=
S
×
P
S
(
S
t
>
K
)
−
K
e
−
r
t
×
P
∗
(
S
t
>
K
)
\begin{aligned} C&=\mathbb{E}^*[S_t\mathbb{I}_{S_t>K}] - Ke^{-rt}\mathbb{E}^*[\mathbb{I}_{S_t>K}]\\ &= S\times \mathbb{P}^S(S_t>K)-Ke^{-rt}\times \mathbb{P}^*(S_t>K) \end{aligned}
C=E∗[StISt>K]−Ke−rtE∗[ISt>K]=S×PS(St>K)−Ke−rt×P∗(St>K)