Professional Documents
Culture Documents
Introduction To Stochastic Calculus
Introduction To Stochastic Calculus
Arun Kumar
IIT Ropar
1 Ordinary Calculus
3 References
Stochastic Calculus
Since Brownian sample paths are not differentiable, the ordinary calculus intuition will
dB(t)
not work because there is no process B ′ (t) = dt , and hence we cannot express
Z t
g(s)dB(s)
0
as Z t
g(s)B ′ (s)ds.
0
Ordinary Calculus
Suppose
c if t ∈ (a, b]
f (t) = cI(a,b] (t) =
0 otherwise .
Suppose
c if t ∈ (a, b]
f (t) = cI(a,b] (t) =
0 otherwise .
df (x(t)) dx(t)
= f ′ (x(t))
dt dt
or in differential form
df (x(t)) = f ′ (x(t))dx(t).
We prove this by using Taylor series expansion such that
1 ′′
df (x(t)) = f ′ (x(t))dx(t) + f (x(t))dx(t)dx(t) + · · · .
2!
Here df(x(t)) denote the infinitesimal increment of f (x(t)). Note that dx(t)dx(t)
corresponds to the quadratic variation of the function x(·).
Ito’s Formula
For f : R → R be a function which has derivatives of all orders. Then using Taylor
series expansion of f , we get
1 ′′ 1
df (X (t)) = f ′ (X (t))dX (t)+ f (X (t))dX (t)dX (t)+ f ′′′ (X (t))dX (t)dX (t)dX (t)+· · · ,
2! 3!
since variation of X (·) of order 3 and more are zero, we get
1 ′′
df (X (t)) = f ′ (X (t))dX (t) + f (X (t))dX (t)dX (t).
2!
1
df (X (t)) = [Q(t)f ′ (X (t)) + P(t)2 f ′′ (X (t))]dt + P(t)f ′ (X (t))dB(t).
2
Ito Formula Extended
Proposition
For an Ito process dX (t) = P(t)dB(t) + Q(t)dt, the Ito’s formula is given by
Proof.
Using Taylor series expansion, we have
1
df (t, X (t)) = ft (t, X (t))dt + fx (t, X (t))dX (t) + fxx (t, X (t))dX (t)dX (t).
2
• Under Bachelier model stock price is governed by the equation dS(t) = σdB(t).
• σ̂ = sd(St − St−1 ).
Geometric Brownian motion
1 1 1 −1 2
d ln(S(t)) = 0 + P(t)dB(t) + Q(t)dt + P (t)dt.
S(t) S(t) 2 S 2 (t)
• By substituting P(t) = σS(t) and Q(t) = µS(t), it follows
1 1 1 −1
d ln(S(t)) = σS(t)dB(t) + µS(t)dt + σS 2 (t)dt
S(t) S(t) 2 S 2 (t)
σ2
= σdB(t) + (µ − )dt.
2
σ2
• Which in turn implies S(t) = S(0)e(µ− 2
)t+σB(t)
.
Estimation of parameters of GBM
σ2
• We have S(t) = S(0)e(µ− 2
)t+σB(t)
.
σ2
• Thus the log-returns r (t) = log S(t)
S(t−1)
= (µ − 2
) + σ(B(t) − B(t − 1)).
σ2
• Hence r (t) is normal with mean µ − and variance σ 2 .
2
• Hence under the GBM model, the parameters can be estimated by daily log
returns of the equity.
Simulation of GBM
• Generate N = T /∆t normal random numbers with mean 0 and variance σ 2 ∆t.
• Take the cumulative sum of these random numbers. Call this vector as p1 .
2
• Calculate µ − σ j∆t, j = 1, 2, . . . , N. Call this vector p2 .
2
• https://cran.r-project.org/web/packages/sde/sde.pdf
THANK YOU