随机过程笔记（二）：布朗运动

Random number Generation

• Normal distribution：np.random.normal(loc=0., scale=0.25, size=10000)
• Uniform distribution：np.random.uniform(a=0, b=1, size=10000)
• Random Integrate： np.random.randint(0, 2, size = 1000)

1. 正态分布随机数：np.random.normal(loc=0., scale=0.25, size=10000)
   – 默认标准正态分布
   – loc位置参数，表示均值 $\mu$，默认为0
   – scale 尺度参数，表示标准差 $\sigma$，默认为1

1 Brownian Motion

1.1 Simulation of a Brownian Motion Path<?font>

draw 𝑧1,𝑧2,…,𝑧𝑘 drawn 𝑖.𝑖.𝑑. from 𝑁(0,1), so that we can define:

• $B_{t_1}=B_0+\sqrt{t_1}{\varphi }_1$
• $B_{t_2}=B_{t_1}+\sqrt{t_2-t_1}{\varphi }_2=B_0+\sqrt{t_1}{\varphi }_1+\sqrt{t_2-t_1}{\varphi }_2$
• $B_{t_3}=B_{t_2}+\sqrt{t_3-t_2}{\varphi }_3=B_0+\sqrt{t_1}{\varphi }_1+\sqrt{t_2-t_1}{\varphi }_2+\sqrt{t_3-t_2}{\varphi }_3$
• $B_{t_n}=B_{t_{n-1}}+\sqrt{t_n-t_{n-1}}{\varphi }_n=B_0+\sqrt{t_1}{\varphi }_1+\sqrt{t_2-t_1}{\varphi }_2+...+\sqrt{t_n-t_{n-1}}{\varphi }_n$

Hence, we know:
$$B_{t_n}=\sum _{i=1}^n{\varphi }_i\sqrt{t_i-t_{i-1}}$$

## path for diffusion item, assume the drift equal to zero
k = 10
z = np.random.normal(size=k-1) # draw 9 samples from standard normal distribution
t_max = 20. # set T
t = np.arange(0, t_max, t_max/k) # generate time sequence
dt = np.diff(t) # dt1 = t1 - t0; dt2 = t2 - t1; ...
# np.sqrt(dt) * z : the diffusion item
# np.concatenate([[0.0],x]) : Assign 0.0 to the initial value
plt.plot(t, np.concatenate([[0.0], np.cumsum(np.sqrt(dt) * z)]), '+-', 'b')
plt.xlabel('$t$'); plt.ylabel('$B_t$'); plt.show()

1.2 Brownian Motion with Drift in Python

$X_t=\sigma \varphi B_t+\mu t$

• $X_{t_1}=X_0+\sigma {\varphi }_1\sqrt{t_1}+\mu t_1$
• $X_{t_2}=X_{t_1}+\sigma {\varphi }_2\sqrt{t_2-t_1}+\mu t_2=B_0+\sigma {\varphi }_1\sqrt{t_1}+\sigma {\varphi }_2\sqrt{t_2-t_1}+\mu t_1+\mu t_2$
• $X_{t_3}=X_{t_2}+\sigma {\varphi }_3\sqrt{t_3-t_2}+\mu t_3=B_0+\sigma {\varphi }_1\sqrt{t_1}+\sigma {\varphi }_2\sqrt{t_2-t_1}+\sigma {\varphi }_3\sqrt{t_3-t_2}+\mu t_1+\mu t_2+\mu t_3$
• $X_{t_n}=X_{t_{n-1}}+\sigma {\varphi }_n\sqrt{t_n-t_{n-1}}+\mu t_n=B_0+\sigma {\varphi }_1\sqrt{t_1}+\sigma {\varphi }_2\sqrt{t_2-t_1}+...+\sigma {\varphi }_n\sqrt{t_n-t_{n-1}}+\mu t_1+\mu t_2+...+\mu t_n$
  $$X_{t_n}=\sum _{i=1}^n\sigma {\varphi }_i\sqrt{t_i-t_{i-1}}+\mu (t_i-t_{i-1})$$

sigma = 0.005; mu = 0.005; k = 100; T = 10.
z = np.random.normal(size=k-1)
t = np.arange(0, T, T/k)
dt = np.diff(t)
plt.plot(t[1:], np.cumsum(sigma*np.sqrt(dt)*z + mu*dt))
plt.xlabel('$t$'); plt.ylabel('$X_t$'); plt.title('$\mu=+0.005$')
plt.show()

1.3 Brownian Motion with Multiple Paths in Python

import numpy as np
import matplotlib.pyplot as plt
from numpy.random import normal

initial_value = 100.0
t_max = 100
n = 10
sigma = 0.005
mu = 0

random_numbers = normal(loc=mu, scale=sigma, size=(t_max,n))  # 每个随机过程占 1 列
multipliers = 1 + random_numbers
values = initial_value * np.cumprod(multipliers,axis=0) # 按列累积（累乘）
values_multiple_initial = np.concatenate((np.matrix([initial_value]*n), values),axis=0) # 加一行初始值（给每列加一个）
plt.xlabel('$t$')
plt.ylabel('$y_t$')
plt.plot(values_multiple_initial)
plt.plot

2 Geometric Brownian Motion

2.1 Simulating Geometric Brownian Motion

The $\mu$ below is the drift of S (lognormal)
$$dS=\mu Sdt+\sigma SdX$$

• S t 1 = S 0 e x p { σ ϕ 1 t 1 + μ t 1 } S_{t_1} = S_0 exp\{ \sigma \phi_1\sqrt{t_1} + \mu t_1\} St1​​=S0​exp{σϕ1​t1​ ​+μt1​}
• S t 2 = S t 1 e x p { σ ϕ 2 t 2 − t 1 + μ t 2 } = S 0 e x p { σ ϕ 1 t 1 + σ ϕ 2 t 2 − t 1 + μ t 1 + μ t 2 } S_{t_2} = S_{t_1} exp\{\sigma \phi_2\sqrt{t_2 - t_1} + \mu t_2 \}= S_0 exp\{ \sigma \phi_1\sqrt{t_1} + \sigma \phi_2\sqrt{t_2 - t_1} + \mu t_1 + \mu t_2\} St2​​=St1​​exp{σϕ2​t2​−t1​ ​+μt2​}=S0​exp{σϕ1​t1​ ​+σϕ2​t2​−t1​ ​+μt1​+μt2​}
• S t 3 = S t 2 e x p { σ ϕ 3 t 3 − t 2 + μ t 3 } = S 0 e x p { σ ϕ 1 t 1 + σ ϕ 2 t 2 − t 1 + σ ϕ 3 t 3 − t 2 + μ t 1 + μ t 2 + μ t 3 } S_{t_3} = S_{t_2} exp \{ \sigma \phi_3\sqrt{t_3 - t_2} + \mu t_3 \}= S_0 exp\{ \sigma \phi_1\sqrt{t_1} + \sigma \phi_2\sqrt{t_2 - t_1} + \sigma \phi_3\sqrt{t_3 - t_2} + \mu t_1 + \mu t_2 + \mu t_3\} St3​​=St2​​exp{σϕ3​t3​−t2​ ​+μt3​}=S0​exp{σϕ1​t1​ ​+σϕ2​t2​−t1​ ​+σϕ3​t3​−t2​ ​+μt1​+μt2​+μt3​}
• S t n = S t n − 1 e x p { σ ϕ n t n − t n − 1 + μ t n } = S 0 e x p { σ ϕ 1 t 1 + σ ϕ 2 t 2 − t 1 + . . . + σ ϕ n t n − t n − 1 + μ t 1 + μ t 2 + . . . + μ t n } S_{t_n} = S_{t_{n-1}} exp\{\sigma \phi_n\sqrt{t_n - t_{n-1}} +\mu t_n \}= S_0 exp\{ \sigma \phi_1\sqrt{t_1} + \sigma \phi_2\sqrt{t_2 - t_1} +...+ \sigma \phi_n\sqrt{t_n - t_{n-1}} + \mu t_1 + \mu t_2 +...+ \mu t_n\} Stn​​=Stn−1​​exp{σϕn​tn​−tn−1​ ​+μtn​}=S0​exp{σϕ1​t1​ ​+σϕ2​t2​−t1​ ​+...+σϕn​tn​−tn−1​ ​+μt1​+μt2​+...+μtn​}
  S t k = S 0 ∏ i = 1 k e x p { σ ϕ t i − t i − 1 + μ ( t i − t i − 1 ) } S_{t_k} = S_0 \prod\limits_{i=1}^{k}exp\{\sigma \phi \sqrt{t_i-t_{i-1}}+\mu (t_i - t_{i-1})\} Stk​​=S0​i=1∏k​exp{σϕti​−ti−1​ ​+μ(ti​−ti−1​)}

sigma = 0.4 # relevantly small
mu = 0.1; k = 1000; t_max = 12.0; S0 = 100.0
dt = t_max / k
T = np.arange(0, t_max, dt)
z = np.random.normal(size=len(T))
S = S0 * np.cumprod(np.exp(sigma*np.sqrt(dt)*z + mu*dt))
plt.plot(T, S, '+-')
plt.xlabel('$t$'); plt.ylabel('$S_t$'); plt.show()

2.2 GBM with Multiple Paths in Python

The $r$ here is the drift of $X_t$(normal)
$$dX=rdt+\sigma d{B}_t$$
$$dS=(r-\frac{1}{2}{\sigma }^2)Sdt+\sigma \varphi SdX$$

def gbm(sigma, r, k, t_max, S0, I=1):
    phi = np.random.normal(size=(k-1, I))  
    dt = t_max/k
    y = sigma * np.sqrt(dt)*phi + (r - sigma**2 / 2.) * dt
    return S0 * np.exp(np.cumsum(y, axis=0))  
sigma = 0.05; r = 0.01; k = 100; t_max = 10.; S0 = 100.
T = np.arange(0, t_max - t_max/k, t_max/k)
ax = plt.plot(T, gbm(sigma, r, k, t_max, S0, I=50))
plt.xlabel('$t$'); plt.ylabel('$S_t$'); plt.show()