Itô Integral, Itô Formula
约 641 个字 2 张图片 预计阅读时间 3 分钟
Itô Integral¶
可以将 diffusion 公式化为
\[ \begin{equation} \mathrm{d}X_t = \mu(t, X_t)\mathrm{d}t + \sigma(t, X_t)\mathrm{d}B_t \end{equation} \]
这里 \(\{ B_t\}_{t \ge 0}\) 为 SBM
对式(1)积分得
\[ \begin{equation} \forall T, X_T = X_0 + \int_{0}^{T}\mu(t, X_t)\mathrm{d}t + \int_{0}^{T}\sigma (t, X_t)\mathrm{d}B_t \end{equation} \]
\[ \begin{equation*} \forall T, \forall \omega \in \Omega, X_T(\omega) = X_0(\omega) + \int_{0}^{T}\mu(t, X_t(\omega))\mathrm{d}t + \int_{0}^{T}\sigma (t, X_t(\omega))\mathrm{d}B_t \end{equation*} \]
- \(\int_{0}^{T}\mu(t, X_t)\mathrm{d}t\) 为一个常见的黎曼积分
- \(\int_{0}^{T}\sigma (t, X_t(\omega))\mathrm{d}B_t\) 为一个 Itô Integral
Riemann-Stieltjes Integral¶
记 \(F: [0, T] \rightarrow \mathbb{R}\) 是一个导函数有界的函数,区间 \([0, T]\) 的一个划分为:\(0 = t_0 \le t_1 \le \cdots \le t_n = T\),定义
\[ \begin{align*} \int_{0}^{T}X\mathrm{d}F &= \lim_{n \rightarrow \infty} \sum_{i=1}^n X(t_i^*)(F(t_i) - F(t_{i-1})) \\ &= \lim_{n \rightarrow \infty} \sum_{i=1}^{n} X(t_i^*)F'(t_{i-1})(t_i - t_{i-1}) \\ &= \int_{0}^{T}X(t)F'(X(t))\mathrm{d}t \end{align*} \]
常用于计算期望:\(E[X] = \int_{\Omega}X\mathrm{d}F\)
关于 \(\int_{0}^{T}F(x)\mathrm{d}F(x)\) 容易得到:
Itô Integral¶
这里 \(\int_{0}^{T}B_t\mathrm{d}B_t = \frac{1}{2}B_T^2 - \frac{1}{2}T\) 是因为 \(\lim_{\Delta \rightarrow 0}\frac{1}{2}\sum_{i=1}^{n}(B(t_i) - B(t_{i-1}))^2\) 不一定收敛了,因此需要引入 mean square convergence 的概念来处理,记 \(Q_n = \sum_{i=1}^{n}(B(t_i) - B(t_{i-1}))^2\),可以算出 \(E[Q_n] = T, \operatorname{Var}[Q_n] \rightarrow 0\),故最终出现了 \(-\frac{1}{2}T\) 这一项。
Itô Formula¶
由于对任意 \(T\)
\[ \lim_{n \rightarrow \infty}Q_n = \int_{0}^{T}(\mathrm{d}B_t)^2 = T = \int_{0}^{T}\mathrm{d}t \]
可推断出 \((\mathrm{d}B_t)^2 \approx \mathrm{d}t\)
故对于式(1),以及函数 \(f\),有
\[ \begin{align*} \mathrm{d}f(X_t) &= f(X_t + \mathrm{d}X_t) - f(X_t) \\ &= f'(X_t)\mathrm{d}X_t + \frac{1}{2}f''(X_t)(\mathrm{d}X_t)^2 + o((\mathrm{d}X_t)^2) \\ &= f'(X_t)(\mu_t \mathrm{d}t + \sigma_t\mathrm{d}B_t) + \frac{1}{2}f''(X_t)(\mu_t \mathrm{d}t + \sigma_t\mathrm{d}B_t)^2 + o((\mathrm{d}X_t)^2) \\ & \stackrel{\mathrm{d}t \rightarrow 0}{\longrightarrow} \left(f'(X_t)\mu + \frac{1}{2}f''(X_t) \sigma _t^2\right)\mathrm{d}t + f'(X_t)\sigma _t\mathrm{d}B_t \end{align*} \]
这就是 Itô Formula。
可以用这个公式计算:
- \(\mathrm{d}(B_t^2) = 2 B_t \mathrm{d}B_t + \mathrm{d}t\)
- Geometric Brownian Motion: 对 \(Y_t = e^{B_t}\),有 \(\mathrm{d}Y_t = Y_t \mathrm{d}B_t + \frac{1}{2}Y_t \mathrm{d}t\)
- Ornstein-Uhlenbeck Process: 对 \(f(t, X_t) = e^t X_t\),\(\{ X_t\}\) 为 Ornstein-Uhlenbeck process 且 \(X_0 = 0\),有 \(\mathrm{d}f(t, X_t) = 2e^t\mathrm{d}B_t\)