Bootcamp

Topics related to measure theory.

略去,详见测度论专栏中的文章


Expectations

令 \(X\) 为 \((\Omega, \mathcal{F}, P)\) 上的随机变量,\(\mathbb{E}[X]\) 为其期望。一些期望的特殊表示如下:

  • \(X: \Omega \rightarrow \mathbb{R}\) 为简单函数,即,\(X\) 在有限集 \(\left\{x_{1},\ldots, x_{n} \right\}\) 中取值,则:

    \[\mathbb{E}[X] := \sum\limits^{n}_{i=1} x_{i} P(X = x_{i})
    \]
  • \(X \geq 0\) almost surely,则:

    \[\mathbb{E}[X] := \sup \left\{ \mathbb{E}[Y]: ~ Y \mbox{ is simple, } ~ 0 \leq Y \leq X \mbox{ almost surely. } \right\}
    \]

    注意,非负随机变量的期望可能为 \(\infty\)。

  • \(\mathbb{E}[X^{+}]\) 或 \(\mathbb{E}[X^{-}]\) 其中之一是有限的,则:

    \[\mathbb{E}[X] := \mathbb{E}[X^{+}] - \mathbb{E}[X^{-}]
    \]
  • \(X\) 为一个向量,且 \(\mathbb{E}[|X|] < \infty\),则:

    \[\mathbb{E}\Big[\left(X_{1}, \ldots, X_{d}\right)\Big] := \Big( \mathbb{E}[X_{1}], \ldots, \mathbb{E}[X_{d}] \Big)
    \]

Jensen's Inequality (琴生不等式)

令 \(X\) 为一个随机变量,\(g: \mathbb{R} \rightarrow \mathbb{R}\) 为一个凸函数。那么当 \(X\) 的期望存在时:

\[\mathbb{E}[g(X)] \geq g\left(\mathbb{E}[X] \right)
\]

若 \(g\) 为严格凸函数,则以上不等式可随之写为严格大于的形式(除非 \(X\) 取常数值)。


  • 注(Convex function):

    函数 \(f: X \rightarrow \mathbb{R}\) 称作一个凸函数,如果:

    \[\forall ~ t \in [0, ~ 1]: ~ \forall ~ x_{1}, x_{2} \in X: ~ f\Big( tx_{1} + (1-t) x_{2} \Big) \leq t\cdot f(t x_{1}) + (1-t) \cdot f(x_{2})
    \]

Self-Financing Condition

A self-financing strategy is defined as a consumption stream \((c_{t})_{t\geq 0}\) which follows:

\[(c_{t} - c_{t+1})\cdot P_{t} = 0 \qquad \quad \mbox{for } \forall t \geq 0
\]

Numeraire (计价单位)

  • \((\eta_t)_{t\geq 0}\) 为 previsible process.

  • \(\eta_{t} \cdot P_{t} > 0\) almost surely, i.e., \(P(\eta_t \cdot P_{t} > 0) = 1\).

  • \((\eta_{t})_{t\geq 0}\) 满足 self-financing condition, i.e.,

    \[(\eta_{t} - \eta_{t+1}) \cdot P_{t} = 0 \qquad \quad \mbox{for } \forall t\geq 0
    \]

    这实际上意味着:

    \[\eta_{t} \cdot P_{t} = \eta_{t+1} \cdot P_{t} \qquad \qquad \text{for } ~ \forall t \geq 0
    \]

    注意,以上式子中两侧的 \(P_{t}\) 不能随手约去,因为等式两边是两个向量的内积运算。


Numeraire Asset

  • A numeraire asset is an asset with strictly positive price.

  • 若 asset \(i\) 为一个 numeraire asset,那么对于 \(\forall t \geq 0\),定义 constant portfolio \(\eta\):

    \[\eta_{t}^{j} = \begin{cases}
    1 \qquad \text{if } j = i\\
    0 \qquad \text{otherwise}
    \end{cases}
    \]

    为一个 numeraire portfolio。


Investment-Consumption Strategy

\[\begin{align*}
c_{0} & = x - H_{1} \cdot P_{0}\\
c_{t} & = (H_{t} - H_{t+1}) \cdot P_{t} \qquad \qquad \mbox{for } t \geq 1
\end{align*}
\]

其中 \(x\) 为初始财富。


Terminal Consumption Strategy

\[\begin{align*}
c_{0} & = -H_{1} \cdot P_{0} = 0\\
c_{t} & = (H_{t} - H_{t+1}) \cdot P_{t} = 0 \qquad \qquad \mbox{for } 1 \leq t \leq T-1\\
c_{T} & = H_{T} \cdot P_{T} \geq 0 \\
\mbox{and} \qquad \qquad \\
P( &c_{T} > 0) > 0
\end{align*}
\]

其中 \(H\) 为 previsible process,non-random \(T > 0\) 使得以上 holds almost surely。


Pure Investment Strategy

对于 \(\forall t \geq 0\),每一期持仓 \(H_{t}\),但将每一期的 consumption \(c_{t}\) 不用于消费,而是用于投资 numeraire portfolio \(\eta_{t}\)。


Theorem. 局部鞅 \(\rightarrow\) 鞅的充分条件 (local martingales to true martingales: sufficient condition)

令 \(X\) 为一个离散或连续的 local martingale,令过程 \((Y_{t})_{t\geq 0}\) 满足:

\[\mbox{for } ~ \forall ~ s,t, ~ 0 \leq s \leq t: ~ |X_{s}| \leq Y_{t} \mbox{ almost surely}
\]

若 \(\mathbb{E}[Y_{t}] \leq \infty, ~ \mbox{ for } ~ \forall ~ t \geq 0\),那么 \(X\) 为一个 true martingale。


证明:

由于 \((X_{t})_{t\leq 0}\) 为一个 local martingale,根据定义存在一个 stopping time series (localizing sequence):\((\tau_{N})_{N\geq0}\),满足 \(\lim \limits_{N \rightarrow \infty} \tau_{N} = \infty\),使得对于 \(\forall ~ N \geq 0\),\(\Big(X^{\tau_{N}}_{t}\Big)_{t \geq 0} = \Big(X_{t \land \tau_{N}}\Big)_{t\geq 0}\) 为 true martingale。

首先证明 \((X_{t})_{t\geq 0}\) 可积。对于任意 \(t \geq 0\),取任意 \(T \geq t\),根据条件:\(|X_{t}| \leq Y_{T}\) almost surely。又因为:\(\forall ~ T \geq 0: ~ \mathbb{E}[Y_{T}] < \infty\),那么:

\[\mbox{for } ~ \forall ~ t \geq 0: ~ |X_{t}| \leq Y_{T} \quad \implies \quad \mathbb{E}[X_{t}] \leq \mathbb{E}[Y_{T}] < \infty
\]

因此 \((X_{t})_{t\geq 0}\) integrable。

将 \(X_{t\land\tau_{N}}\) 视作一个下标为 \(N\) 的序列,即:

\[\Big\{ X_{t\land \tau_{N}} \Big\}_{N\geq 0} = X_{t\land \tau_{1}}, ~ X_{t\land \tau_{2}}, ~ X_{t\land \tau_{3}}, ~ \ldots
\]

注意到 \(X_{t\land \tau_{N}} = X_{\min(t, \tau_{N})} \longrightarrow X_{t}\) almost surely with \(N \longrightarrow \infty\),即:

\[\mbox{for } ~ \forall ~ t \geq 0: ~ \forall ~ \varepsilon > 0: ~ P\left( \lim\limits_{N \rightarrow \infty} \left| X_{t\land \tau_{N}} - X_{t} \right| > \varepsilon \right) = 0
\]

这是因为 \(\lim \limits_{N \rightarrow \infty} \tau_{N} = \infty\),\(t \land \tau_{N} = \min(t, \tau_{N})\) 自然随 \(N\) 增大而收敛于 \(t\)。

所以对于 \(\forall ~ 0 \leq s \leq t\):

\[\begin{align*}
\mathbb{E}[X_{t} ~ | ~ \mathcal{F}_{s}] & = \mathbb{E}\Big[\lim\limits_{N\rightarrow \infty}X_{t\land \tau_{N}} ~ | ~ \mathcal{F}_{s}\Big]\\
& = \lim\limits_{N \rightarrow \infty} \mathbb{E}\Big[ X_{t\land\tau_{N}} ~ | ~ \mathcal{F}_{s}\Big] \quad (\mbox{Dominated Convergence Theorem})\\
& = \lim\limits_{N \rightarrow \infty} X_{s \land \tau_{N}} \quad (\mathbf{*})\\
& = X_{s}
\end{align*}
\]

因此:local martingale \((X_{t})_{t\geq 0}\) 在给定的条件下也为一个 true martingale。


  • 注意:

    以上带星号的那一步推导中,鞅 \(\Big(X_{t\land\tau_{N}}\Big)_{t\geq 0}\) 的下标依然是 \(t\),尽管现在复合为 \(t\land \tau_{N}\)。因此在这一步中我们只需将 \(t\) 替换为 \(s\) 即可。


Corollary.

假设 \(X\) 一个 离散 时间 local martingale,使对于 \(\forall ~ t \geq 0: ~ \mathbb{E}[|X_{t}|] < \infty\),那么 \(X\) 是一个 true martingale。


证明:

令 \(Y_{t} = |X_{0}| + |X_{1}| + \cdots + |X_{t}|\)。Trivially:

\[Y_{t} = |X_{0}| + |X_{1}| + \cdots + |X_{t}| \geq |X_{s}| ~ \mbox{ for } ~ \forall s \in \left\{0, 1, \ldots, t \right\}
\]

并且由于:\(\forall ~ t \geq 0: ~ \mathbb{E}[|X_{t}|] < \infty\),那么:

\[\begin{align*}
\mathbb{E}[Y_{t}] & = \mathbb{E}\Big[ \left|X_{0}\right| + \left|X_{1}\right| + \cdots + \left|X_{t}\right| \Big]\\
& = \sum\limits^{t}_{s=0}\mathbb{E}\big[ \left| X_{s} \right| \big] < \infty
\end{align*}
\]

所以 \((Y_{t})_{t\geq 0}\) 可积,并且此时 \((X_{t})_{t \leq 0}\) 和 \((Y_{t})_{t\geq 0}\) 恰满足上述 Sufficient Condition,因此 \((X_{t})_{t\geq 0}\) 为一个 true martingale。


Supermartingale and Submartingale (上鞅与下鞅)

上鞅(Supermartingale)

相关于 filtration \(\mathcal{\left\{ F_{t} \right\}}_{t\geq 0}\) 的一个 supermartingale(上鞅)是一个 adapted stochastic process \((U_{t})_{t\geq 0}\),满足以下性质:

  • (Integrability)

    \[\forall ~ t \geq 0: ~ \mathbb{E}\big[\left| U_{t} \right|\big] < \infty
    \]
  • (Decrease in average)

    \[\forall ~ 0 \leq s \leq t: ~ \mathbb{E}\big[U_{t} ~ | ~ \mathcal{F}_{s}\big] \leq U_{s}
    \]

下鞅(Submartingale)

相关于 filtration \(\mathcal{\left\{ F_{t} \right\}}_{t\geq 0}\) 的一个 submartingale(下鞅)是一个 adapted stochastic process \((V_{t})_{t\geq 0}\),满足以下性质:

  • (Integrability)

    \[\forall ~ t \geq 0: ~ \mathbb{E}\big[ | V_{t} | \big] < \infty
    \]
  • (Increase in average)

    \[\forall ~ 0 \leq s \leq t: ~ \mathbb{E}\big[V_{t} ~ | ~ \mathcal{F}_{s}\big] \geq V_{s}
    \]

鞅、上鞅、下鞅

A martingale is a stochastic process that is both a supermartingale and a submartingale.


Theorem.

假设 \(X\) 是一个连续或离散时间上的 local martingale。如果 \(X_{t} \geq 0\) 对于 \(\forall ~ t \geq 0\) 都成立,那么 \(X\) 是一个 supermartingale(上鞅)。


证明:

令 \((\tau_{N})_{N\geq 0}\) 为相关于 local martingale \((X_{t})_{t\geq 0}\) 的 localizing sequence,即:

\[\forall ~ N \geq 0: ~ \Big(X^{\tau_{N}}_{t} \Big)_{t\geq 0} ~ \mbox{ is a true martingale.}
\]

首先证明 \((X_{t})_{t \geq 0 }\) 可积。由 Fatou's Lemma

\[\begin{align*}
\mathbb{E}\big[|X_{t}|\big] & = \mathbb{E}[X_{t}] \\
& = \mathbb{E}\Big[\lim\limits_{N \rightarrow \infty} X_{t \land \tau_{N}}\Big] \\
& = \mathbb{E}\Big[\liminf\limits_{N \rightarrow \infty} X_{t \land \tau_{N}}\Big] \\
& \leq \liminf\limits_{N \rightarrow \infty} \mathbb{E}\Big[X_{t\land \tau_{N}}\Big] \\
& = \liminf\limits_{N \rightarrow \infty} \mathbb{E}\Big[X_{t\land \tau_{N}} ~ \Big| ~ \mathcal{F}_{0} \Big] \\
& = X_{0} < \infty
\end{align*}
\]

在条件期望上运用 Fatou's Lemma,对于 \(\forall ~ 0 \leq s \leq t:\)

\[\begin{align*}
\mathbb{E}\big[X_{t} ~ | ~ \mathcal{F}_{s}\big] & = \mathbb{E}\Big[ \lim\limits_{N \rightarrow \infty} X_{t\land \tau_{N}} ~ \Big| ~ \mathcal{F}_{s} \Big] \\
& = \mathbb{E}\Big[ \liminf\limits_{N \rightarrow \infty} X_{t\land \tau_{N}} ~ \Big| ~ \mathcal{F}_{s} \Big] \\
& \leq \liminf_{N \rightarrow \infty} \mathbb{E}\Big[ X_{t\land \tau_{N}} ~ \Big| ~ \mathcal{F}_{s} \Big] \\
& = \liminf_{N \rightarrow \infty} X_{s \land \tau_{N}} \\
& = X_{s}
\end{align*}
\]

因此 \((X_{t})_{t\geq 0}\) 为一个 supermartingale(上鞅)。


Corollary.

如果 \((X_{t})_{t\geq 0}\) 是一个离散时间 local martingale,且对于任意 $ t \geq 0$,有 \(X_{t} \geq 0\) almost surely,那么 \((X_{t})_{t\geq 0}\) 是一个 true martingale。


证明:

通过上述 Theorem,我们有:

\[\mathbb{E}\big[|X_{t}|\big] = \mathbb{E}[X_{t}] \leq X_{0} < \infty
\]

由于 \(X\) 是可积的,通过上一条 Corollary 可以得出 \((X_{t})_{t\geq 0}\) 是一个 martingale 的结论。


Theorem.

假设:

\[X_{t} = X_{0} + \sum\limits^{t}_{s=1} K_{s} (M_{s} - M_{s-1})
\]

其中,\(K\) 是一个 previsible process,\(M\) 是一个 local martingale,\(X_{0}\) 是一个常数。

如果对于某些非随机的 \(T > 0\),有:\(X_{T} \geq 0\) almost surely,那么 \((X_{t})_{0\leq t \leq T}\) 是一个 true martingale。


证明:

略。(太长了,以后有机会补上。)


随机贴现因子(Stochastic Discount Factor / Pricing Kernel / State Price Density)

在一个没有股息的市场中,在时刻 \(s\) 和 \(t\) 间(\(0 \leq s < t\))的随机贴现因子是一个 adapted positive \(\mathcal{F}_{t}-\) measurable random variable \(\rho_{s,t}\), 使得:

\[P_{s} = \mathbb{E}\big[\rho_{s,t}P_{t} ~ | ~ \mathcal{F}_{s}\big]
\]

  • 令 \(Y\) 为一个 martingale deflator(i.e. \(\forall 0 \leq s < t: ~ \mathbb{E}[Y_{t}P_{t} ~ | ~ \mathcal{F}_{s}] = Y_{s}P_{s}\)),令 \(\rho_{s,t} = \frac{Y_{t}}{Y_{s}}\),若 \(\rho_{s,t}P_{t}\) 可积,那么 \(\rho_{s,t}\) 为时间 \(s\) 与 \(t\) 间的 pricing kernel。

    • 证明:

      对于 positivity,由于 \(Y\) 为 martingale deflator,则 \(\forall t \geq 0: ~ Y_{t} > 0\),所以 \(\rho_{s,t} = \frac{Y_{t}}{Y_{s}} > 0\),并且:

      \[\begin{align*}
      \mathbb{E} \big[ \rho_{s,t} P_{t} ~ | ~ \mathcal{F}_{s} \big] & = \mathbb{E} \Big[ \frac{Y_{t}}{Y_{s}} P_{t} ~ | ~ \mathcal{F}_{s} \Big] \\
      & = \frac{1}{Y_{s}} \mathbb{E} \big[ Y_{t}P_{t} ~ | ~ \mathcal{F}_{s} \big] \\
      & = \frac{1}{Y_{s}} \cdot Y_{s} P_{s} \\
      & = P_{s}
      \end{align*}
      \]

      因此 \(\rho_{s,t}\) 为一个 pricing kernel。

  • 相反地,对于 \(s\geq 0\),假设 \(\rho_{s, s+1}\) 为 时间 \(s\) 与 \(s+1\) 间的 pricing kernel,令 \(Y_{t} = \rho_{0,1} \rho_{1,2} \ldots \rho_{t-1, t}\),且 \(YP\) 可积,那么 \(Y\) 为一个 martingale deflator。

    • 证明:

      对于 \(\forall t \geq 0\),由于 pricing kernel 为正随机变量,则 \(Y_{t} = \rho_{0,1} \rho_{1,2} \ldots \rho_{t-1, t} > 0\),并且:

      \[\begin{align*}
      \mathbb{E} \big[Y_{t+1}P_{t+1} ~ \big| ~ \mathcal{F}_{t} \big] & = \mathbb{E} \big[\rho_{0,1} \rho_{1,2} \ldots \rho_{t-1, t} \rho_{t, t+1} \cdot P_{t+1} ~ \big| ~ \mathcal{F}_{t} \big] \\
      & = \rho_{0,1} \rho_{1,2} \ldots \rho_{t-1, t} \cdot \mathbb{E} \big[\rho_{t, t+1} \cdot P_{t+1} ~ \big| ~ \mathcal{F}_{t} \big] \qquad \text{(adaptness)}\\
      & = Y_{t} \cdot P_{t} \qquad \text{(by definition)}
      \end{align*}
      \]

      因此,\((Y_{t})_{t\geq 0}\) 为一个 martingale deflator。


Proposition.

考虑存在一个 numeraire \(\eta\) 的市场,且令:\(N_{t} = \eta_{t} \cdot P_{t} \quad \forall t \geq 0\)。令 \(H\) 为一个 investment-consumption strategy,即,\(H\) 的 consumption stream 定义为:

\[\begin{align*}
c_{0} & = x - H_{1} \cdot P_{0}\\
c_{t} & = (H_{t} - H_{t+1}) \cdot P_{t}
\end{align*}
\]

其中 \(x\) 为初始财富。令:

\[K_{t} = H_{t} + \eta_{t} \sum\limits_{s=0}^{t-1} \frac{c_{s}}{N_{s}}
\]

那么,\(K\) 为一个 pure-investment strategy from the same initial wealth \(x\)。

特殊地,当且仅当 \(K\) 为一个 terminal-consumption arbitrage 时,\(H\) 为一个 arbitrage。


证明:

\[\begin{align*}
(K_{t} - K_{t+1}) \cdot P_{t} & = \Big( H_{t} + \eta_{t}\sum\limits_{s=0}^{t-1}\frac{c_{s}}{N_{s}} - H_{t+1} - \eta_{t+1}\sum\limits_{s=0}^{t}\frac{c_{s}}{N_{s}} \Big) \cdot P_{t} \\
& = (H_{t} - H_{t+1}) \cdot P_{t} + \Big( \eta_{t}\sum\limits_{s=0}^{t-1}\frac{c_{s}}{N_{s}} - \eta_{t+1}\sum\limits_{s=0}^{t}\frac{c_{s}}{N_{s}} \Big) \cdot P_{t} \\
& = (H_{t} - H_{t+1}) \cdot P_{t} + \Big( \eta_{t}\sum\limits_{s=0}^{t}\frac{c_{s}}{N_{s}} - \eta_{t+1}\sum\limits_{s=0}^{t}\frac{c_{s}}{N_{s}} - \eta_{t} \frac{c_{t}}{N_{t}} \Big) \cdot P_{t} \\
& = (H_{t} - H_{t+1}) \cdot P_{t} + \Big( \big( \eta_{t} - \eta_{t+1} \big) \sum\limits_{s=0}^{t} \frac{c_{s}}{N_{s}} - \eta_{t} \frac{c_{t}}{N_{t}} \Big) \cdot P_{t} \\
& = (H_{t} - H_{t+1}) \cdot P_{t} - \eta_{t} \cdot P_{t} \frac{c_{t}}{N_{t}} + \big( \eta_{t} - \eta_{t+1} \big) \cdot P_{t} \sum\limits_{s=0}^{t}\frac{c_{s}}{N_{s}} \\
& = (H_{t} - H_{t+1}) \cdot P_{t} - \eta_{t} \cdot P_{t}\frac{c_{t}}{N_{t}} \qquad \text{(Investment-consumption strategy)} \\
& = c_{t} \cdot P_{t} - N_{t} \cdot \frac{c_{t}}{N_{t}} \qquad \text{(By definition)} \\
& = 0
\end{align*}
\]

因此,对于 \(\forall t \geq 0\),有:

\[(K_{t} - K_{t+1}) \cdot P_{t} = 0
\]

由假设:\((\eta_{t})_{t\geq 0}\) 为 pure-investment strategy,则 \((K_{t})_{t\geq 0}\) 亦为 pure-investment strategy。

假设对于 non-random \(T\),有:\(c_{T} = H_{T}\cdot P_{T}\),那么:

\[\begin{align*}
K_{T} \cdot P_{T} & = \Big( H_{T} + \eta_{T}\sum\limits_{s=0}^{T-1}\frac{c_{s}}{N_{s}} \Big) \cdot P_{T} \\
& = H_{T} \cdot P_{T} + \eta_{T} \cdot P_{T} \sum\limits_{s=0}^{T-1}\frac{c_{s}}{N_{s}} \\
& = c_{T} + N_{T} \sum\limits_{s=0}^{T-1}\frac{c_{s}}{N_{s}} \\
& = N_{T} \frac{c_{T}}{N_{T}} + N_{T} \sum\limits_{s=0}^{T-1}\frac{c_{s}}{N_{s}} \\
& = N_{T} \sum\limits_{s=0}^{T}\frac{c_{s}}{N_{s}} \\
\end{align*}
\]
\[\implies K_{T} \cdot P_{T} = N_{T} \sum\limits_{s=0}^{T}\frac{c_{s}}{N_{s}}
\]

则:当且仅当 某些 \(c_{t} ~ (0 \leq t \leq T)\) 取值为 strictly positive 时, 等式左侧 \(K_{T} \cdot P_{T}\) 为 strictly positive。


Lemma. (Bayes formula; from homework 5.)

令 \(\mathbb{P}\) 和 \(\mathbb{Q}\) 为定义在 \((\Omega, ~ \mathcal{F})\) 上的 equivalent probability measures,令 Radon - Nikodym derivative: \(Z = \frac{d\mathbb{Q}}{d\mathbb{P}}\),令 \(\mathcal{G} \subset \mathcal{F}\) 为一个 \(\sigma-\)field。那么:

\[\mathbb{E}^{\mathbb{Q}}\big[ X ~ \big| ~ \mathcal{G} \big] = \frac{\mathbb{E}^{\mathbb{P}}[ZX ~ | ~ \mathcal{G}]}{\mathbb{E}^{\mathbb{P}}[Z ~ | ~ \mathcal{G}]}
\]

证明:

令 \(Y = \frac{\mathbb{E}^{\mathbb{P}}[ZX ~ | ~ \mathcal{G}]}{\mathbb{E}^{\mathbb{P}}[Z ~ | ~ \mathcal{G}]}\),欲证:\(\mathbb{E}^{\mathbb{Q}}\big[ X ~ \big| ~ \mathcal{G} \big] = Y\),这等价于:

对于 \(\forall G \in \mathcal{G}\):

\[\begin{align*}
& \mathbb{E}^{\mathbb{Q}}\big[ X ~ \big| ~ \mathcal{G} \big] \cdot \mathbb{I}_{G} = Y \cdot \mathbb{I}_{G} \\
\iff \quad & \mathbb{E}^{\mathbb{Q}}\Big[ \mathbb{E}^{\mathbb{Q}}\big[ X ~ \big| ~ \mathcal{G} \big] \cdot \mathbb{I}_{G} \Big] = \mathbb{E}^{\mathbb{Q}} \Big[ Y \cdot \mathbb{I}_{G} \Big] \\
\iff \quad & \mathbb{E}^{\mathbb{Q}}\Big[ \mathbb{E}^{\mathbb{Q}}\big[ X \cdot \mathbb{I}_{G} ~ \big| ~ \mathcal{G} \big] \Big] = \mathbb{E}^{\mathbb{Q}} \Big[ Y \cdot \mathbb{I}_{G} \Big] \\
\iff \quad & \mathbb{E}^{\mathbb{Q}} \big[ X \cdot \mathbb{I}_{G} \big] = \mathbb{E}^{\mathbb{Q}} \Big[ Y \cdot \mathbb{I}_{G} \Big] \\
\iff \quad & \int_{G} ~ X ~ d\mathbb{Q} = \int_{G} ~ Y ~ d\mathbb{Q}
\end{align*}
\]

由 Radon-Nikodym derivative \(Z = \frac{d\mathbb{Q}}{d\mathbb{P}} \implies d\mathbb{Q} = Z \cdot d\mathbb{P}\):

\[\begin{align*}
& \int_{G} ~ X ~ d\mathbb{Q} = \int_{G} ~ Y ~ d\mathbb{Q} \\
\iff \quad & \int_{G} ~ X Z ~ d\mathbb{P} = \int_{G} ~ YZ ~ d\mathbb{P} \\
\iff \quad & \mathbb{E}^{\mathbb{P}}\big[ XZ \cdot \mathbb{I}_{G} \big] = \mathbb{E}^{\mathbb{P}}\big[ YZ \cdot \mathbb{I}_{G} \big]
\end{align*}
\]

因此,目标等价于证明:对于 \(\forall G \in \mathcal{G}\),有:

\[\mathbb{E}^{\mathbb{P}}\big[ XZ \cdot \mathbb{I}_{G} \big] = \mathbb{E}^{\mathbb{P}}\big[ YZ \cdot \mathbb{I}_{G} \big]
\]

注意到 \(Y = \mathbb{E}^{\mathbb{Q}}\big[ X ~ \big| ~ \mathcal{G} \big]\) 为 \(\mathcal{G}-\)measurable,那么RHS:

\[\begin{align*}
\mathbb{E}^{\mathbb{P}}\big[ YZ \cdot \mathbb{I}_{G} \big] & = \mathbb{E}^{\mathbb{P}} \Big[ \mathbb{E}^{\mathbb{P}}\big[ YZ \cdot \mathbb{I}_{G} ~ \big| ~ \mathcal{G} \big] \Big] \qquad \text{(Tower property)} \\
& = \mathbb{E}^{\mathbb{P}} \Big[ \mathbb{I}_{G}Y \cdot \mathbb{E}^{\mathbb{P}}\big[ Z ~ \big| ~ \mathcal{G} \big] \Big] \qquad \text{($\mathbb{I}_{G}Y$ is $\mathcal{G}-$measurable)} \\
& = \mathbb{E}^{\mathbb{P}} \Big[ \mathbb{I}_{G} \cdot \frac{\mathbb{E}^{\mathbb{P}}[ZX ~ | ~ \mathcal{G}]}{\mathbb{E}^{\mathbb{P}}[Z ~ | ~ \mathcal{G}]} \cdot \mathbb{E}^{\mathbb{P}}\big[ Z ~ \big| ~ \mathcal{G} \big] \Big] \\
& = \mathbb{E}^{\mathbb{P}} \Big[ \mathbb{I}_{G} \cdot \mathbb{E}^{\mathbb{P}} \big[ZX ~ \big| ~ \mathcal{G} \big] \Big] \\
& = \mathbb{E}^{\mathbb{P}} \Big[ \mathbb{E}^{\mathbb{P}} \big[ZX \cdot \mathbb{I}_{G} ~ \big| ~ \mathcal{G} \big] \Big] \qquad \text{($\mathbb{I}_{G}$ is $\mathcal{G}-$measurable)} \\
& = \mathbb{E}^{\mathbb{P}} \big[ ZX \cdot \mathbb{I}_{G} \big] \qquad \text{(Tower property)}
\end{align*}
\]

证毕。

Stochastic Methods in Finance (1)的更多相关文章

  1. 自然语言15.1_Part of Speech Tagging 词性标注

    QQ:231469242 欢迎喜欢nltk朋友交流 https://en.wikipedia.org/wiki/Part-of-speech_tagging In corpus linguistics ...

  2. 词性标注 parts of speech tagging

    In corpus linguistics, part-of-speech tagging (POS tagging or POST), also called grammatical tagging ...

  3. Deep Learning中的Large Batch Training相关理论与实践

    背景 [作者:DeepLearningStack,阿里巴巴算法工程师,开源TensorFlow Contributor] 在分布式训练时,提高计算通信占比是提高计算加速比的有效手段,当网络通信优化到一 ...

  4. Introduction To Monte Carlo Methods

    Introduction To Monte Carlo Methods I’m going to keep this tutorial light on math, because the goal ...

  5. Computational Methods in Bayesian Analysis

    Computational Methods in Bayesian Analysis Computational Methods in Bayesian Analysis  [Markov chain ...

  6. Coursera Deep Learning 2 Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization - week2, Assignment(Optimization Methods)

    声明:所有内容来自coursera,作为个人学习笔记记录在这里. 请不要ctrl+c/ctrl+v作业. Optimization Methods Until now, you've always u ...

  7. (转) Ensemble Methods for Deep Learning Neural Networks to Reduce Variance and Improve Performance

    Ensemble Methods for Deep Learning Neural Networks to Reduce Variance and Improve Performance 2018-1 ...

  8. History of Monte Carlo Methods - Part 1

    History of Monte Carlo Methods - Part 1 Some time ago in June 2013 I gave a lab tutorial on Monte Ca ...

  9. Stochastic Optimization Techniques

    Stochastic Optimization Techniques Neural networks are often trained stochastically, i.e. using a me ...

  10. 《Graph Neural Networks: A Review of Methods and Applications》阅读笔记

    本文是对文献 <Graph Neural Networks: A Review of Methods and Applications> 的内容总结,详细内容请参照原文. 引言 大量的学习 ...

随机推荐

  1. MessagePack 和System.Text.Json 序列号 反序列化对比

    本博客将测试MessagePack 和System.Text.Json 序列号 反序列化性能 项目文件: Program.cs代码: using BenchmarkDotNet.Running; us ...

  2. 2022-11-05 Acwing每日一题

    本系列所有题目均为Acwing课的内容,发表博客既是为了学习总结,加深自己的印象,同时也是为了以后回过头来看时,不会感叹虚度光阴罢了,因此如果出现错误,欢迎大家能够指出错误,我会认真改正的.同时也希望 ...

  3. 关于Intent.setDataAndType参数问题

    关于Intent.setDataAndType参数问题 install取设置属于和类型,数据就是获取到的uri,更具文件类型不同,type参数也不相同,具体参考下表 {后缀名,MIME类型} ​ {& ...

  4. mindxdl---common--test_tools.go

    // Copyright (c) 2021. Huawei Technologies Co., Ltd. All rights reserved.// Package common define co ...

  5. std C++11 生成随机数组

    #include <algorithm> #include <array> #include <iostream> #include <iterator> ...

  6. 【DL论文精读笔记】Image Segmentation Using Deep Learning: A Survey 图像分割综述

    深度学习图像分割综述 Image Segmentation Using Deep Learning: A Survey 原文连接:https://arxiv.org/pdf/2001.05566.pd ...

  7. Java网络编程:Socket 通信 2

    client----发送数据(输出流)------------(输入)-[管道流处理数据]-(输出)------接收数据(输入流)------server 文件传输: 客户端: 创建Socket连接对 ...

  8. .NET MAUI 安卓应用开发初体验

    一..NET MAUI开发环境搭建&安卓SDK和安卓模拟器安装提示网络连接失败问题解决 引言 本节目标是帮助第一次搭建.NET MAUI开发环境,在下载安卓SDK和安卓模拟器过程中一直提示网络 ...

  9. JDBC Request 中 Variable names 以及 Result variable name 的使用方法

    1.Variable name 的使用方法 设置好JDBC Connection Configuration.JDBC Request  具体配置百度 如果数据库查询的结果不止一列那就在Variabl ...

  10. 最新 2022 年 Kubernetes 面试题高级面试题及附答案解析

    题1:Kubernetes Service 都有哪些类型? 通过创建Service,可以为一组具有相同功能的容器应用提供一个统一的入口地址,并且将请求负载分发到后端的各个容器应用上.其主要类型有: C ...