What is a Girsanov's theorem?
- User
What is a Girsanov's theorem?
- Assistant
Girsanov's theorem is a cornerstone of modern probability theory and stochastic calculus, named after the Russian mathematician Igor Girsanov (1933–1964). It provides a powerful tool for changing the probability measure in a way that transforms a given stochastic process (like a Brownian motion) into another process with a desired drift, while preserving its martingale properties. This is essential for applications in mathematical finance (e.g., option pricing under risk-neutral measures), filtering theory, and solving stochastic differential equations (SDEs).
Intuitive Idea
Imagine you have a "drunkard's random walk" modeled by a standard Brownian motion (W_t) under a probability measure (\mathbb{P}). Girsanov's theorem lets you "tilt" the probabilities (via a Radon-Nikodym derivative) to create a new measure (\mathbb{Q}) where the path looks like a Brownian motion with a specific drift (\theta_t), say (\tilde{W}_t = W_t - \int_0^t \theta_s , ds), which becomes a martingale (driftless) under (\mathbb{Q}).
This is like switching from the "real-world" measure to a "risk-neutral" one in finance, where assets grow at the risk-free rate.
Precise Statement (One-Dimensional Case)
Let ((\Omega, \mathcal{F}, {\mathcal{F}t}, \mathbb{P})) be a filtered probability space satisfying the usual conditions. Suppose (W = (W_t){t \geq 0}) is a standard (\mathbb{P})-Brownian motion, and (\theta = (\theta_t)) is an adapted process such that the stochastic exponential [ Z_t = \exp\left( \int_0^t \theta_s , dW_s - \frac{1}{2} \int_0^t \theta_s^2 , ds \right) ] is a (\mathbb{P})-martingale (Novikov's condition ensures this: (\mathbb{E}[\exp(\frac{1}{2} \int_0^T \theta_s^2 , ds)] < \infty)).
Define a new measure (\mathbb{Q}) by (d\mathbb{Q}|_{\mathcal{F}t} = Z_t , d\mathbb{P}|{\mathcal{F}_t}). Then:
- (\tilde{W}_t := W_t - \int_0^t \theta_s , ds) is a standard (\mathbb{Q})-Brownian motion.
- Conversely, (W_t) is a (\mathbb{Q})-Brownian motion with drift (\theta).
This extends to multi-dimensional cases, semimartingales, and more general diffusions.
Multidimensional and General Forms
For a (d)-dimensional Brownian motion and vector-valued (\theta), it's analogous. The theorem also applies to Itô processes (X_t = X_0 + \int \mu_s , ds + \int \sigma_s , dW_s), where under (\mathbb{Q}), the drift changes by (\sigma \theta).
Key Applications
- Black-Scholes Model: Switches to risk-neutral measure for derivative pricing.
- Stochastic Control: Optimal control under different measures.
- Malliavin Calculus: Differentiability of solutions to SDEs.
For proofs and deeper dives, check Karatzas & Shreve's Brownian Motion and Stochastic Calculus or Øksendal's Stochastic Differential Equations. If you have a specific context or example in mind (e.g., finance or proof sketch), let me know! 🔍