Stochastic Processes Unit 10 Review: Martingales

Key Concepts and Definitions

• Martingales are stochastic processes where the expected value of the next observation equals the last observation, given all prior observations
• Filtration $(\mathcal{F}_n)_{n \geq 0}$ represents the information available up to time $n$, with $\mathcal{F}_n \subseteq \mathcal{F}_{n+1}$ for all $n \geq 0$
  • Example: In a coin toss experiment, $\mathcal{F}_n$ could represent the outcomes of the first $n$ tosses
• Adapted processes have the property that $X_n$ is $\mathcal{F}_n$-measurable for all $n \geq 0$, meaning the value of $X_n$ is known given the information in $\mathcal{F}_n$
• Submartingales and supermartingales are related processes where the expected value of the next observation is greater than or equal to (submartingale) or less than or equal to (supermartingale) the last observation
• Doob decomposition theorem states that any submartingale can be decomposed into a martingale and an increasing predictable process
• Optional stopping theorem allows for the extension of martingale properties to random stopping times under certain conditions

Probability Foundations

• Probability space $(\Omega, \mathcal{F}, \mathbb{P})$ consists of a sample space $\Omega$, a $\sigma$-algebra $\mathcal{F}$, and a probability measure $\mathbb{P}$
  • Sample space $\Omega$ is the set of all possible outcomes
  • $\sigma$-algebra $\mathcal{F}$ is a collection of subsets of $\Omega$ that includes $\Omega$ and is closed under complementation and countable unions
  • Probability measure $\mathbb{P}$ assigns probabilities to events in $\mathcal{F}$, with $\mathbb{P}(\Omega) = 1$ and countable additivity property
• Conditional expectation $\mathbb{E}[X \mid \mathcal{G}]$ is the expected value of a random variable $X$ given the information in a sub-$\sigma$-algebra $\mathcal{G}$
  • Properties include linearity, taking out what is known, and tower property
• Stochastic processes are collections of random variables $(X_t)_{t \in T}$ indexed by a set $T$, often representing time
• Markov property states that the future behavior of a process depends only on its current state, not on its past history
  • Example: In a simple random walk, the probability of moving up or down depends only on the current position, not on the path taken to reach that position

Types of Martingales

• Discrete-time martingales are defined for integer time steps $n = 0, 1, 2, \ldots$, with $\mathbb{E}[X_{n+1} \mid \mathcal{F}_n] = X_n$ for all $n \geq 0$
  • Example: Simple random walk $S_n = \sum_{i=1}^n X_i$, where $X_i$ are i.i.d. with $\mathbb{P}(X_i = 1) = \mathbb{P}(X_i = -1) = \frac{1}{2}$
• Continuous-time martingales are defined for real-valued time $t \geq 0$, with $\mathbb{E}[X_t \mid \mathcal{F}_s] = X_s$ for all $0 \leq s \leq t$
  • Example: Brownian motion $(B_t)_{t \geq 0}$ is a continuous-time martingale with respect to its natural filtration
• Square-integrable martingales satisfy $\mathbb{E}[X_n^2] < \infty$ for all $n \geq 0$ (discrete-time) or $\mathbb{E}[X_t^2] < \infty$ for all $t \geq 0$ (continuous-time)
  • Allows for the application of $L^2$ theory and Hilbert space techniques
• Local martingales are processes that are martingales when stopped at a sequence of stopping times converging to infinity
  • Relaxes the integrability condition of martingales while retaining some of their properties
• Martingale transforms involve multiplying a martingale by a predictable process, resulting in a new martingale under certain conditions
  • Used in the construction of stochastic integrals and financial models

Martingale Properties

• Martingales have constant expectation over time, i.e., $\mathbb{E}[X_n] = \mathbb{E}[X_0]$ for all $n \geq 0$ (discrete-time) or $\mathbb{E}[X_t] = \mathbb{E}[X_0]$ for all $t \geq 0$ (continuous-time)
• Martingales are closed under linear combinations, i.e., if $(X_n)_{n \geq 0}$ and $(Y_n)_{n \geq 0}$ are martingales and $a, b \in \mathbb{R}$, then $(aX_n + bY_n)_{n \geq 0}$ is also a martingale
• Martingale convergence theorem states that a bounded martingale converges almost surely and in $L^1$ to a limit random variable
  • Example: In a fair casino betting game, the gambler's fortune is a bounded martingale and converges to a random variable representing the final fortune
• Maximal inequality for submartingales: $\mathbb{P}(\max_{0 \leq k \leq n} X_k \geq \lambda) \leq \frac{\mathbb{E}[X_n^+]}{\lambda}$ for any submartingale $(X_n)_{n \geq 0}$ and $\lambda > 0$
  • Provides a bound on the probability of the maximum value exceeding a certain level
• Martingale central limit theorem: Under certain conditions, a scaled and centered sequence of martingale differences converges in distribution to a standard normal random variable
  • Allows for the application of normal approximation techniques to martingales

Stopping Times and Optional Sampling

• Stopping time $\tau$ is a random variable with values in $\{0, 1, 2, \ldots\} \cup \{\infty\}$ such that the event $\{\tau \leq n\} \in \mathcal{F}_n$ for all $n \geq 0$
  • Intuition: The decision to stop at time $n$ depends only on the information available up to time $n$
• Optional stopping theorem: If $(X_n)_{n \geq 0}$ is a martingale and $\tau$ is a bounded stopping time, then $\mathbb{E}[X_\tau] = \mathbb{E}[X_0]$
  • Allows for the computation of expected values at random stopping times
• Optional sampling theorem: If $(X_n)_{n \geq 0}$ is a martingale and $\sigma \leq \tau$ are bounded stopping times, then $(X_{\min(\sigma, n)})_{n \geq 0}$ and $(X_{\min(\tau, n)})_{n \geq 0}$ are also martingales
  • Generalizes the optional stopping theorem to multiple stopping times
• Localization technique: If $(X_n)_{n \geq 0}$ is a local martingale and $(\tau_k)_{k \geq 1}$ is a sequence of stopping times increasing to infinity, then $(X_{\min(\tau_k, n)})_{n \geq 0}$ is a martingale for each $k$
  • Allows for the application of martingale properties to local martingales by stopping them at appropriate times
• Wald's identity: If $(X_n)_{n \geq 1}$ are i.i.d. random variables with $\mathbb{E}[|X_1|] < \infty$ and $\tau$ is a stopping time with $\mathbb{E}[\tau] < \infty$, then $\mathbb{E}[\sum_{n=1}^\tau X_n] = \mathbb{E}[X_1] \mathbb{E}[\tau]$
  • Relates the expected value of a stopped sum to the expected values of the summands and the stopping time

Applications in Finance

• Efficient market hypothesis states that asset prices fully reflect all available information, implying that price changes are unpredictable and follow a martingale
  • Example: In a perfectly efficient market, stock prices would be martingales, and it would be impossible to consistently outperform the market
• Martingale pricing theory uses martingales to derive fair prices for financial derivatives based on the principle of no arbitrage
  • Example: The Black-Scholes formula for pricing European call options is derived using a martingale approach
• Martingale representation theorem states that any martingale can be represented as a stochastic integral with respect to a Brownian motion, under certain conditions
  • Provides a foundation for the construction of continuous-time financial models
• Girsanov's theorem allows for the change of probability measure while preserving the martingale property, enabling the use of equivalent martingale measures in pricing and hedging
  • Example: In the Black-Scholes model, the risk-neutral measure is an equivalent martingale measure under which discounted stock prices are martingales
• Optimal stopping problems in finance involve finding the best time to make a decision (e.g., exercising an American option) based on the available information
  • Martingale methods, such as the Snell envelope and the free boundary approach, are used to solve these problems
• Martingale methods are used in the analysis of trading strategies, such as pairs trading and statistical arbitrage, to identify potential mispricing and generate profits
  • Example: A pairs trading strategy might involve forming a mean-reverting spread between two stocks and trading based on the spread's deviation from its long-term mean

Common Examples and Problems

• Gambler's ruin problem: A gambler with initial fortune $a$ plays against an opponent with initial fortune $b$, with the probability of winning each round being $p$. The game continues until one player is ruined. The probability of the gambler's ruin and the expected duration of the game can be analyzed using martingale methods.
• Polya's urn model: An urn contains $a$ red balls and $b$ blue balls. At each step, a ball is drawn randomly and replaced along with an additional ball of the same color. The proportion of red balls in the urn after $n$ draws can be shown to converge almost surely to a random variable with a Beta distribution, using martingale convergence theorems.
• De Moivre's martingale: Consider a sequence of i.i.d. random variables $(X_n)_{n \geq 1}$ with $\mathbb{P}(X_n = 1) = p$ and $\mathbb{P}(X_n = -1) = 1-p$. The process $M_n = (\frac{1-p}{p})^{\sum_{i=1}^n X_i}$ is a martingale, which can be used to derive properties of the binomial distribution and prove the weak law of large numbers.
• Azuma-Hoeffding inequality: If $(X_n)_{n \geq 0}$ is a martingale with bounded differences (i.e., $|X_n - X_{n-1}| \leq c_n$ for all $n \geq 1$), then for any $\epsilon > 0$, $\mathbb{P}(|X_n - X_0| \geq \epsilon) \leq 2 \exp(-\frac{\epsilon^2}{2 \sum_{i=1}^n c_i^2})$. This inequality provides concentration bounds for martingales with bounded jumps.
• Martingale betting systems: Various betting systems, such as the d'Alembert system and the Martingale system (doubling the bet after each loss), are based on the idea of increasing or decreasing bets according to the outcome of the previous bet. However, these systems do not change the expected value of the game and cannot overcome the house edge in the long run.
• Branching processes: A branching process models the growth of a population where each individual independently gives rise to a random number of offspring according to a fixed probability distribution. The martingale $Z_n = \frac{X_n}{\mu^n}$, where $X_n$ is the population size at generation $n$ and $\mu$ is the mean number of offspring per individual, can be used to study the long-term behavior of the process.

Advanced Topics and Extensions

• Martingale representation theorem in continuous time: Any square-integrable martingale $(M_t)_{t \geq 0}$ adapted to a Brownian filtration can be represented as $M_t = M_0 + \int_0^t H_s dB_s$, where $(H_t)_{t \geq 0}$ is a predictable process and $(B_t)_{t \geq 0}$ is a Brownian motion
  • Provides a powerful tool for constructing and analyzing continuous-time martingales
• Stochastic integration: The theory of stochastic integrals extends the concept of integration to integrands that are stochastic processes, with the Itô integral being a key example
  • Martingale properties play a crucial role in the development and application of stochastic calculus
• Martingale problem: Given a operator $A$ acting on functions, a solution to the martingale problem for $A$ is a stochastic process $(X_t)_{t \geq 0}$ such that $f(X_t) - \int_0^t Af(X_s) ds$ is a martingale for a suitable class of functions $f$
  • Provides a characterization of stochastic processes in terms of their infinitesimal generators
• Martingale capacity: A set function that assigns a non-negative value to each subset of a probability space, satisfying certain properties related to martingales
  • Used in the study of exceptional sets and the extension of probabilistic results to more general settings
• Martingale optimal transport: An extension of optimal transport theory that incorporates martingale constraints, allowing for the modeling of dynamic transport problems
  • Applications in mathematical finance, such as the construction of model-free bounds on derivative prices
• Martingale duality: The duality between martingales and convex functions, which allows for the characterization of martingales in terms of their associated convex functions and vice versa
  • Used in the study of martingale inequalities and the geometry of Banach spaces
• Martingale Hardy spaces: Function spaces consisting of martingales satisfying certain integrability conditions, analogous to classical Hardy spaces
  • Provide a framework for studying the boundary behavior and convergence properties of martingales
• Nonlinear expectations and $g$-expectation: Generalizations of the classical expectation operator that satisfy nonlinear properties, often related to martingales and backward stochastic differential equations
  • Used in the modeling of risk measures and the study of stochastic control problems under uncertainty