Key Concepts in Stochastic Differential Equations

Why This Matters

Stochastic differential equations sit at the intersection of probability theory, calculus, and real-world modeling. When you're tested on SDEs, you're really being tested on your ability to understand how randomness interacts with continuous change: a stock price fluctuating over time, a particle diffusing through a medium, or an interest rate reverting to its long-term average. The core principles here (drift versus diffusion, measure changes, mean reversion, and martingale properties) show up repeatedly in both theoretical proofs and applied problems.

For each concept, ask yourself: What type of randomness does this capture? What real-world behavior does it model? How does it connect to other SDEs? If you can explain why geometric Brownian motion differs from an Ornstein-Uhlenbeck process, or why Girsanov's theorem matters for option pricing, you're thinking at the level the exam demands.

---

Foundational Building Blocks

These concepts form the mathematical foundation upon which all SDEs are built. Without understanding continuous-time random processes and how to differentiate functions of them, you cannot work with any SDE.

Brownian Motion and Wiener Processes

Brownian motion is the canonical "noise" driving SDEs. Its paths are continuous but nowhere differentiable, which is what makes stochastic calculus necessary in the first place (ordinary calculus can't handle these paths).

• Independent, stationary increments: $W(t) - W(s) \sim N(0, t-s)$. The change over any interval depends only on the interval's length, not its location in time.
• Variance grows linearly with time: $\text{Var}(W(t)) = t$. This is why longer time horizons produce greater uncertainty in diffusion-driven models.
• Starts at zero: $W(0) = 0$ almost surely.

Itô's Formula

Itô's formula is the stochastic chain rule. If you have a smooth function $f(t, X_t)$ where $X_t$ satisfies an SDE, Itô's formula tells you how $f$ evolves. The key difference from ordinary calculus is an extra second-derivative correction term.

For a process $dX_t = \mu \, dt + \sigma \, dW_t$, the formula reads:

$df = \left(\frac{\partial f}{\partial t} + \mu \frac{\partial f}{\partial x} + \frac{1}{2}\sigma^2 \frac{\partial^2 f}{\partial x^2}\right)dt + \sigma \frac{\partial f}{\partial x} \, dW_t$

That $\frac{1}{2}\sigma^2 \frac{\partial^2 f}{\partial x^2}$ term exists because $(dW)^2 = dt$ in the Itô calculus framework. Brownian motion's quadratic variation is non-zero, so second-order terms survive rather than vanishing as they do in ordinary calculus. You'll use this constantly to transform processes, verify solutions, and derive pricing equations.

---

Asset Price Models

These SDEs describe how prices evolve over time, balancing deterministic trends (drift) with random fluctuations (diffusion). The key distinctions are whether prices can go negative and whether they exhibit mean reversion.

Geometric Brownian Motion

The SDE $dS = \mu S \, dt + \sigma S \, dW$ models multiplicative (percentage) changes rather than additive ones. Because both drift and diffusion are proportional to $S$, the process has an exponential structure that keeps prices strictly positive.

• Log-normal distribution: $\ln(S_t)$ follows a normal distribution, so $S_t$ itself is log-normally distributed.
• Constant drift $\mu$ and volatility $\sigma$ are the model's two parameters. The explicit solution is $S_t = S_0 \exp\left((\mu - \frac{1}{2}\sigma^2)t + \sigma W_t\right)$.
• Foundation of Black-Scholes: GBM's tractability makes it the standard assumption for equity prices, though it ignores fat tails and volatility clustering.

Ornstein-Uhlenbeck Process

The SDE $dX = \theta(\mu - X) \, dt + \sigma \, dW$ pulls the process toward a long-term mean $\mu$ with speed $\theta$. When $X$ is above $\mu$, the drift is negative; when below, the drift is positive.

• Can go negative, making it unsuitable for asset prices but ideal for interest rates, spreads, and volatility that fluctuate around equilibrium values.
• Stationary distribution exists: unlike Brownian motion, the variance stabilizes at $\frac{\sigma^2}{2\theta}$ rather than growing indefinitely. The stationary distribution is $N\left(\mu, \frac{\sigma^2}{2\theta}\right)$.
• The Vasicek interest rate model is exactly an O-U process applied to the short rate.

---

Probability and Measure Theory

These concepts address how we describe and transform the probability distributions governing stochastic processes. Mastering them is essential for derivative pricing and risk analysis.

Martingales and the Martingale Representation Theorem

A martingale satisfies $E[X_t \mid \mathcal{F}_s] = X_s$ for $s \leq t$. The best prediction of future values is the current value: there's no systematic drift up or down. Brownian motion itself is a martingale, but a Brownian motion with drift is not.

• Representation theorem: Any square-integrable martingale adapted to the Brownian filtration can be written as $M_t = M_0 + \int_0^t H_s \, dW_s$ for some adapted process $H$.
• Why this matters for hedging: The theorem guarantees that derivative payoffs can be replicated by trading the underlying asset. This is the mathematical backbone of no-arbitrage pricing and the reason complete markets allow unique option prices.

Girsanov Theorem

Girsanov's theorem tells you how to change the probability measure so that a process with drift becomes a martingale (or vice versa). Concretely, if $X_t$ has drift $\mu$ under measure $\mathbb{P}$, you can find a new measure $\mathbb{Q}$ under which the drift is removed (or replaced).

• Risk-neutral pricing relies on this: Under the risk-neutral measure $\mathbb{Q}$, discounted asset prices become martingales, which means $E^{\mathbb{Q}}[\text{discounted payoff}]$ gives you the fair price.
• The Radon-Nikodym derivative $\frac{d\mathbb{Q}}{d\mathbb{P}}$ quantifies how likely events are under the new measure relative to the original. It takes the form of an exponential martingale involving the market price of risk.
• Novikov's condition (or Kazamaki's condition) ensures the measure change is well-defined.

---

Evolution Equations

These equations describe how probability distributions or prices evolve over time, connecting SDEs to partial differential equations.

Fokker-Planck Equation

Given an SDE with drift $\mu(x)$ and diffusion $\sigma(x)$, the Fokker-Planck (or forward Kolmogorov) equation describes how the probability density $p(x,t)$ of the state variable changes:

$\frac{\partial p}{\partial t} = -\frac{\partial}{\partial x}[\mu(x)p] + \frac{1}{2}\frac{\partial^2}{\partial x^2}[\sigma^2(x)p]$

The first term on the right captures how drift transports probability, and the second captures how diffusion spreads it. This equation is essential in statistical mechanics for analyzing equilibrium distributions and relaxation times.

Black-Scholes Equation

The Black-Scholes PDE for the price $V(S,t)$ of a derivative on an underlying following GBM is:

$\frac{\partial V}{\partial t} + \frac{1}{2}\sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} + rS\frac{\partial V}{\partial S} - rV = 0$

It's derived by constructing a riskless portfolio (delta hedging) and applying Itô's formula. Notice the drift $\mu$ of the stock doesn't appear; only the risk-free rate $r$ does. That's a direct consequence of risk-neutral pricing.

• Assumes GBM dynamics, constant volatility, continuous trading, and no transaction costs.
• Closed-form solutions exist for European calls and puts (the famous Black-Scholes formulas). For more complex payoffs, you'll need numerical methods.

---

Computational Methods and Applications

When analytical solutions don't exist, numerical methods become essential. Understanding when and why to simulate is as important as knowing the theory.

Numerical Methods for SDEs (Euler-Maruyama)

The Euler-Maruyama scheme discretizes an SDE into time steps of size $\Delta t$:

$X_{n+1} = X_n + \mu(X_n)\Delta t + \sigma(X_n)\Delta W_n, \quad \text{where } \Delta W_n \sim N(0, \Delta t)$

• Strong convergence order 0.5: Pathwise accuracy improves as $O(\Delta t^{0.5})$. This matters when you care about individual sample paths.
• Weak convergence order 1.0: Distributional accuracy (expectations of functions of $X$) improves as $O(\Delta t)$. This is what matters for pricing.
• Milstein scheme adds a correction term $\frac{1}{2}\sigma(X_n)\sigma'(X_n)[(\Delta W_n)^2 - \Delta t]$ to improve strong convergence to order 1.0. It requires computing $\sigma'$, the derivative of the diffusion coefficient.

Applications in Finance and Physics

The SDE framework unifies a wide range of problems where deterministic dynamics are perturbed by continuous random forcing.

• Finance: Option pricing via Monte Carlo, portfolio optimization, interest rate modeling (Vasicek, CIR), and credit risk analysis.
• Physics: Particle diffusion (Langevin equation), thermal fluctuations, population dynamics, and noise-driven phase transitions.
• Choosing your approach: Use closed-form results (like Black-Scholes) when available for speed and precision. Fall back on Monte Carlo simulation with Euler-Maruyama for path-dependent options, complex payoffs, or non-standard dynamics where no analytical solution exists.

---

Quick Reference Table

---

Self-Check Questions

1. Both geometric Brownian motion and Ornstein-Uhlenbeck are driven by Brownian motion. What fundamental property distinguishes their long-term behavior, and which would you choose to model a mean-reverting interest rate?

2. Explain why Itô's formula includes a second-derivative term that doesn't appear in the ordinary chain rule. What property of Brownian motion causes this?

3. Compare the Fokker-Planck equation and the Black-Scholes equation: what does each one describe, and how do their "directions" in time differ?

4. If you need to price a derivative under the risk-neutral measure, which theorem justifies changing from $\mathbb{P}$ to $\mathbb{Q}$? What happens to the drift of the underlying asset under this transformation?

5. When would you use the Euler-Maruyama method instead of a closed-form solution, and what is its strong convergence order? How does the Milstein scheme improve upon it?