Stochastic Differential Equations

Motivation: why this matters in quant finance

A stochastic differential equation is how a continuous-time model says what happens next. The Black-Scholes stock model, Vasicek short rates, Heston variance, local volatility, and diffusion approximations to high-frequency order flow all have the same skeleton:

dX_t=m(t,X_t)\,dt+\sigma(t,X_t)\,dW_t.

The

dt

term is predictable drift; the

dW_t

term is unpredictable innovation. Without this split, phrases like "hedge the Brownian risk", "change the drift under the risk-neutral measure", and "derive the pricing PDE" are only slogans. With it, Itô's lemma becomes a machine for turning a probabilistic model into calculus.

Lawler introduces diffusions exactly this way: a diffusion is a solution to an SDE, its simulation uses the stochastic Euler rule, and its generator leads directly to partial differential equations. That is the quant-finance payoff. The same equation that simulates price paths also produces the Black-Scholes PDE, the Feynman-Kac formula, and the Kolmogorov equations.

The informal idea

An ordinary differential equation says that the next small change is mostly deterministic:

dy_t=F(t,y_t)\,dt.

An SDE adds a second source of motion: a Brownian shock whose typical size is $\sqrt{dt}$ , not $dt$ . The coefficient $\sigma(t,X_t)$ says how strongly the state responds to that shock. A large $\sigma$ makes paths spread quickly; a state-dependent $\sigma$ makes uncertainty itself depend on where the process currently sits.

The differential notation is convenient, but the actual definition is integral. The equation

dX_t=m(t,X_t)\,dt+\sigma(t,X_t)\,dW_t

means

X_t=X_0+\int_0^t m(s,X_s)\,ds+\int_0^t \sigma(s,X_s)\,dW_s.

The first integral is ordinary time accumulation. The second is a stochastic integral, evaluated with information available before the next Brownian increment arrives. That non-anticipative convention is what makes the model usable for trading strategies.

Formal definitions

A one-dimensional Itô SDE is an equation of the form

dX_t=m(t,X_t)\,dt+\sigma(t,X_t)\,dW_t,\qquad X_0=x_0,

where

W_t

is Brownian motion,

m

is the drift function, and

\sigma

is the diffusion or volatility function. In integral form:

X_t=x_0+\int_0^t m(s,X_s)\,ds+\int_0^t \sigma(s,X_s)\,dW_s.

The process is time-homogeneous if

m

and

\sigma

do not depend explicitly on

t

dX_t=m(X_t)\,dt+\sigma(X_t)\,dW_t.

A strong solution is built on a given Brownian motion and filtration. Roughly, once

W_t

and

X_0

are fixed, the solution path is fixed. This is the solution concept behind simulation and most finance models.

Key properties

Drift and diffusion play different mathematical roles

The drift contributes finite-variation motion. The diffusion contributes quadratic variation:

d\langle X\rangle_t=\sigma^2(t,X_t)\,dt.

This is why volatility, not drift, controls the second-derivative term in pricing PDEs.

Diffusions are Markov under their own state

For a diffusion with coefficients depending only on

(t,X_t)

, the current state contains the relevant information for the future law. Past path details matter only through their effect on

X_t

. In finance, this is what makes state-variable models tractable: a one-factor short-rate model needs the current short rate, not the whole yield-curve history.

The generator is the local drift of test functions

For a time-homogeneous diffusion

dX_t=m(X_t)\,dt+\sigma(X_t)\,dW_t,

the infinitesimal generator applied to a smooth function $f$ is

\mathcal{L}f(x)=m(x)f'(x)+\frac{1}{2}\sigma^2(x)f''(x).

This follows from Itô's lemma and is the bridge from path models to PDEs.

Existence is not automatic

Lipschitz-type conditions on $m$ and $\sigma$ guarantee that Picard iteration converges to a solution, echoing the deterministic ODE proof. If the coefficients are only locally Lipschitz, the solution may exist only until an explosion or boundary hitting time. Models such as square-root diffusions require boundary-specific care.

Worked examples

Example 1: geometric Brownian motion as an SDE

Geometric Brownian motion solves

dS_t=\mu S_t\,dt+\sigma S_t\,dW_t.

Applying Itô's lemma to $\log S_t$ gives

d\log S_t=\left(\mu-\frac{1}{2}\sigma^2\right)dt+\sigma\,dW_t,

S_t=S_0\exp\left(\left(\mu-\frac{1}{2}\sigma^2\right)t+\sigma W_t\right).

Lawler notes that the SDE form is often more useful than the closed-form expression because it generalises to state-dependent models where no closed form exists.

Example 2: Ornstein-Uhlenbeck mean reversion

The Ornstein-Uhlenbeck process satisfies

dX_t=-\kappa X_t\,dt+\sigma\,dW_t,\qquad \kappa>0.

Multiplying by $e^{\kappa t}$ and integrating gives

X_t=X_0e^{-\kappa t}+\sigma\int_0^t e^{-\kappa(t-s)}\,dW_s.

The mean is

\mathbb{E}[X_t]=X_0e^{-\kappa t},

and the variance is

\text{Var}(X_t)=\frac{\sigma^2}{2\kappa}\left(1-e^{-2\kappa t}\right).

This is the mathematical core of the Vasicek short-rate model and many spread-trading mean-reversion models.

Example 3: Euler simulation

Lawler's stochastic Euler rule approximates the SDE over a small step $\Delta t$ by

X_{t+\Delta t}=X_t+m(t,X_t)\Delta t+\sigma(t,X_t)\sqrt{\Delta t}\,Z,

where $Z\sim\mathcal{N}(0,1)$ . The formula is not a heuristic replacement for the SDE; it is a discrete approximation to the integral equation. It keeps the Brownian scale $\sqrt{\Delta t}$ explicit, which is the most common source of simulation bugs.

Example 4: multi-factor SDEs

If $W^1,\ldots,W^d$ are Brownian drivers, a vector process can satisfy

dX_t^k=H_t^k\,dt+\sum_{i=1}^d A_t^{i,k}\,dW_t^i.

The covariation is

d\langle X^j,X^k\rangle_t=\sum_{i=1}^d A_t^{i,j}A_t^{i,k}\,dt.

This is how multi-asset equity baskets, multi-factor interest-rate models, and stochastic-volatility models encode correlation.

Common confusions and pitfalls

" $dX_t/dt$ is a noisy derivative." It is not a classical derivative. Brownian paths are nowhere differentiable, so the SDE is shorthand for an integral equation.

"The drift is the important part because it predicts the next move." For pricing, the diffusion coefficient is often more important. It drives quadratic variation, option convexity, and the second-derivative term in the generator.

"Euler simulation is exact if $\Delta t$ is small." It is exact only for special models or in the limit. For GBM with constant coefficients, the exact log-space update is better than Euler.

"A solution exists because the equation is written down." Coefficient regularity matters. Non-Lipschitz coefficients can produce non-uniqueness, boundary issues, or explosion.

"Every SDE is Markov." The standard diffusion form with coefficients depending on current state is Markov. Path-dependent coefficients or hidden factors can break Markovity unless the state vector is enlarged.

Where this goes next

Geometric Brownian Motion: The canonical stock-price SDE and the main example with a closed-form solution.
Infinitesimal Generators and Kolmogorov Equations: Turns the SDE coefficients into the local PDE operator.
Feynman-Kac Formula: Converts expectations over SDE paths into PDE solutions.
Girsanov's Theorem: Changes the drift of Brownian-driven SDEs while preserving volatility.
Jump-Diffusion Processes: Adds jump terms when continuous diffusion paths cannot capture market gaps.

References

Lawler, G. F. (2023). Stochastic Calculus: An Introduction with Applications. Ch. 3 §3.4 (More versions of Itô's formula), §3.5 (Diffusions), §3.7 (Several Brownian motions).