Stochastic Calculus

Stochastic calculus is the branch of mathematics that extends classical calculus to handle functions of random processes. In quantitative finance, these random processes model the unpredictable evolution of asset prices, interest rates, and volatility over time.

Why can't we just use ordinary calculus? Because Brownian motion — the standard model for random price movements — is continuous but nowhere differentiable. The ordinary chain rule, integration, and differential equations break down when applied to such jagged paths. Stochastic calculus provides modified rules that work correctly.

The field was developed primarily by Japanese mathematician Kiyosi Ito in the 1940s-50s. His key contribution — Ito's lemma — is the stochastic version of the chain rule and is arguably the most important formula in quantitative finance. It is the mathematical tool that Black, Scholes, and Merton used to derive their famous option pricing formula.

In ordinary calculus, if f is a function of x and x changes by dx, then df = f'(x) dx (the chain rule). In stochastic calculus, we need a correction term.

If X_t follows the stochastic differential equation:

dX = μ dt + σ dW

and f(X, t) is a twice-differentiable function, then Ito's lemma states:

df = (f_t + μf_X + ½σ²f_XX) dt + σf_X dW

The crucial difference from ordinary calculus is the ½σ²f_XX term — this "Ito correction" arises because (dW)² = dt (the quadratic variation of Brownian motion). In ordinary calculus, (dx)² is negligible and dropped. In stochastic calculus, it cannot be dropped because the Brownian motion increment dW is of order √dt, so (dW)² is of order dt — not negligible.

Key application: If a stock price follows geometric Brownian motion dS = μS dt + σS dW, then using f(S) = ln(S) in Ito's lemma gives:

d(ln S) = (μ - σ²/2) dt + σ dW

This shows that log-returns are normally distributed — the foundation of the Black-Scholes model. The -σ²/2 term (the "Ito correction") is why the geometric mean return is lower than the arithmetic mean return.

A stochastic differential equation (SDE) describes how a random process evolves over time. The general form is:

dX = μ(X, t) dt + σ(X, t) dW

where μ(X, t) is the drift (deterministic trend) and σ(X, t) is the diffusion (randomness). Key SDEs in finance include:

• Geometric Brownian Motion (GBM): dS = μS dt + σS dW. The standard model for stock prices. Solution: S_T = S₀ exp((μ - σ²/2)T + σW_T).
• Ornstein-Uhlenbeck (OU): dX = θ(μ - X) dt + σ dW. Models mean reversion — used for interest rates, volatility, and spreads.
• Cox-Ingersoll-Ross (CIR): dr = κ(θ - r) dt + σ√r dW. Mean-reverting process that stays positive — used for interest rates and the Heston stochastic volatility model.
• Heston model: A system of two SDEs — one for the stock price and one for its variance — capturing the volatility smile.

SDEs are solved analytically when possible (GBM has a closed-form solution) or numerically using Monte Carlo simulation (the Euler-Maruyama scheme).

Stochastic calculus is the mathematical engine behind all modern derivatives pricing:

• Deriving the Black-Scholes PDE: Apply Ito's lemma to an option's price V(S, t) where S follows GBM. Construct a delta-hedged portfolio (long option, short Δ shares). Since the portfolio is risk-free, it must earn the risk-free rate. This gives the Black-Scholes PDE: V_t + ½σ²S²V_SS + rSV_S - rV = 0.
• Girsanov's theorem: Allows changing from the real-world probability measure to the risk-neutral measure. Under the risk-neutral measure, the stock drift changes from μ to r (the risk-free rate), which is why option prices don't depend on the stock's expected return.
• Martingale representation theorem: States that in a complete market, every derivative can be replicated by trading the underlying. This justifies risk-neutral pricing.
• Feynman-Kac theorem: Connects SDEs to PDEs — the solution to certain PDEs can be expressed as an expectation of a functional of a stochastic process. This links the PDE approach and the risk-neutral expectation approach to derivatives pricing.