Itô Isometry

Motivation: why this matters in quant finance

The Itô isometry is the variance ledger for Brownian trading gains. If a hedging error is written as a stochastic integral, the isometry turns its mean-square size into an ordinary time integral of squared exposure.

That is a practical statement. In a model where discounted prices are martingales, a self-financing strategy has gains of the form $\int_0^T H_t\,dW_t$ after the relevant scaling. The expected gain is zero, but the risk is not zero. The Itô isometry says exactly how much second-moment risk the adapted exposure $H$ creates.

Lawler presents the identity first as the variance rule for stochastic integrals. This lesson isolates that rule because it is the technical bridge from stochastic integrals to square-integrable martingales, quadratic variation, and later stochastic differential equations.

The informal idea

The word "isometry" means distance-preserving. Here the distance is the $L^2$ size of a random variable or process. Brownian integration maps an adapted process $A_t$ to a random payoff $\int_0^T A_t\,dW_t$ , and the isometry says its squared $L^2$ size is exactly the squared $L^2$ size of the integrand over time.

For simple strategies, the reason is orthogonality. Brownian increments over disjoint intervals are independent and mean zero. When the square of the sum is expanded, all cross terms disappear in expectation. Only the squared stakes times interval lengths remain.

This is the continuous-time analogue of a fair-game variance calculation: if you bet $Y_i$ before an independent fair increment, the expected squared winnings add interval by interval.

Formal definitions

Let $W_t$ be standard Brownian motion with respect to $\{\mathcal{F}_t\}$ . For a simple adapted process

A_t=Y_j,\qquad t_j\leq t<t_{j+1},

with each $Y_j$ square-integrable and $\mathcal{F}_{t_j}$ -measurable, the Itô integral is

Z_t=\int_0^t A_s\,dW_s.

For these simple processes, Lawler's variance rule gives

\mathbb{E}[Z_t^2]=\int_0^t \mathbb{E}[A_s^2]\,ds.

Equivalently,

\mathbb{E}\left[\left(\int_0^t A_s\,dW_s\right)^2\right] =\mathbb{E}\left[\int_0^t A_s^2\,ds\right].

For bounded adapted continuous processes, this extends by $L^2$ approximation with simple processes. For broader classes, it holds under the square-integrability condition that both sides are finite.

This identity is the Itô isometry.

Key properties

It controls the construction of the integral

The isometry is not only a property after the integral is defined. It is what lets the construction work. If $A^{(n)}$ approximates $A$ in

\mathbb{E}\left[\int_0^t\left(A_s^{(n)}-A_s\right)^2\,ds\right]\to 0,

then the corresponding stochastic integrals are Cauchy in $L^2$ . That makes the $L^2$ limit well defined.

It gives exact second moments

Z_t=\int_0^t A_s\,dW_s,

then

\text{Var}(Z_t)=\mathbb{E}[Z_t^2] =\int_0^t\mathbb{E}[A_s^2]\,ds,

provided

Z_0=0

and the integral is square-integrable. For hedging error, this is the cleanest route from a strategy's exposure process to a risk number.

It rests on adaptedness

The cross terms vanish because the stake is measurable before the future Brownian increment and the future increment has conditional mean zero. If the integrand can look into the future, the orthogonality argument breaks.

It connects to quadratic variation

For the same integral,

\langle Z\rangle_t=\int_0^t A_s^2\,ds.

Thus the isometry can be read as

\mathbb{E}[Z_t^2]=\mathbb{E}[\langle Z\rangle_t].

Lawler also notes the martingale characterization: for square-integrable martingales of this form, $Z_t^2-\langle Z\rangle_t$ is a martingale.

Worked examples

Example 1: deterministic linear integrand

Let $A_s=s$ . Then

\mathbb{E}\left[\left(\int_0^T s\,dW_s\right)^2\right] =\int_0^T s^2\,ds =\frac{T^3}{3}.

The integral itself is random and mean zero. The isometry computes its second moment without finding its distribution first.

Example 2: piecewise-constant trading exposure

Suppose a strategy holds exposure $h_i$ on $[t_i,t_{i+1})$ , with deterministic $h_i$ . Then

\int_0^T A_s\,dW_s =\sum_i h_i\left(W_{t_{i+1}}-W_{t_i}\right),

and

\mathbb{E}\left[\left(\int_0^T A_s\,dW_s\right)^2\right] =\sum_i h_i^2(t_{i+1}-t_i).

This is exactly the discrete variance calculation a trader would expect: squared position times variance of the price shock, summed across intervals.

Example 3: the integral of Brownian motion against itself

For

Z_t=\int_0^t W_s\,dW_s,

the isometry gives

\mathbb{E}[Z_t^2]=\int_0^t\mathbb{E}[W_s^2]\,ds =\int_0^t s\,ds =\frac{t^2}{2}.

Itô's formula separately gives $Z_t=\frac{1}{2}(W_t^2-t)$ . The isometry verifies its second moment while keeping the stochastic-integral construction front and center.

Common confusions and pitfalls

"The isometry is a pathwise identity." It is an $L^2$ identity. It compares expectations of squares, not individual sample paths.

"The stochastic integral equals the time integral of $A^2$ ." The random variable $\int A\,dW$ is not $\int A^2\,dt$ . Only its second moment is controlled by the latter.

"Adaptedness is a technical side condition." Adaptedness is the reason the cross terms vanish. It is the mathematical version of placing the bet before the fair-game increment is observed.

"Mean zero makes the variance formula unnecessary." Mean zero says the strategy is fair. The isometry says how risky the fair strategy is.

"The isometry proves every local stochastic integral is a true martingale." Lawler separates square-integrable martingales from local martingales. The isometry requires the relevant square-integrability to be finite.

Where this goes next

Stochastic Integrals: Defines the integral whose $L^2$ size the isometry measures.
Quadratic Variation: Gives the pathwise variance clock $\langle Z\rangle_t=\int_0^tA_s^2\,ds$ .
Itô's Lemma: Uses stochastic integrals and quadratic variation to transform functions of Brownian paths.
Local Martingales and Semimartingales: Explains what changes when square-integrability is weakened.
Stochastic Differential Equations: Uses the same $L^2$ control to reason about diffusion terms.

References

Lawler, G. F. (2023). Stochastic Calculus: An Introduction with Applications. Ch. 3 §3.2 (Stochastic integral), especially Proposition 3.2.1, Proposition 3.2.3, Proposition 3.2.5, and Theorem 3.2.6.