Chapter 17 · Exercises — The Proving Ground

From Theory to Practice

Five exercise clusters that transform everything you have learned into tools you can actually use.
The proving ground

You now have the entire theoretical framework — Itô integrals, the Itô-Doeblin formula, Black-Scholes, correlation, and the Brownian Bridge. The exercises in this chapter are the proving ground: the problems that force you to use these tools under pressure, in the same way a new instrument is tested before it goes on stage. We break them into five clusters, each targeting a different skill.

Five exercise clusters mapped to skills Five coloured boxes showing the five exercise clusters: portfolio mechanics, solving SDEs, Greeks verification, correlation geometry, and the Ito-Stratonovich trap 4.1 & 4.10 Portfolio Mechanics 4.5 & 4.8 Solving SDEs 4.9 Greeks & BSM PDE 4.13–4.17 Correlation & Dimension 4.4 Itô vs Stratonovich

Five clusters — five distinct skills. Master all five and you are ready for Risk-Neutral Pricing.

Cluster 1 · Portfolio mechanics (exercises 4.1 & 4.10)
Ex 4.1
Trading a Martingale produces a Martingale

The problem: Show that if you trade an asset M(t) that is already a Martingale, your capital gains process I(t) = ∫₀ᵗ Δ(u) dM(u) is also a Martingale — no matter what strategy Δ you use.

Step-by-step insight
1
The Itô Integral theorem (Theorem 10.1) states directly: if M(t) is a Martingale, then ∫Δ dM is a Martingale.
2
The intuition is the coin-flip argument from Chapter 10: changing your bet size on a fair game cannot make the game unfair. No strategy converts a zero-expected-return process into a positive-expected-return process.
3
This is the mathematical foundation of no-arbitrage: if prices are Martingales, profitable strategies without risk cannot exist.
𝔼[ I(t) | ℱ(s) ] = I(s) for all s ≤ t ✓
Ex 4.10
Why the self-financing condition ignores dΔ·dS

The problem: The general Itô product rule says d(ΔS) = Δ dS + S dΔ + dΔ dS. But in the BSM self-financing portfolio, we write dX = Δ dS + r(X − ΔS) dt with no dΔ dS term. Why?

The answer: A self-financing portfolio satisfies dX = Δ dS + r(X − ΔS) dt by construction — it means you never inject or withdraw cash. Any rebalancing (changes in Δ) is funded by simultaneously adjusting the cash account. The dΔ dS term does not disappear mathematically — it is absorbed into the accounting: the gain from changing positions is exactly offset by the cost of funding that change. The self-financing condition enforces this balance as a constraint.

The self-financing condition is the bridge between discrete-time portfolio theory (where you explicitly track cash flows) and continuous-time stochastic calculus (where everything happens simultaneously). It is what allows us to write the clean BSM master equation — and it is the constraint that a real delta-hedger must maintain at every rebalancing step.

Cluster 2 · Solving SDEs (exercises 4.5 & 4.8)
Ex 4.5
Solving GBM — the log-linearisation trick

The problem: Given dS = αS dt + σS dW, find the explicit solution for S(t).

The log-linearisation method
1
Let f(S) = log S. Apply Itô-Doeblin: d(log S) = (1/S)dS − ½(1/S²)(dS)²
2
Substitute dS = αS dt + σS dW:
d(log S) = (1/S)(αS dt + σS dW) − ½(1/S²)(σS)² dt
3
Simplify: d(log S) = (α − ½σ²) dt + σ dW — no S remaining! The SDE is now linear.
4
Integrate directly: log S(t) = log S(0) + (α − ½σ²)t + σW(t)
S(t) = S(0) · exp{ (α − ½σ²)t + σW(t) }
The −½σ² is the Itô correction: it appears because applying the chain rule to log S generates a second-derivative term −½(1/S²)·(σS)² dt = −½σ² dt.
Ex 4.8
Solving the Vasicek model — the integrating factor

The problem: Solve dR = (α − βR) dt + σ dW explicitly.

The integrating factor method
1
Rewrite as: dR + βR dt = α dt + σ dW. This looks like a first-order linear ODE — except for the dW term.
2
Multiply through by the integrating factor e^(βt): d(e^(βt) R) = αe^(βt) dt + σe^(βt) dW
3
Integrate both sides from 0 to t:
R(t) = R(0)e^(−βt) + (α/β)(1 − e^(−βt)) + σ∫₀ᵗ e^(−β(t−u)) dW(u)
Three terms: decay of initial rate + pull toward long-run mean α/β + accumulated random shocks. The rubber-band analogy in closed form.
Cluster 3 · Verifying the BSM formula solves the PDE (exercise 4.9)

This is the most important exercise for a practitioner. You take the closed-form BSM formula c = x·N(d₁) − Ke⁻ʳᵀN(d₂) and verify — by computing partial derivatives — that it actually satisfies the PDE: cₜ + rx·cₓ + ½σ²x²·cₓₓ = rc.

Θ
Theta = cₜ
∂c/∂t = −xN'(d₁)σ/2√τ − rKe^(−rτ)N(d₂)
Time decay: negative for option buyers. The surface sinks every day.
Δ
Delta = cₓ
∂c/∂x = N(d₁)
Hedge ratio: shares needed to delta-neutralise. Ranges 0 to 1.
Γ
Gamma = cₓₓ
∂²c/∂x² = N'(d₁) / xσ√τ
Curvature: the cost of hedging. Highest at-the-money.
BSM PDE balance sheet — Theta plus Delta contribution plus Gamma equals r times c Three coloured bars representing Theta (negative, time decay), Delta contribution (positive, price gain), and Gamma (positive, curvature gain) summing to r times c the risk-free return BSM PDE: cₜ + rx·cₓ + ½σ²x²·cₓₓ = r·c cₜ Theta (−) Time decay + rx·cₓ Delta (Δ) Price gain + ½σ²x²·cₓₓ Gamma (Γ) Convexity profit = r·c Risk-free return No arbitrage The Gamma-Theta trade-off: options buyers pay Theta daily to receive Gamma profit from every wiggle

The BSM PDE is a balance sheet equation. Three forces — Theta (−), Delta (+), Gamma (+) — must always sum to the risk-free return r·c.

The verification tells you something profound: the Gamma term ½σ²x²·cₓₓ is the "cost of hedging" — the amount you must earn from the curvature of the option to cover the cost of continuous rebalancing. If Gamma is high (option is near the money), hedging is expensive and must be precisely offset by Theta. This Gamma-Theta balance is why options markets price implied volatility so carefully.

Cluster 4 · Correlation and dimension (exercises 4.13, 4.15, 4.17)
Ex 4.13
Decomposing correlated stocks into independent drivers

The problem: Given two correlated stocks dS₁ = σ₁S₁ dW₃ and dS₂ = σ₂S₂ dW₄ where W₃ and W₄ have correlation ρ, find the independent Brownian Motions W₁ and W₂ that drive them.

The method: Use the Cholesky decomposition of the correlation matrix. Set W₃ = W₁ and W₄ = ρW₁ + √(1−ρ²)W₂. Then:

Why it matters: Every multi-asset simulation starts here. To generate correlated random scenarios on a computer, you generate independent standard normals first, then apply the correlation mix. This decomposition is the engine inside every Monte Carlo simulator that runs multiple correlated assets.

Ex 4.17
Instantaneous correlation: why local constants work globally

The problem: If volatility σ(t) and correlation ρ(t) are changing continuously, show that the cross-variation formula dW₁·dW₂ = ρ(t) dt still holds locally.

The insight: Over an infinitesimal window dt, all smooth functions are effectively constant. Even if σ and ρ change from moment to moment, within each dt they behave as fixed parameters. The Multiplication Table applies with the current instantaneous values. This is why stochastic volatility models (where σ itself is a random process) are still tractable — you use the current σ at each step, not a global constant.

Exercise 4.15 asks you to construct a two-dimensional Brownian Motion from two independent one-dimensional motions. This is the foundation of every multi-factor risk model: the Fama-French three-factor model, the APT, and all modern factor-based strategies decompose asset returns into independent shocks mixed with a correlation structure. The stochastic calculus of Chapter 15 is the mathematical rigour behind what practitioners implement daily.

Cluster 5 · The trap: Itô vs Stratonovich (exercise 4.4)

This exercise is a warning for anyone who comes to finance from physics or engineering. Two different conventions for defining stochastic integrals produce answers that differ by exactly ½T — and in finance, using the wrong convention leads to systematic mispricing.

Itô Convention

Evaluates the integrand at the left endpoint of each interval — the value before the price move. This is the only physically realisable convention for trading: you must set your position before you see the price move.

∫₀ᵀ W dW = ½W²(T) − ½T

The −½T is the Itô correction — the "cost of jaggedness."

Stratonovich Convention

Evaluates the integrand at the midpoint of each interval — an average of beginning and end. Produces ordinary calculus rules, with no Itô correction.

∫₀ᵀ W ∘ dW = ½W²(T)

Cleaner formula — but requires knowledge of the future price within each interval.

Ito vs Stratonovich — endpoint evaluation vs midpoint evaluation in one price interval Two panels each showing one price interval. The Ito panel shows the position being set at the left end, before the price move. The Stratonovich panel shows the position averaging the left and right endpoints — requiring knowledge of the future price. Itô: evaluate at LEFT end Stratonovich: evaluate at MIDPOINT tⱼ (left) tⱼ₊₁ Position = W(tⱼ) — known ✓ midpoint Position = avg(W(tⱼ),W(tⱼ₊₁)) — needs future! ✗

Itô uses what you know (left endpoint). Stratonovich uses what you cannot yet know (midpoint). Only Itô is physically realisable in trading.

The difference between Itô and Stratonovich is exactly ½T for the integral ∫W dW. This is not a rounding error — it is a systematic gap that grows with time. A model that accidentally uses Stratonovich calculus in a finance context will consistently underprice options by an amount proportional to ½σ²T. In a 1-year option with σ = 20%, that is a mispricing of 2% of the notional — substantial by any standard.

Try it — measure the Itô vs Stratonovich gap

Simulate many Brownian paths and compute both the Itô integral ∫W dW and the Stratonovich integral ∫W ∘ dW. Confirm that their average difference is exactly ½T — the Itô correction measured empirically.

500
500
Avg Itô ∫W dW
Avg Strat. ∫W∘dW
Gap (should = ½T)
Predicted ½T

Red histogram = Itô integral values  |  Blue histogram = Stratonovich values  |  Both centred but offset by exactly ½T

The final readiness check

Before moving to Risk-Neutral Pricing, confirm you can answer each of these questions instinctively. They represent the minimum fluency needed for Chapter 5.

✅ Readiness Checklist for Chapter 5
Why does d(log S) have a −½σ² dt term? Because Itô-Doeblin applied to f(S) = log S generates a second-derivative correction −½(1/S²)(σS)² dt = −½σ² dt. This is the volatility drag.
Why is the geometric average always below the arithmetic average? Because the log of the mean exceeds the mean of the logs (Jensen's Inequality on a concave function). The −½σ² term is the exact quantification of this gap.
Why does α not appear in the BSM formula? Because delta hedging eliminates all directional risk. The only surviving uncertainty is volatility σ — which is why sigma, not alpha, determines option price.
What does the self-financing condition mean practically? No cash is injected or withdrawn during rebalancing. Every change in position is funded by offsetting trades within the portfolio itself.
What is the Gamma-Theta trade-off? Options buyers pay Theta (daily time decay) in exchange for Gamma (profit from each market wiggle due to convexity). The BSM PDE says these must exactly balance at the risk-free rate r.

Chapter 17 — five clusters, five skills:

You are ready. The final readiness test is the −½σ² correction: if you can explain why d(log S) has this term, you have the full fluency needed for Risk-Neutral Pricing and Girsanov's Theorem — the magic that transforms the real world into a Martingale world, and makes fair option pricing possible.

← Back to Home