Chapter 12 · Itô Calculus — Working Examples

GBM, Vasicek & CIR

Three models. Three real-world problems. The Itô-Doeblin Formula put to work.
From theory to tools

Chapter 11 gave us the Itô-Doeblin Formula — the new chain rule for a random world. Now we apply it to three of the most important models in quantitative finance. Each model solves a different real problem: how stocks grow, how interest rates stay bounded, and how to prevent a model from producing physically impossible results. These are not just textbook exercises. They are the building blocks of pricing engines used in financial institutions worldwide.

Three models and their use cases — GBM for stocks, Vasicek for interest rates, CIR for safe interest rates Three boxes side by side showing GBM with a stock chart, Vasicek with a rubber band, and CIR with a zero-floor bounce Example 12.1 Generalised GBM The stock engine Models NSE stock prices Example 12.2 Vasicek Model The rubber band rate Mean-reverting interest rates Example 12.3 CIR Model The safe rate Rates that never go negative

Three models, three problems, one mathematical engine — the Itô-Doeblin Formula.

Example 12.1 · Generalised GBM — the stock engine

In Chapter 3 we saw GBM with constant drift and volatility. In practice, a stock's expected return and riskiness change over time — as the company grows, market conditions shift, or volatility spikes during a crisis. Generalised GBM allows both α(t) and σ(t) to vary continuously. Think of it as upgrading from a car with a fixed speed to one with a working accelerator and a variable engine noise.

S(t)
The Car
The stock price itself. Moves forward through time, driven by the two forces below.
α(t)
Gas Pedal
The drift — the expected average return. The driver's foot deciding how fast to accelerate.
σ(t)
Engine Noise
The volatility — the random vibration of the engine. Can be loud or quiet depending on market conditions.
S(t) = S(0) · exp{ ∫₀ᵗ σ(u)dW(u) + ∫₀ᵗ (α(u) − ½σ²(u)) du }
The Itô-Doeblin formula automatically inserts the −½σ²(u) correction.
Without it, the expected return would be distorted by the volatility drag.

The most important feature of this formula is the correction term −½σ². It exists because of an asymmetry in how percentage returns compound. A concrete example makes this clear:

Volatility drag — why a 10% gain followed by 10% loss leaves you below your starting price A number line showing a $100 stock gaining 10% to $110, then losing 10% to $99, ending below $100 despite symmetric percentage moves $100 Start $110 +10% ↑ × 1.10 $99 After −10% ↓ × 0.90 −$1 volatility drag! The −½σ² correction accounts for this $1 shortfall exactly

Symmetric percentage moves are not symmetric in dollar terms. The Itô correction −½σ² is the mathematical fix for this compounding asymmetry.

The −½σ² term is not an approximation or a fudge. It is the exact result of applying the Itô-Doeblin formula to the exponential function. When the formula hits the second derivative of exp(x) — which equals exp(x) itself — the ½f''dt term becomes ½σ²dt, which then subtracts from the drift. The math enforces accounting honesty.

Example 12.2 · The Vasicek model — the rubber band interest rate

Interest rates behave very differently from stock prices. A stock can double, triple, or go to zero. An interest rate set by a central bank tends to stay within a corridor — it is pulled back toward a long-run average whenever it strays too far. The Vasicek model captures this mean-reverting behaviour using a single elegant equation.

dR(t) = (α − βR(t)) dt + σ dW(t)
α/β — The target level (long-run average rate the rubber band pulls toward)
β — The strength of the pull (how quickly rates return to the target)
σ dW(t) — The random shock (news, policy surprises, market moves)
Vasicek rubber band analogy — rate above target gets pulled down, rate below target gets pulled up A diagram showing an interest rate path oscillating around a central target level, with arrows showing the rubber band pull toward the mean whenever the rate deviates Target α/β = 5% Rate too high pulled DOWN ↓ Rate too low pulled UP ↑ 8% 5% 2% ⚠ Vasicek flaw: normal distribution allows negative rates (below zero line) A small but non-zero probability — which we observed in Europe and Japan after 2008

The rubber band pulls the rate back toward the target whenever it strays. But the normal distribution's tails extend to −∞, allowing negative rates.

Vasicek's flaw — the possibility of negative interest rates — was long considered a theoretical curiosity. Then the European Central Bank and Bank of Japan introduced negative policy rates after 2008. This shows that model flaws can sometimes turn out to be features in disguise: Vasicek accidentally predicted something economists once thought impossible. The CIR model was designed specifically to prevent this.

Example 12.3 · The CIR model — the safe interest rate

The Cox-Ingersoll-Ross model (CIR) makes one targeted change to Vasicek: it replaces the constant volatility σ with σ√R(t). This single modification makes the model self-correcting at zero. As the interest rate approaches zero, the volatility term shrinks toward zero too — which means the random shocks get weaker just as the rate hits bottom, allowing the positive drift to push it back up. The rate never crosses zero.

dR(t) = (α − βR(t)) dt + σ√R(t) dW(t)
Same rubber-band drift as Vasicek — but the volatility term σ√R(t) depends on the rate itself.
When R(t) → 0: the volatility → 0 too. The wiggle vanishes just before the floor.
CIR vs Vasicek near zero — Vasicek crosses zero but CIR bounces off the floor Two rate paths shown: Vasicek in red dips below zero while CIR in gold approaches zero then bounces back up as volatility weakens R = 0% (Floor) Vasicek — crosses zero! ⚠ CIR — bounces off zero ✓ 0% Volatility → 0 Drift kicks in ↑ 5% 0%

The CIR model (gold) bounces off zero as σ√R → 0 near the floor. Vasicek (red dashed) passes straight through — a mathematical impossibility in practice.

The CIR model's self-correcting mechanism is elegant: the volatility is endogenous — it grows when rates are high and shrinks when rates are low. This means that at low rates, the market is actually calmer, and the positive mean-reversion drift dominates. The model is internally consistent in a way Vasicek is not.

Vasicek vs CIR — the definitive comparison

Both models capture mean reversion. Both are Itô Processes driven by the same rubber-band drift. The single change — from constant σ to σ√R — creates a cascade of different properties. Understanding exactly why they differ is more important than memorising which one to use.

Feature Vasicek CIR
Mean Reversion Yes — rubber band Yes — rubber band
Volatility structure Constant σ — same wiggle at any rate σ√R — wiggle grows with the rate
Distribution of R(t) Normal (Bell Curve) Non-central Chi-squared (skewed)
Negative rates possible? Yes — a model flaw No — guaranteed positive
Mathematical tractability Very clean — closed-form solutions Slightly more complex — Bessel functions
Best used for Quick pricing, regime analysis Bond pricing, option pricing on rates
Theorem 12.1 · The deterministic strategy result

There is one final insight from this chapter that connects all three models to trading strategy. If a trader uses a deterministic strategy — a position plan that is fixed in advance and does not respond to what the market actually does — then the resulting P&L has a very special property.

THEOREM 12.1
Deterministic strategies produce Normal P&L
If Δ(t) is a deterministic function (fixed in advance, not adapted to the path of W), then the Itô Integral I(T) = ∫₀ᵀ Δ(t) dW(t) follows a perfect Normal Distribution with mean zero and variance ∫₀ᵀ Δ²(t) dt.

The moment Δ(t) becomes random — responding to the current price W(t) — the distribution of I(T) is no longer Normal. It distorts. The shape depends on the specific strategy chosen.
Deterministic vs random strategy — Bell Curve output vs distorted distribution Two panels. Left shows a symmetric Normal distribution for P&L from a deterministic strategy. Right shows an asymmetric skewed distribution from a random adaptive strategy. Deterministic Δ(t) Adaptive Δ(t) = f(W(t)) Perfect Bell Curve ✓ Distorted — skewed & curved mean = 0 shape depends on strategy

Fixing your strategy in advance produces a clean Bell Curve in your P&L. Adapting to the market distorts that curve — sometimes in your favour, sometimes not.

This theorem is why passive index investing (a deterministic strategy — same allocation every period) has a mathematically predictable and symmetric risk profile. Active management (an adaptive strategy — adjusting based on market signals) distorts the P&L distribution away from Normal. Whether that distortion is an improvement — positive skew, fatter right tail — depends entirely on the skill of the signal being used.

Try it — simulate and compare all three models

Select a model, set the parameters, and simulate multiple paths. For Vasicek and CIR, watch the mean-reversion rubber band at work. Try pushing the starting rate far from the target and observe the different speeds of convergence. Then lower the rate close to zero — and see how CIR bounces while Vasicek crosses the floor.

10%
20%
5
Model
Mean final
Std dev
Min value

GBM paths grow exponentially with drift. Vasicek/CIR paths oscillate around the target — but only CIR stays above zero.

Chapter 12 — three models, three lessons:

What comes next: We now have all the ingredients — GBM for the stock, a risk-free bank account for the alternative, and the Itô-Doeblin formula as our calculator. The Black-Scholes-Merton equation combines these three into the formula that tells you the one fair price for an option. The formula that changed Wall Street forever.

Next Chapter: The BSM Portfolio →