Concept 1 of 61 min
Brownian motion: variances, covariances, scaling
Volatility over several days
| Daily volatility | 2% |
|---|---|
| Short horizon | 4 days |
| Long horizon | 25 days |
- 4 days: 2% × √4 = 4%
- 25 days: 2% × √25 = 10%
Volatility grows with the square root of time, not with time.
A Brownian bridge
| Variance | t × (1 − t) |
|---|---|
| Early t | 0.2 |
| Middle t | 0.5 |
| Late t | 0.8 |
- At 0.2: 0.2 × 0.8 = 0.16
- At 0.5: 0.52 = 0.25
- At 0.8: 0.8 × 0.2 = 0.16
Pinned at both ends, the bridge is most uncertain in the middle.
Covariance of W₃ and W₇
| Times | 3 and 7 |
|---|
- Increment variance = 7 − 3 = 4
- E[W₃ W₇] = min(3, 7) = 3
Only the shared part, W₃, links the 2, so the covariance is the variance of W₃.
- 2% a day
- t × (1 − t)
So Cov(Ws, Wt) = min(s, t), its quadratic variation over [0, T] is T, and volatility over independent periods scales with the square root of time.
Write the later value as the earlier one plus an independent increment: the covariance is the variance of the shared part. W₄ and W₁ are not independent, so their variances do not add.
The sum of squared increments does not vanish as the steps shrink, unlike for a smooth curve, which is where Itô's extra term comes from. Because variances add over independent days, volatility grows with √t, not t.
Tying the path to 0 at time 1 gives a Brownian bridge, whose variance t(1 − t) is 0 at both ends and peaks at 1/4 in the middle. The time integral ∫₀ᵀ Wt dt has variance T³/3, used for Asian options.
- Adding the variances of Wt and Ws as if they were independent.
- Scaling volatility with time instead of its square root.
- Taking T²/2 for the variance of ∫ W dt, which is the Itô isometry answer for ∫ W dW.
Why does a stock's volatility over a year scale with the square root of time rather than with time?
Why does volatility scale with the square root of time, and when does that rule break down?
A complete answer covers
- Variances of independent returns add, so the standard deviation grows with √t.
- It needs independent, identically distributed returns.
- Autocorrelation breaks it: momentum makes long-horizon volatility larger, mean reversion smaller.
- Volatility clustering and fat tails make it a rough guide over short horizons.
Also in our technical interviews, on camera, with a score.
Concept 2 of 61 min
Itô's lemma and stochastic integrals
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 3 of 61 min
Martingales, stopping times and hitting problems
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 4 of 61 min
Risk-neutral pricing, Girsanov and Black-Scholes
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 5 of 61 min
Mean reversion and jumps
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 6 of 61 min
AdvancedVolatility drag: why more leverage can grow less
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inRecap
- Brownian increments are independent and normal with variance equal to the elapsed time; Cov(Ws, Wt) = min(s, t).
- Itô's lemma adds ½ f'' dt because (dW)² = dt: d(W²) = 2W dW + dt, d(ln S) = (μ − σ²/2) dt + σ dW.
- W, W² − t and eθW − θ²t/2 are martingales; optional stopping turns them into hitting probabilities and exit times.
- Under the risk-neutral measure every asset grows at r; Girsanov changes the drift, never the volatility.
- Black-Scholes contains no μ, and Feynman-Kac lets you price with a PDE (partial differential equation) or with an expectation.
25 questions on Stochastic calculus
Timed, like the real one, drawn from a bank of 33. Every answer explained.