Chapter 28 · Exotic Options — Lookbacks

The Lookback Option: The Time Machine

An option that always pays — because it lets you look back and pick the best price the market ever offered.
The option that always pays something

A standard call expires worthless if the stock finishes below the strike. A lookback call has no such failure mode. Because its effective "strike" is the highest price ever reached during the contract's life, the terminal payoff is always non-negative. You are always buying at the worst possible historical price — and the option pays your regret.

Vanilla Call

  • Strike K is fixed at inception
  • Payoff = max(S(T) − K, 0)
  • Can expire worthless if S(T) < K
  • Cares only about S(T)
Prob(payoff = 0) > 0

Floating Lookback

  • Strike = Y(T) = max{S(u): 0≤u≤T}
  • Payoff = Y(T) − S(T) ≥ 0 always
  • Always pays something — guaranteed
  • Cares about the entire path history
Payoff = 0 only if S(t) = const.
Lookback option payoff — the gap between the historical peak Y and the final price S(T) A stock price path that rises to a peak then falls back. The peak is labelled Y of T. The final price is labelled S of T. The lookback payoff is the vertical gap between the peak and the final price. Y(T) = peak S(T) = final Payoff Y−S(T) t=0 T → The option pays your regret — how much higher the peak was than the final price

The stock peaks at Y(T) then falls to S(T). The lookback payoff is the gap — always non-negative. It pays exactly the "regret" of not having sold at the peak.

The lookback is the option that every retail investor wishes they had. Imagine buying NIFTY in January and receiving a payoff equal to the January-to-December high minus the December closing price. Because markets are volatile, this payoff is usually substantial. The cost reflects this: lookbacks are typically 2–4× more expensive than vanilla calls on the same underlying.

1 · The singular process Y(t) — flat then jumps

The running maximum Y(t) = max{S(u): 0 ≤ u ≤ t} is unlike any process we have encountered in Chapters 1–27. It is singular: it stays perfectly flat for long periods, then suddenly increases when the stock hits a new all-time high. This unusual behaviour requires a special differential rule and produces a unique boundary condition in the PDE.

The Singular Differential Rule for Y(t)
📉
Flat periods: When S(t) < Y(t), the stock is below its historical peak. Y(t) doesn't change. dY(t) = 0. The running maximum sits frozen at the previous high.
📈
New high moments: When S(t) = Y(t), the stock is at an all-time high. Y(t) can increase. dY(t) = dS(t) at those instants — the maximum tracks the stock exactly when they are equal.
The key constraint: Y(t) only grows on a set of times with zero Lebesgue measure — the "new high" instants. This is what "singular" means. The differential is concentrated on a set of measure zero.
dY(t) ≥ 0  ,  dY(t) = 0 when S(t) < Y(t)  ,  (S(t) − Y(t)) dY(t) = 0
Y of t (running maximum) versus S of t — Y stays flat then jumps to match S at each new high Two lines on the same chart. S of t is a jagged Brownian path moving up and down. Y of t is a step function that stays flat and only rises when S reaches a new all-time high, forming a staircase pattern tracking the upper envelope of S. Y flat (S below peak) New high! Y jumps S(t) Y(t) t = 0 T →

Y(t) (gold staircase) stays perfectly flat when S(t) is below the peak, then jumps to match S(t) at each new all-time high. The flat segments have zero measure — this is the "singular" growth.

The singular nature of Y(t) is what makes the lookback PDE different from the barrier PDE. Y(t) generates no diffusion term — it has zero quadratic variation. The PDE has no fᵧᵧ or fᵧ terms from Y itself. The only influence of Y enters through the boundary condition vᵧ(t,y,y) = 0 — the singular process is "felt" at the boundary, not in the interior.

2 · The three-dimensional PDE — and the singular boundary

The lookback option price v(t, x, y) depends on three variables: time t, current stock price x, and the running maximum y. The PDE in the interior (x < y) is the standard BSM equation. The magic — and mathematical subtlety — is at the boundary x = y, where the singular growth of Y imposes a precise condition.

The Lookback PDE System
vₜ + rx·vₓ + ½σ²x²·vₓₓ = r·v   (for 0 ≤ x ≤ y)
Interior PDE
Standard BSM
Same physics as vanilla options. Holds strictly inside the region x < y.
★ Singular Boundary
vᵧ(t, y, y) = 0
When x = y (stock at its all-time high), the option's sensitivity to the maximum must be zero. This prevents a "jump" in value as new highs are set.
Terminal Condition
v(T, x, y) = y − x
At maturity, collect the peak minus the final price. Always ≥ 0.

The singular boundary condition vᵧ = 0 at x = y has a beautiful economic interpretation: the option's value should not jump discontinuously just because the stock hits a new record high. If it did, you could make an instantaneous profit by being in the option exactly at new-high moments. The condition vᵧ = 0 is the no-free-lunch requirement applied to the running maximum process.

3 · The quant trick — dimension reduction from 3D to 2D

Solving a PDE in three variables (t, x, y) requires a 3D grid — computationally expensive and mathematically complex. The lookback option has a special structure that enables a powerful simplification: linear scaling. If you double both the stock price and the maximum, the option value doubles exactly.

The Linear Scaling Trick
1
Observe the homogeneity: v(t, λx, λy) = λ·v(t, x, y) for any λ > 0. The lookback payoff (y − x) scales linearly — doubling all prices doubles the payoff, so the option value doubles too.
2
Define the ratio: z = x/y — the current stock price as a fraction of its all-time high. By construction, 0 ≤ z ≤ 1. When z = 1 the stock is at its all-time high; when z is small the stock has fallen far below the peak.
z = x / y  ∈  [0, 1]
3
Factor out y: Write v(t, x, y) = y · u(t, z). The three-variable problem collapses into a two-variable problem in u(t, z). The PDE for u is a standard 2D PDE — directly solvable by the methods of Chapters 23–24.
v(t, x, y) = y · u(t, x/y)
4
The new boundary conditions: The singular condition vᵧ = 0 at x = y becomes uᵤ(t, 1) − u(t, 1) = 0 in z-coordinates, which is a standard Robin boundary condition. The terminal condition becomes u(T, z) = 1 − z.

Dimension reduction is one of the most important computational techniques in quantitative finance. It exploits the symmetry of the pricing problem to collapse high-dimensional PDEs into tractable low-dimensional ones. The same technique (identifying the homogeneous scaling) applies to Asian options, exchange options, and any option whose payoff is a homogeneous function of the underlying assets.

4 · The lookback pricing formula

After solving the 2D PDE for u(t, z) via Feynman-Kac and integrating against the joint density of (S(T), Y(T)) from Chapter 27, Shreve derives the closed-form lookback price. It resembles Black-Scholes but with additional terms that capture the path-dependency through the running maximum y.

Floating Strike Lookback Formula (Theorem 28.1)
v(t, x, y) = xN(δ₊(x,y)) − e^(−rτ)y·N(δ₋(x,y))
+ (σ²/2r)·x [ N(δ₊(x,y)) − e^(−rτ)(y/x)^(2r/σ²)·N(δ₋(y²/x, y)) ]
δ₊(x,y) = [log(x/y) + (r + ½σ²)τ] / σ√τ
Like BSM d₁ but using y (the running max) as the effective strike.
δ₋(x,y) = δ₊ − σ√τ
Like BSM d₂ — the risk-neutral exercise probability adjusted for the floating strike.
xN(δ₊) − e^(−rτ)y·N(δ₋)
The "vanilla-like" core — a call with floating strike y instead of fixed K.
(σ²/2r)·x[…] correction
The path-dependency premium. Accounts for the expected future new highs. Grows with σ² — lookbacks are very sensitive to volatility.

The (σ²/2r) prefactor in the correction term reveals the critical insight: the lookback premium is directly proportional to σ². Double the volatility → four times the path-dependency premium. This is why lookback options are so expensive in volatile markets — a high-volatility stock is likely to reach a very high peak and then fall significantly, making the Y(T) − S(T) payoff large. In calm markets, the lookback approaches a vanilla call in price.

5 · Vanilla vs lookback — the full comparison
FeatureVanilla CallFloating Lookback
StrikeFixed K at inceptionFloating Y(T) = historical max
Path dependent?No — only S(T) mattersYes — full path history matters
Can expire worthless?Yes — if S(T) ≤ KNo — Y(T) − S(T) ≥ 0 always
State variablest, S(t) — 2Dt, S(t), Y(t) — 3D (reducible to 2D)
Key boundary conditionv(T,x) = (x−K)⁺vᵧ(t,y,y) = 0 (singular) + v(T,x,y) = y−x
Sensitivity to σLinear in σ (Vega ≈ S·√T·N'(d₁))Quadratic in σ — σ² appears in the correction term
Relative costBaseline2–4× more expensive — the "perfect hindsight" premium
HedgeΔ = N(d₁) sharesΔ = ∂v/∂x (more complex — depends on x and y)
Try it — lookback vs vanilla vs barrier

Simulate GBM paths and compare the three option types side by side. Watch how the lookback option always produces a non-negative payoff regardless of market direction, while the vanilla can expire worthless and the barrier can be killed. Increase volatility and observe the lookback premium grow quadratically.

$100
$100
25%
$135
Vanilla call
Lookback call
Lookback premium
Always positive?

Blue histogram = vanilla payoff distribution (mass at 0 = worthless expiry)  |  Gold histogram = lookback payoff distribution (zero mass at 0 — always pays)  |  Lookback distribution has no spike at zero

Chapter 28 — lookback options in five ideas:

What comes next: The Change of Numeraire technique — a powerful measure-change that simplifies complex payoffs by choosing a different "unit of account" (numeraire). It is the key tool for pricing exchange options, spread options, and Quanto derivatives.