Brownian Bridge™

Setup

Market Context

Barrier options are the most common path-dependent exotic in equity and FX derivatives markets. A barrier option is a vanilla option that is either activated (knock-in) or extinguished (knock-out) when the underlying crosses a prescribed barrier level $H$ during the option's life. Because the barrier introduces a path-dependency — the option's existence depends on whether $S_t$ has ever reached $H$ — their pricing requires tracking the full distribution of the running maximum or minimum of the process, not just the terminal distribution.

Barriers are everywhere on structured product desks: capital-protected notes embed down-and-out puts, autocallable structures embed up-and-out calls, reverse convertibles embed down-and-in puts. Understanding barrier pricing — including the analytic formula under Black-Scholes, the Greeks and their discontinuities near the barrier, and the smile adjustment — is non-negotiable for any equity exotics quant.

INSIGHT

Financial insight. The single most important practical issue with barrier options is the barrier Greek instability: delta, gamma, and vega become extremely large and oscillatory as spot approaches the barrier near expiry. On a live desk, managing a barrier option in the last few days before expiry near a barrier is one of the most challenging hedging situations. The model choice — Black-Scholes vs local vol vs stochastic vol — materially affects the barrier price, and the smile adjustment can easily move the price by 10–30% of the Black-Scholes value.

Assumptions

The underlying follows Black-Scholes GBM: $dS_t = (r - q)S_t\,dt + \sigma S_t\,dW_t$ , with constant volatility $\sigma$ , risk-free rate $r$ , and dividend yield $q$ .
The barrier $H$ is monitored continuously. Discrete monitoring (daily barrier fixes) requires a separate correction (Broadie-Glasserman-Kou, 1997), which shifts the effective barrier by a factor $e^{\pm 0.5826\,\sigma\sqrt{\Delta t}}$ .
No jumps. A jump process that crosses the barrier without touching it would require an entirely different framework.
The barrier is single: one level $H$ , either above (up-barrier) or below (down-barrier) the initial spot. Double barriers exist but are not treated here.
European exercise: the option can only be exercised at expiry $T$ . American-barrier and Bermudan-barrier options require finite-difference methods.
Rebate is zero unless stated otherwise. Some barrier contracts pay a fixed rebate $R$ when the barrier is hit; this adds a separate term to the pricing formula.

Theory

1. Barrier Option Taxonomy

DEFINITION

Definition 5.1 (Barrier Option Types). Let $M_T = \max_{0 \leq t \leq T} S_t$ and $m_T = \min_{0 \leq t \leq T} S_t$ .

Type	Activation condition	Payoff at $T$ (if alive)
Up-and-Out Call (UOC)	Extinguished if $M_T \geq H$	$(S_T - K)^+ \cdot \mathbf{1}_{M_T < H}$
Up-and-In Call (UIC)	Activated only if $M_T \geq H$	$(S_T - K)^+ \cdot \mathbf{1}_{M_T \geq H}$
Down-and-Out Put (DOP)	Extinguished if $m_T \leq H$	$(K - S_T)^+ \cdot \mathbf{1}_{m_T > H}$
Down-and-In Put (DIP)	Activated only if $m_T \leq H$	$(K - S_T)^+ \cdot \mathbf{1}_{m_T \leq H}$
Down-and-Out Call (DOC)	Extinguished if $m_T \leq H$	$(S_T - K)^+ \cdot \mathbf{1}_{m_T > H}$
Down-and-In Call (DIC)	Activated only if $m_T \leq H$	$(S_T - K)^+ \cdot \mathbf{1}_{m_T \leq H}$

Knock-in / Knock-out parity. For any barrier level $H$ and vanilla option $V$ :

THEOREM

Theorem 5.1 (Knock-In / Knock-Out Parity). For the same strike $K$ , barrier $H$ , and maturity $T$ :

$\text{Knock-In} + \text{Knock-Out} = \text{Vanilla}$

Proof. At expiry, exactly one of $\{M_T \geq H\}$ or $\{M_T < H\}$ occurs. The two payoffs partition the vanilla payoff, so their sum equals the vanilla payoff in every scenario. Discounting preserves the equality. $\square$

This parity is a static replication identity — no model assumptions. It allows pricing the knock-in from the knock-out (or vice versa) once the vanilla is known. In practice, knock-outs are typically the more liquid leg (structured products embed DOCs and DOPs), so knock-ins are derived via parity.

2. The Black-Scholes Barrier Formula

The closed-form formula under Black-Scholes uses the reflection principle for Brownian motion: the probability that a Brownian path hits a level $b$ before time $T$ can be computed by reflecting the portion of paths that do cross, using the strong Markov property.

Define the standard Black-Scholes building blocks for an option with spot $S$ , forward $F = Se^{(r-q)T}$ , strike $K$ , vol $\sigma$ , maturity $T$ :

$d_1(S, K) = \frac{\ln(S/K) + (r - q + \tfrac{1}{2}\sigma^2)T}{\sigma\sqrt{T}}, \qquad d_2 = d_1 - \sigma\sqrt{T}$

$c_{BS}(S, K) = Se^{-qT}N(d_1) - Ke^{-rT}N(d_2)$

$p_{BS}(S, K) = Ke^{-rT}N(-d_2) - Se^{-qT}N(-d_1)$

For barrier options, the key auxiliary quantity is the reflection parameter:

$\mu = \frac{r - q - \frac{1}{2}\sigma^2}{\sigma^2}, \qquad \lambda = \sqrt{\mu^2 + \frac{2r}{\sigma^2}}$

In the zero-dividend, zero-rate special case: $\mu = -\frac{1}{2}$ , and the reflection reduces to simply replacing $S$ by $H^2/S$ in the vanilla formula.

THEOREM

Theorem 5.2 (Black-Scholes Down-and-Out Call, $H \leq K$ ). For a DOC with $H \leq \min(S_0, K)$ :

$C_{DOC}(S_0, K, H) = c_{BS}(S_0, K) - \left(\frac{H}{S_0}\right)^{2(\mu+1)} c_{BS}\!\left(\frac{H^2}{S_0}, K\right)$

where $\mu = (r - q)/\sigma^2 - \frac{1}{2}$ .

The second term is the image term: it prices the contribution from reflected paths that would have crossed the barrier, which must be subtracted to enforce the knock-out condition.

For a DOC with $H > K$ (the barrier is above the strike — the option can knock out even while being in-the-money at the barrier), the formula includes additional terms. The full general case for all eight barrier/call/put combinations is given by Merton (1973) and Reiner-Rubinstein (1991); the structure is always:

$\text{Barrier option} = \text{Vanilla terms} - \text{Image terms}$

where image terms replace $S_0$ by $H^2/S_0$ and multiply by the factor $(H/S_0)^{2\mu+1}$ .

REMARK

Remark. In the risk-neutral measure with $r = q = 0$ and $\sigma$ constant, $\mu = -\frac{1}{2}$ , and $(H/S_0)^{2\mu+1} = H/S_0$ . The image term simplifies to $\frac{H}{S_0} c_{BS}(H^2/S_0, K)$ . This special case is the easiest to remember for interview purposes: the DOC equals the vanilla call minus $\frac{H}{S_0}$ times a call struck at $K$ with spot $H^2/S_0$ .

3. Greeks Near the Barrier

The most important practical property of barrier options is that their Greeks become large and discontinuous as spot approaches the barrier near expiry. Understanding this qualitatively is critical for risk management.

Delta and Gamma. At expiry, the DOC has value $(S_T - K)^+$ if $S_T > H$ (and the barrier was not hit), and zero if $S_T < H$ or the barrier was hit. The delta profile therefore has a large negative spike just above the barrier near expiry: if spot approaches $H$ from above, the option is about to knock out to zero value, so delta is large and negative (the option becomes worthless). The gamma (second derivative) has a corresponding spike.

WARNING

Warning: Barrier delta discontinuity. For a DOC with $H < K$ , as spot approaches $H$ from above near expiry, the option value falls rapidly toward zero (the option is about to become worthless). Delta becomes large and negative. At the barrier itself, the option is worth zero by definition. The transition from positive delta (call-like) to near-zero value through large negative delta creates an enormous hedging demand: to delta-hedge, the desk must hold a large short position in the underlying near the barrier, which is impossible to unwind quickly if spot moves suddenly. This is called the barrier delta hedge problem and is why barrier options near expiry near the barrier are the most dangerous contracts on a structured products book.

Vega near the barrier. The vega of a knock-out option changes sign near the barrier: far from the barrier, vega is positive (as for a vanilla call); near the barrier, vega becomes negative — higher vol increases the probability of hitting the barrier and knocking out, which reduces the option value. This sign change creates a hedging complexity: the vega hedge must be reversed as spot approaches the barrier.

4. The Smile Adjustment

The Black-Scholes barrier formula assumes constant volatility. In reality, the vol surface is not flat, and the choice of which volatility to use materially affects the barrier price. The key issue: a barrier option's value depends on the probability of $S_t$ hitting $H$ during $[0, T]$ , which is sensitive to the entire distribution of paths — not just the terminal distribution.

Two standard approaches to incorporate the smile:

Static replication (Carr-Ellis-Gupta, 1998). A knock-out barrier can be statically replicated by a portfolio of vanilla options struck along the barrier, with weights determined by the boundary condition that the replication portfolio equals zero when $S = H$ for all $t \in [0, T]$ . This portfolio is model-free up to the vanilla prices used and is the theoretically cleanest approach for liquid barriers.

Local vol / Stochastic vol pricing. Calibrate a local vol (Dupire) or stochastic vol (Heston) model to the full vanilla surface, then price the barrier option by Monte Carlo or PDE under that model. The price difference between Black-Scholes (with ATM vol) and local vol is the smile correction — typically large for barriers near the money or near the barrier.

INSIGHT

Financial insight. For a down-and-out call with the barrier below the initial spot ( $H < S_0$ ), the smile correction depends crucially on the skew: a negative skew (OTM puts more expensive) means higher implied vol at lower strikes — near the barrier. Local vol, which reproduces the skew, generates higher local vol near the barrier, increasing the probability of touching the barrier and therefore reducing the DOC price relative to Black-Scholes with ATM vol. The smile correction for DOCs is therefore typically negative (the product is cheaper under smile-consistent pricing than BS). For up-and-out calls, where the barrier is above the initial spot, higher implied vol at higher strikes increases the probability of barrier crossing, again reducing the UOC value.

5. Finite Difference Pricing

For non-constant volatility (local vol, time-dependent vol) or American exercise, the barrier is incorporated into the finite-difference PDE as a Dirichlet boundary condition:

$V(H, t) = 0 \quad \text{for all } t \in [0, T] \quad \text{(knock-out)}$

The PDE grid is solved with this condition enforced at the barrier grid node nearest to $H$ . The accuracy depends on the grid spacing: a coarser grid near the barrier introduces larger errors. Standard practice is to use a non-uniform grid that concentrates nodes near both the barrier and the strike.

WARNING

Warning: Discrete barrier monitoring. The standard Black-Scholes formula assumes the barrier is monitored continuously. Exchange-traded barrier options typically fix the barrier daily (close of business). The Broadie-Glasserman-Kou (1997) continuity correction shifts the barrier by a factor of $e^{\pm\beta\sigma\sqrt{\Delta t}}$ where $\beta \approx 0.5826$ (the absolute value of the negative zero of the Riemann zeta function) and $\Delta t$ is the monitoring interval. For weekly monitoring ( $\Delta t = 1/52$ , $\sigma = 0.20$ ): the barrier shift is $e^{0.5826 \times 0.20 \times \sqrt{1/52}} \approx 1.016$ — a 1.6% upward shift for a down-barrier. Ignoring this correction systematically misprices discretely-monitored barriers.

Validation

The companion notebook verifies:

Knock-in/out parity: confirm numerically that DOC + DIC = vanilla call for various parameter combinations.
BS formula round-trip: compute DOC prices via the Black-Scholes formula and verify against Monte Carlo with continuous barrier monitoring.
Greeks near the barrier: compute the delta profile of a DOC as a function of spot and show the large negative delta spike as $S \to H^+$ near expiry.
Smile sensitivity: compare DOC prices under flat vol versus a skewed vol surface (via local vol) and quantify the smile correction.

PRACTICE

Before opening the notebook. Consider a DOC with $S_0 = 100$ , $K = 100$ , $H = 80$ , $T = 1$ , $r = q = 0$ , $\sigma = 20\%$ .

(a) Using the simplified formula for $r = q = 0$ ( $\mu = -\frac{1}{2}$ ): the DOC = $c_{BS}(100, 100) - \frac{80}{100} c_{BS}(6400/100, 100)$ . The second call has spot $64$ , strike $100$ . Is this second term large or small? (Hint: $64/100 = 0.64$ , deep OTM call.)

(b) Using knock-in/out parity: what is $C_{DIC}(100, 100, 80, 1)$ ?

Limitations

WARNING

Flat vol assumption. The Black-Scholes barrier formula assumes constant vol. The smile correction for barrier options is large — often 10–30% of the BS value — because the barrier price is sensitive to the vol near and at the barrier level, which is typically away from ATM (where the vol surface is not flat). Always use local vol or stochastic vol for production pricing.

WARNING

Continuous monitoring. The formula prices continuous barriers. Most exchange-listed and OTC barrier options have discrete (daily) monitoring. The BGK continuity correction works well for barriers not too close to the current spot; near-the-money barriers with weekly or less frequent monitoring may require direct Monte Carlo with discrete monitoring.

WARNING

Near-barrier vanna/volga risk. Near the barrier, the option's sensitivity to changes in vol and spot simultaneously (vanna: $\partial^2 V / \partial S \partial \sigma$ ) and to changes in vol-squared (volga: $\partial^2 V / \partial \sigma^2$ ) become large. These second-order vol sensitivities are not captured by a first-order vega hedge alone. Barrier desks explicitly monitor vanna and volga P&L explain.

WARNING

Jump risk. The barrier formula assumes continuous paths. In the presence of jumps, the underlying can skip over the barrier — the barrier condition is hit without the option being knocked out. Jump-diffusion models (Merton 1976, Bates 1996) require modified pricing formulas that account for the possibility of the underlying jumping past $H$ . Ignoring jumps underestimates the probability of barrier breach for equity indices with fat-tailed return distributions.

Interview Angle

PRACTICE

L1 — Junior Quant / Quant Developer.

"What is knock-in/knock-out parity, and does it depend on the model?" Expected: KI + KO = Vanilla. This is model-free — it follows from the partition of paths and discounting, with no assumption on the dynamics of $S_t$ .
"For a down-and-out call, what happens to delta as spot approaches the barrier from above, very close to expiry?" Expected: delta becomes large and negative. The option is about to knock out (value → 0), so a small upward move in spot increases the option value (positive delta far from barrier), but as spot approaches the barrier, the dominant effect is that the option is about to become worthless, driving delta strongly negative. This large delta creates a hedging problem.
"Does increasing volatility always increase the price of a barrier option?" Expected: No. For knock-out options (e.g., DOC), higher vol increases the probability of hitting the barrier and being knocked out. For options with barrier near the initial spot, this can dominate and reduce the value. Vega can be negative for knock-out options near the barrier.

PRACTICE

L2 — Senior Quant / Structurer.

"Derive the down-and-out call formula from the reflection principle. State the key probabilistic result you need." Expected: under log-normal dynamics, $\ln(S_t/S_0)$ is a Brownian motion with drift $\mu\sigma$ and diffusion $\sigma$ . The reflection principle for Brownian motion with drift states: $\mathbb{P}(\min_{t\leq T} W_t \leq -b, W_T > x) = \mathbb{P}(W_T > x + 2b) \cdot e^{-2\mu b}$ (for $b, x > 0$ ). Applying this to $\ln(S_T/S_0)$ with $b = \ln(S_0/H)$ gives the image term in the formula.
"What is the smile correction for a DOC with H = 80, S₀ = 100, under a negatively-skewed vol surface? In which direction does the price move?" Expected: negative skew → higher implied vol at $K = 80$ (barrier level) than at ATM. Local vol near the barrier is higher, increasing the probability of barrier breach. The DOC is cheaper under a skewed surface than under flat BS with ATM vol. Quantitatively, the correction depends on the skew slope; typical values are 5–20% of the BS price.
"A structured product desk books a DOC with H very close to S₀ (e.g., S₀ = 100, H = 98, T = 1w). What risk management issues arise?" Expected: extreme delta and gamma near the barrier — the option can swing from near-vanilla value to zero in minutes. Vega sign changes near $H$ . The discretely-monitored correction (BGK) is large relative to the option premium for near-the-money barriers. Intraday monitoring may be required. The desk would typically require wide bid-offer spreads or decline to trade such a product.

PRACTICE

L3 — Quant Researcher.

"Explain the Carr-Ellis-Gupta (1998) static replication of a barrier option. What is the theoretical advantage over dynamic hedging, and what is the practical limitation?" Expected: a knock-out barrier option can be exactly replicated by a static portfolio of vanillas struck at the barrier $H$ across all maturities $t \in [0, T]$ , with weights chosen so the portfolio value is zero when $S = H$ for all $t$ . In practice, a finite set of maturities is used. The theoretical advantage: no model assumptions beyond the vanilla prices used; no dynamic rehedging required. Practical limitation: the required strikes and maturities may not be liquid; the portfolio must be entered at inception and cannot be adjusted.
"How does the Broadie-Glasserman-Kou continuity correction for discrete barrier monitoring work, and what is its derivation?" Expected: BGK (1997) showed that a discretely-monitored barrier (monitoring interval $\Delta t$ ) is approximately equivalent to a continuously-monitored barrier with the level shifted by $e^{\beta\sigma\sqrt{\Delta t}}$ where $\beta = -\zeta(1/2)/\sqrt{2\pi} \approx 0.5826$ and $\zeta$ is the Riemann zeta function. The derivation uses the Euler-Maclaurin formula to evaluate the discrete maximum distribution and match it to the continuous distribution with a shifted barrier. The shift is upward for down-barriers, downward for up-barriers.
"In a local vol model, what is the qualitative effect on a DOC price when the local vol near the barrier is higher than ATM? How would you compute the exact smile correction?" Expected: higher local vol near $H$ → higher probability of touching $H$ → lower DOC price. The exact correction: calibrate Dupire's local vol surface to the vanilla quotes; price the DOC by Monte Carlo under the local vol SDE (or by finite differences with the local vol PDE and Dirichlet boundary condition at $H$ ); the smile correction is the difference between this price and the BS price with ATM vol. For a desk with a calibrated local vol surface, this is a daily computation; the smile correction is typically provided as a percentage of the BS price for each barrier level.