Brownian Bridge™

Setup

Market Context

Beyond vanilla calls and puts, the equity derivatives market encompasses a rich taxonomy of exotic payoffs — contracts whose payoff depends on the path of the underlying, on its maximum or minimum, on its realised variance, or on returns measured at multiple fixing dates. These products exist because clients have genuine economic exposures that vanilla options cannot hedge efficiently: a commodity producer wants to hedge average prices (Asian), a fund wants exposure to realised volatility (variance swap), a structured note issuer wants to embed a capital-protection mechanism (cliquet).

Each exotic payoff type has a canonical pricing model, a set of model sensitivities that matter in practice, and a known set of model-dependency risks. Understanding which exotic is sensitive to which model feature — and why — is the core quant skill for equity exotics desks.

INSIGHT

Financial insight. The four exotic payoffs covered here — Asian options, lookbacks, variance/volatility swaps, and cliquets — represent four distinct dimensions of path-dependency: average path (Asian), extreme path (lookback), quadratic variation of path (variance swap), and sequential reset of strikes (cliquet). Each one breaks a different assumption of the Black-Scholes framework. A quant who can derive each payoff's replication strategy or pricing formula, state the model sensitivities, and identify what breaks in each model is interview-ready at the L2/L3 level for any equity exotics role.

Assumptions

Throughout: the underlying $S_t$ follows a continuous Itô diffusion under the risk-neutral measure $\mathbb{Q}$ . Specific assumptions are stated per product.
Interest rate $r$ and dividend yield $q$ are deterministic unless stated.
We price at $t = 0$ with maturity $T$ . Monitoring dates $0 = t_0 < t_1 < \cdots < t_N = T$ are equally spaced: $\Delta = T/N$ .
Currency: all prices are in the domestic currency. FX adjustments for multi-currency exotics are not treated here.

Theory

1. Asian Options

An Asian option has a payoff based on the arithmetic average of the underlying over monitoring dates $t_1, \ldots, t_N$ :

DEFINITION

Definition 6.1 (Asian Option Payoffs).

Asian call (fixed strike): $\left(\bar{S} - K\right)^+$ where $\bar{S} = \frac{1}{N}\sum_{i=1}^N S_{t_i}$
Asian put (fixed strike): $\left(K - \bar{S}\right)^+$
Asian call (floating strike): $\left(S_T - \bar{S}\right)^+$
Asian call (geometric average): $\left(\hat{S} - K\right)^+$ where $\hat{S} = \left(\prod_{i=1}^N S_{t_i}\right)^{1/N}$

Why Asians? Commodity producers and consumers hedge their exposure to average prices over a period, not spot prices at a single date. Energy markets quote swaps referencing monthly average prices. For equity, Asian options appear in employee stock option plans and some structured notes as a cost-reduction mechanism (lower premium than vanilla due to averaging).

Geometric average — closed form. Under Black-Scholes, the geometric average $\hat{S} = \exp\!\left(\frac{1}{N}\sum_i \ln S_{t_i}\right)$ is log-normally distributed, so a geometric Asian call has a Black-Scholes-type closed form. Let $\hat{\sigma}$ and $\hat{\mu}$ be the effective vol and drift of $\ln \hat{S}$ :

THEOREM

Theorem 6.1 (Geometric Asian — Closed Form). Under Black-Scholes with constant $\sigma$ , $r$ , $q$ , the geometric average $\hat{S}$ satisfies $\ln \hat{S} \sim \mathcal{N}(\hat{\mu}, \hat{\sigma}^2)$ with:

$\hat{\sigma}^2 = \frac{\sigma^2}{N^2}\sum_{i=1}^N \sum_{j=1}^N \min(t_i, t_j), \qquad \hat{\mu} = \ln S_0 + \frac{1}{N}\sum_{i=1}^N \left[(r-q)t_i - \frac{\sigma^2 t_i}{2}\right]$

The geometric Asian call price is $e^{-rT}\,\mathbb{E}[(\hat{S} - K)^+]$ , evaluated as a standard Black-Scholes formula with $(\hat{\mu}, \hat{\sigma})$ .

Arithmetic average — no closed form. The arithmetic average is a sum of correlated log-normals, which has no closed-form distribution. Standard approaches: (1) Monte Carlo simulation; (2) moment-matching approximation (Turnbull-Wakeman, 1991): match the first two moments of $\bar{S}$ to a log-normal and apply the geometric Asian formula with adjusted parameters.

REMARK

Remark: Averaging reduces vol. The key intuition for Asian options: averaging over $N$ dates reduces the effective volatility of the payoff. In the continuous limit ( $N \to \infty$ ), the arithmetic average $\frac{1}{T}\int_0^T S_t\,dt$ has variance scaling as $\sigma^2 T / 3$ rather than $\sigma^2 T$ for a single terminal observation. Asian options are therefore cheaper than vanilla options with the same strike — typically 50–70% of the vanilla premium for at-the-money contracts.

2. Lookback Options

A lookback option has a payoff depending on the running maximum or minimum of the underlying over the option's life:

DEFINITION

Definition 6.2 (Lookback Payoffs).

Floating-strike lookback call: $S_T - m_T$ where $m_T = \min_{t \leq T} S_t$
Floating-strike lookback put: $M_T - S_T$ where $M_T = \max_{t \leq T} S_t$
Fixed-strike lookback call: $(M_T - K)^+$
Fixed-strike lookback put: $(K - m_T)^+$

The floating-strike lookback call pays the difference between the final price and the lowest price over the period — the payoff of buying at the bottom and selling at the end. The fixed-strike lookback call pays the excess of the maximum price over the strike.

Closed form under Black-Scholes. Using the joint distribution of $(S_T, M_T)$ or $(S_T, m_T)$ — which follows from the reflection principle — lookback options have closed-form prices. For the floating-strike lookback call ( $m_0 = S_0$ , i.e., monitoring starts now):

THEOREM

Theorem 6.2 (Floating-Strike Lookback Call, Goldman-Sosin-Gatto 1979). Under Black-Scholes with constant $\sigma$ , $r$ , $q$ :

$C_{LB} = S_0 e^{-qT} N(d_1) - S_0 e^{-rT}\left[N(d_2) - \frac{\sigma^2}{2(r-q)} N(-d_1)\right] - S_0 e^{-qT} \frac{\sigma^2}{2(r-q)} e^{(-2(r-q)/\sigma^2)\ln(S_0/S_0)} N(-d_1)$

where $d_1 = \frac{(r - q + \sigma^2/2)T}{\sigma\sqrt{T}}$ and $d_2 = d_1 - \sigma\sqrt{T}$ .

In the special case $r = q$ , the formula requires L'Hôpital's rule as $r - q \to 0$ .

REMARK

Remark: Lookbacks are expensive. A floating-strike lookback call is always worth more than a vanilla ATM call: its payoff dominates (it buys at the cheapest price, sells at the final price). Lookback options are among the most expensive path-dependent exotics. They are also highly sensitive to the vol used: the sensitivity to $\sigma$ is large, and under a smile model, the local vol near the historical minimum is the key pricing driver.

3. Variance Swaps

A variance swap is a forward contract on realised variance. At maturity, it pays:

DEFINITION

Definition 6.3 (Variance Swap). A variance swap with notional $N_{var}$ (in vega notional $N_{vega}$ ) pays at maturity:

$\text{Payoff} = N_{var} \cdot \left(\sigma_{realised}^2 - K_{var}\right)$

where $\sigma_{realised}^2 = \frac{252}{N}\sum_{i=1}^N \left(\ln \frac{S_{t_i}}{S_{t_{i-1}}}\right)^2$ is the annualised realised variance (daily log-returns, squared, annualised by 252), and $K_{var}$ is the variance strike fixed at inception.

The fair variance strike $K_{var}$ is set so the swap has zero initial value.

Model-free replication (Demeterfi-Derman-Kamal-Zou, 1999). The most important result for variance swaps is that the fair variance strike can be replicated — and therefore priced — model-free, using only vanilla option prices:

THEOREM

Theorem 6.3 (Model-Free Variance Strike, DDKZ 1999). Under the assumption of continuous paths (no jumps) and a continuously monitored realised variance:

$K_{var} = \frac{2}{T}\left[\int_0^{F_T} \frac{P(K)}{K^2}\,dK + \int_{F_T}^\infty \frac{C(K)}{K^2}\,dK\right]$

where $C(K)$ and $P(K)$ are undiscounted call and put prices (forward prices), and $F_T = S_0 e^{(r-q)T}$ is the forward.

Derivation sketch. The log-contract $\ln(S_T/F_T)$ can be decomposed as:

$\ln(S_T/F_T) = (S_T/F_T - 1) - \int_0^{F_T} \frac{(K - S_T)^+}{K^2}dK - \int_{F_T}^\infty \frac{(S_T - K)^+}{K^2}dK$

By Itô's lemma, the quadratic variation of $\ln S_t$ equals $\int_0^T \sigma_t^2\,dt$ (the realised variance). Taking risk-neutral expectations and identifying the integral terms with put/call prices gives Theorem 6.3.

INSIGHT

Financial insight. The model-free result means that a variance swap position is equivalent to a static portfolio of vanilla options across all strikes. The weights $1/K^2$ overweight OTM options — particularly OTM puts (where the skew is steepest in equities). In practice, the integral is approximated using the discrete set of quoted strikes. Variance swaps are therefore model-independent pricing instruments, and variance swap desks routinely use the DDKZ formula with the quoted vol surface to mark their books.

Variance swap under Black-Scholes. For constant $\sigma$ , $K_{var} = \sigma^2$ — the fair variance strike equals the square of the Black-Scholes vol. The variance swap then gives zero vega at the ATM level and becomes a pure volatility instrument.

Convexity adjustment — volatility swap. A volatility swap pays $\sigma_{realised} - K_{vol}$ (on $\sigma$ , not $\sigma^2$ ). The fair vol strike is not equal to $\sqrt{K_{var}}$ due to the convexity of $x \mapsto \sqrt{x}$ . By Jensen's inequality:

$K_{vol} = \mathbb{E}[\sigma_{realised}] < \sqrt{\mathbb{E}[\sigma_{realised}^2]} = \sqrt{K_{var}}$

The convexity adjustment (vol swap strike below vol-of-var-strike) depends on the vol of variance, which is a model-dependent quantity. Volatility swaps cannot be replicated model-free; their pricing requires a stochastic vol model.

4. Cliquets and Forward-Starting Options

A forward-starting option is an option whose strike is set at a future date, as a proportion of the then-prevailing spot:

DEFINITION

Definition 6.4 (Forward-Starting Call). A forward-starting call struck at $\alpha S_{t_1}$ and maturing at $T > t_1$ pays:

$\left(S_T - \alpha S_{t_1}\right)^+$

Under Black-Scholes, the forward-starting call prices as a multiple of $S_{t_1}$ — the spot at the reset date — because the option is an ATM-forward call from $t_1$ 's perspective. The price at $t=0$ is $S_0 \cdot e^{-qt_1} \cdot c_{BS}(1, \alpha, r, q, \sigma, T - t_1)$ — it depends only on the forward smile from $t_1$ to $T$ , not on the current smile.

DEFINITION

Definition 6.5 (Cliquet). A cliquet (ratchet option) has $N$ reset periods $[t_{i-1}, t_i]$ and pays:

$\sum_{i=1}^N \text{clamp}\!\left(\frac{S_{t_i} - S_{t_{i-1}}}{S_{t_{i-1}}},\; \ell,\; c\right)$

where $\ell$ is the local floor (e.g., $0\%$ ) and $c$ is the local cap (e.g., $5\%$ ). Each period contributes $\max(\ell, \min(c, \text{return}_i))$ to the total payoff.

The cliquet is a sum of forward-starting options — specifically, capped call spreads resetting at each period. Its price is the sum of forward-starting option prices, which depend exclusively on the forward vol surface — the implied vol for options starting at $t_{i-1}$ and maturing at $t_i$ .

WARNING

Warning: Model dependency of cliquets. As established in the local vol and Heston modules: the forward smile generated by different models is dramatically different. Local vol: forward smile collapses to flat. Heston: forward smile decays as $e^{-\kappa t}$ . Bergomi (2005) models: forward smile parameterised directly. For a 5-year cliquet with annual resets, the pricing can vary by 30–50% between local vol and stochastic vol models. This is not a modelling subtlety — it is a real economic risk. Cliquets are the canonical example of products where model choice is a first-order pricing driver.

5. Digital Options and Gap Risk

A digital (binary) option pays a fixed amount $Q$ if a condition is met at expiry:

DEFINITION

Definition 6.6 (Cash-or-Nothing Digital). The cash-or-nothing digital call pays $Q \cdot \mathbf{1}_{S_T \geq K}$ at maturity. Under Black-Scholes:

$C_{digital} = Q \cdot e^{-rT} N(d_2)$

where $d_2 = \frac{\ln(S_0/K) + (r - q - \frac{1}{2}\sigma^2)T}{\sigma\sqrt{T}}$ .

Smile adjustment for digitals. Under a non-flat smile, the digital price is:

$C_{digital} = Q \cdot e^{-rT}\left[N(d_2) - K \frac{\partial \sigma_{impl}}{\partial K}\sqrt{T}\cdot n(d_2)\right] + O\!\left(\left(\frac{\partial \sigma_{impl}}{\partial K}\right)^2\right)$

where the second term is the skew correction: a negative skew (rising vol as $K$ decreases) increases the probability of the underlying finishing above $K$ (because more of the distribution mass is to the left, and the smile-consistent density is shifted right relative to ATM). For a digital call struck at $K$ in a left-skewed market, the smile-adjusted price is higher than the flat-vol BS price.

Gap risk. A one-touch option pays $Q$ at the first time $S_t$ reaches a barrier $H$ . Under continuous monitoring, the one-touch can be replicated by 2 calls struck at $H$ (Carr-Chou, 1997). Under discrete monitoring, it suffers from gap risk: the underlying can jump over the barrier. More practically, even under continuous-path models, the payout discreteness (exactly $Q$ vs 0) creates large gamma near expiry, making the one-touch one of the hardest products to hedge dynamically.

WARNING

Warning: Digital hedging with call spreads. In practice, digital options are hedged using tight call spreads: long a call at $K - \epsilon$ and short a call at $K + \epsilon$ . The width $\epsilon$ is chosen to balance hedge cost (narrower = more expensive due to spread cost) against gamma risk (wider = less accurate hedge). Typically $\epsilon = 2–5\%$ of $K$ . A common structuring mistake is to use a too-narrow spread, creating enormous gamma at expiry and a binary P&L outcome.

Validation

The companion notebook verifies:

Geometric Asian closed form vs Monte Carlo: confirms the geometric Asian formula against MC paths under Black-Scholes.
Arithmetic vs geometric Asian: verifies the averaging effect by comparing arithmetic Asian, geometric Asian, and vanilla prices across strikes.
Variance swap DDKZ formula: computes the model-free variance strike from a parameterised smile surface and compares to the flat-vol result $\sigma^2$ .
Forward-starting option: prices a forward-starting ATM call and verifies it equals $e^{-qt_1} c_{BS}(1, 1, r, q, \sigma, T-t_1) \cdot S_0$ .

PRACTICE

Before opening the notebook. Consider an arithmetic Asian call with $S_0 = 100$ , $K = 100$ , $r = q = 0$ , $\sigma = 20\%$ , $T = 1$ year, monthly monitoring ( $N = 12$ ).

(a) The geometric Asian call has a closed-form price. What effective vol $\hat{\sigma}$ should you use in the Black-Scholes formula? (Hint: for equally-spaced times $t_i = iT/N$ , $\hat{\sigma}^2 \approx \sigma^2 (1 + 1/N)(2 + 1/N)/6 \approx \sigma^2/3$ as $N \to \infty$ .)

(b) Should the arithmetic Asian call price be above or below the geometric Asian call price? (Hint: AM $\geq$ GM.)

Limitations

WARNING

Asian options: no closed form for arithmetic average. The arithmetic average of log-normals is not log-normal. The Turnbull-Wakeman moment-matching approximation is accurate to within 1–2% for ATM options but can have larger errors in the wings, particularly for long maturities or high vol. Monte Carlo is the benchmark but requires variance reduction (control variate: use the geometric Asian as the control) for efficient estimation.

WARNING

Variance swaps and jumps. The DDKZ model-free formula assumes continuous paths. Under jump-diffusion dynamics, the realised variance includes jump contributions that are not captured by the log-contract replication. In a Merton jump-diffusion model, the fair variance strike is $K_{var} = \sigma_{diff}^2 + \lambda \mathbb{E}[(e^{J} - 1 - J)^2]$ where $\lambda$ is the jump intensity and $J$ the log-jump size. Ignoring jumps underestimates $K_{var}$ — variance swaps are priced too cheaply under pure diffusion models for high-kurtosis underlyings.

WARNING

Cliquet model risk. The cliquet pricing model must be chosen with full awareness of its forward smile implications. Using a local vol model for cliquets is not conservative — it is systematically wrong in a direction that could be $30–50\%$ of the product value. Any desk pricing cliquets must use a model with explicit forward vol control (Bergomi, LSV with a stochastic vol component, or Rough Heston).

WARNING

Lookback monitoring frequency. The lookback closed-form prices continuous monitoring. For daily monitoring, the running maximum at daily closes is systematically lower than the true continuous running maximum. The correction (analogous to BGK for barriers) adjusts the starting minimum: $m_0 \to m_0 \cdot e^{-\beta\sigma\sqrt{\Delta t}}$ . Ignoring this overprices lookback options.

Interview Angle

PRACTICE

L1 — Junior Quant / Quant Developer.

"Why is an Asian option cheaper than a vanilla option with the same strike?" Expected: averaging reduces the variance of the payoff — the effective vol of the averaged quantity is lower than the vol of the terminal spot. For geometric averaging, the effective vol is approximately $\sigma/\sqrt{3}$ in the continuous limit. Lower effective vol → lower option premium.
"What does a variance swap pay, and who would buy it?" Expected: pays the difference between realised variance and the variance strike, notionalised by the vega notional. Buyers: investors who want pure vol exposure without delta risk (volatility traders, macro funds); sellers: structured product desks that are naturally long vega from exotic books and want to offset.
"What is a cliquet and why is it sensitive to model choice?" Expected: sum of forward-starting options, each resetting the strike to the current spot at the start of each period. Model-sensitive because the price depends entirely on the forward smile — the smile implied for options whose life starts in the future. Local vol, Heston, and Bergomi generate very different forward smiles, leading to large price differences for the same cliquet.

PRACTICE

L2 — Senior Quant / Structurer.

"Derive the model-free variance swap strike formula. What assumption is critical for the derivation to work?" Expected: (1) apply Itô's lemma to $\ln S_t$ to get $d\ln S_t = (r-q-\frac{1}{2}\sigma_t^2)dt + \sigma_t dW_t$ ; (2) rearrange to identify $\sigma_t^2\,dt = -2\,d\ln S_t + 2(r-q)dt + 2\frac{dS_t}{S_t}$ ; (3) integrate and take expectations — the $dS_t/S_t$ term gives the risk-neutral drift, the $d\ln S_t$ term is the log-contract; (4) the log-contract is replicated by a static strip of puts and calls with weights $1/K^2$ . Critical assumption: no jumps (continuous paths). With jumps, $\int_0^T \sigma_t^2\,dt \neq -2\ln(S_T/S_0) + 2\int_0^T\frac{dS_t}{S_t}$ because the Itô-Tanaka formula has additional jump terms.
"Why is the fair vol swap strike below the square root of the variance swap strike?" Expected: Jensen's inequality applied to the concave function $x \mapsto \sqrt{x}$ : $\mathbb{E}[\sqrt{V}] \leq \sqrt{\mathbb{E}[V]}$ . The fair vol strike $K_{vol} = \mathbb{E}[\sigma_{realised}]$ is less than $\sqrt{\mathbb{E}[\sigma_{realised}^2]} = \sqrt{K_{var}}$ . The difference is the convexity adjustment, proportional to the vol of variance. This adjustment is model-dependent — it requires knowing the distribution of $\sigma_{realised}$ , not just its mean — which is why volatility swaps cannot be replicated model-free.
"For a digital call struck at $K$ in an equity market with negative skew, is the smile-adjusted price above or below the flat-vol BS price? Derive the correction." Expected: the smile correction to the digital call is $-Ke^{-rT}n(d_2)\frac{\partial\sigma_{impl}}{\partial K}\sqrt{T}$ . With negative skew ( $\partial\sigma/\partial K < 0$ ), this correction is positive — the digital call is worth more under the skewed surface than under flat vol. Intuition: negative skew tilts the risk-neutral density toward lower values of $S_T$ , but the digital is a bet on $S_T \geq K$ — the density at $K$ is concentrated from the right by the skewed vol, increasing the probability of finishing above $K$ relative to ATM.

PRACTICE

L3 — Quant Researcher.

"Under Bergomi's stochastic vol model, how are cliquet prices parameterised, and what is the key model input?" Expected: Bergomi (2005) parameterises the forward variance curve $\xi_t^u = \mathbb{E}^{\mathbb{Q}}[\sigma_u^2 \mid \mathcal{F}_t]$ and specifies its dynamics directly. The cliquet price is then a function of the joint distribution of $(S_{t_1}, S_{t_2}, \ldots, S_{t_N})$ , which is determined by the forward variance curve and its stochastic component. The key model input is the vol of forward variance — how much the implied variance at a future date fluctuates — which controls the width of the forward smile. A calibrated Bergomi model matches both the current vanilla surface and the forward smile dynamics observed in the market.
"What is the variance risk premium, and how do variance swaps reveal it?" Expected: the variance risk premium (VRP) is the difference between the risk-neutral expected variance (the variance swap strike $K_{var}$ ) and the physical-measure expected variance (what realised vol actually averages over the next month). Empirically in equity markets, $K_{var} > \mathbb{E}^P[\sigma_{realised}^2]$ : investors pay a premium to buy variance protection (they are net short variance). This VRP is the profit source for variance swap sellers. It is related to the equity risk premium but is distinct: it arises from the stochastic discount factor's covariance with variance, not with the market return directly. The VRP is studied by constructing time series of variance swap P&L and regressing against realised variance.