The Equity Volatility Smile

Setup

Market Context

After the 1987 crash, equity options markets exhibited a persistent phenomenon that Black-Scholes cannot explain: out-of-the-money puts trade at a higher implied volatility than at-the-money options, and OTM calls trade at lower implied vol. This is the equity volatility skew — a systematic distortion of the Black-Scholes flat-vol world. The skew encodes the market's fear of large downward moves. Understanding it, quoting it correctly (in delta space vs. strike space), and decomposing it into its market drivers is fundamental knowledge for every equity derivatives practitioner.

The vol surface is not a model. It is a compact representation of market prices — every option price on the board, condensed into a single function $\sigma_{impl}(K, T)$. Pricing a new contract requires reading the surface correctly, interpolating without introducing arbitrage, and understanding what the surface implies about the risk-neutral distribution of the underlying.

Assumptions

• Implied vol $\sigma_{impl}(K, T)$ is defined as the Black-Scholes vol that reproduces the market price $C_{mkt}(K, T)$ for each $(K, T)$. It is a price quoting convention, not a model parameter.
• The vol surface is assumed arbitrage-free: no calendar spread arbitrage, no butterfly arbitrage. These are necessary conditions for the surface to be consistent with any rational pricing model.
• We work in a single-currency, single-underlying setting with deterministic interest rates $r$ and dividend yield $q$.
• Options are European. American early exercise complicates the implied vol extraction and is not treated here.
• Liquidity is assumed sufficient that quoted prices are tradeable. In practice, deep OTM strikes have wide bid-offer spreads; the implied vol of illiquid options is a modelling artefact, not a market observable.

Theory

1. The Implied Vol Surface: Definition and Market Conventions

For each strike $K$ and maturity $T$, the implied vol $\sigma_{impl}(K,T)$ solves:

$C_{BS}(S_0, K, r, q, \sigma_{impl}(K,T), T) = C_{mkt}(K,T)$

where $C_{BS}$ is the Black-Scholes call formula. The surface $\sigma_{impl}(K,T)$ is the market's primary quotation mechanism. It is not a forecast of future realised vol — it is the risk-neutral pricing vol for each specific contract.

Market conventions for quoting the smile:

• Strike space: $\sigma_{impl}(K, T)$ directly. Intuitive but makes comparison across spot levels and maturities difficult — a strike of 100 means something different when spot is at 95 vs. 105.
• Delta space: $\sigma_{impl}(\Delta, T)$ where $\Delta = e^{-qT}N(d_1)$. Standard in FX; increasingly used for equity index options. Normalises for moneyness in a scale-invariant way.
• Log-moneyness space: $\sigma_{impl}(x, T)$ where $x = \ln(K/F)$ and $F = S_0 e^{(r-q)T}$ is the forward. Natural for parameterisation (SVI, SABR). Setting $x = 0$ always corresponds to the forward-at-the-money level regardless of rates, dividends, or spot level.

Key market quotes for the equity smile:

• ATM vol: $\sigma_{ATM}$ at $K = F$ (the forward-at-the-money level). The single most important vol quote; used for variance swap pricing and as the anchor for the surface.
• 25-delta risk reversal (25d RR): $\sigma_{25d\text{-call}} - \sigma_{25d\text{-put}}$. Measures the skew. Negative for equities: OTM puts are more expensive than OTM calls. A 25d RR of $-3\%$ means the 25-delta put trades 3 vol points above the 25-delta call.
• 25-delta butterfly (25d BF): $\frac{1}{2}(\sigma_{25d\text{-call}} + \sigma_{25d\text{-put}}) - \sigma_{ATM}$. Measures the curvature of the smile. Always positive for a convex smile: both wings trade above ATM.

The two quantities RR and BF together with ATM vol fully specify a three-parameter smile family — sufficient for vanilla hedging and risk management of linear payoffs in vol space.

2. Why the Skew Exists: Three Structural Drivers

Leverage effect (Black 1976): when a company's stock price falls, its leverage (debt/equity ratio) increases, making the equity riskier and the firm's asset volatility higher. This creates a negative empirical correlation between spot returns and volatility returns. For major equity indices, $\text{Corr}(\Delta S / S, \Delta\sigma) \approx -0.7$. In the risk-neutral world, this translates to OTM puts (which pay in the event of a large downward move, i.e., high vol) being more valuable than symmetric pricing would suggest.

Crash risk and tail fear: post-1987, market participants pay a structural premium for OTM downside protection. Institutional investors — pension funds, insurance companies, asset managers with risk mandates — are systematic buyers of OTM puts to hedge their long equity exposures. Supply of these hedges is limited; dealers who sell them must be compensated for bearing the crash risk. The supply/demand imbalance drives OTM put IVs persistently above the level justified by historical crash frequencies alone.

Stochastic volatility and jumps: models with mean-reverting stochastic vol (Heston, SABR) produce a smile with negative skew when the spot-vol correlation $\rho < 0$. The magnitude of skew increases with $|\rho|$. Jump-diffusion models (Merton, Kou) produce a more pronounced short-dated skew: at short maturities, the probability of a single large jump dominates and creates extreme left-tail pricing. At long maturities, the central limit theorem dampens the jump impact and the surface flattens.

3. Arbitrage Constraints on the Vol Surface

A vol surface $\sigma(K,T)$ is arbitrage-free if and only if it satisfies two conditions:

Calendar spread no-arbitrage: total implied variance $w(K,T) = \sigma^2(K,T) \cdot T$ must be non-decreasing in $T$ for every fixed $K$:

$\frac{\partial w}{\partial T}(K, T) \geq 0 \quad \forall K, T$

Intuition: a call with longer maturity is always worth at least as much as a shorter-dated call at the same strike (for the same spot). If $w$ decreases in $T$, the longer-dated call is cheaper — selling the long-dated call and buying the short-dated one is a calendar spread with negative cost: a direct arbitrage.

Butterfly no-arbitrage (Breeden-Litzenberger): the risk-neutral density $p(K,T)$ derived from the call price surface via the Breeden-Litzenberger formula must be non-negative:

$p(K,T) = e^{rT} \frac{\partial^2 C}{\partial K^2}(K,T) \geq 0 \quad \forall K, T$

A negative density at $K$ means that a butterfly spread centred at $K$ (long call at $K-\varepsilon$, short two calls at $K$, long call at $K+\varepsilon$) has negative price — a direct cash-and-carry arbitrage with positive payoff and no cost.

4. The Breeden-Litzenberger Formula

The risk-neutral density of $S_T$ is recoverable directly from the call price surface without any model assumption beyond no-arbitrage:

$p(S_T = K) = e^{rT} \frac{\partial^2 C(K,T)}{\partial K^2}$

Derivation. The call price is the discounted expected payoff under the risk-neutral measure $\mathbb{Q}$:

$C(K,T) = e^{-rT} \mathbb{E}^{\mathbb{Q}}[(S_T - K)^+] = e^{-rT} \int_K^\infty (s - K) p(s) \, ds$

Differentiating once with respect to $K$:

$\frac{\partial C}{\partial K} = e^{-rT} \int_K^\infty (-1) p(s) \, ds = -e^{-rT} \mathbb{Q}(S_T > K)$

Differentiating again:

$\frac{\partial^2 C}{\partial K^2} = e^{-rT} p(K)$

Rearranging: $p(K) = e^{rT} \frac{\partial^2 C}{\partial K^2}$.

This result has a direct practical interpretation: buying a tight butterfly spread (for small $\varepsilon$) at strike $K$ is equivalent to buying a digital payoff of $\varepsilon^2 \cdot p(K)$ at maturity — the butterfly price is, to leading order, proportional to the risk-neutral density at that strike.

5. Term Structure of Vol and the VIX

The ATM vol term structure $\sigma_{ATM}(T)$ encodes information about the expected path of volatility under the risk-neutral measure:

• Upward sloping: current vol is low; the market expects mean reversion upward. Typical in calm, low-volatility regimes (e.g., 2017, 2019).
• Downward sloping (inverted): current vol is elevated and expected to revert downward. Typical during crises (March 2020, October 2008). The most expensive vol is at the short end.
• Flat: the market expects vol to remain roughly at current levels — rare and short-lived.

The VIX index is the model-free 30-day expected quadratic variation (annualised), computed directly from the strip of call and put prices across all strikes:

$\text{VIX}^2 = \frac{2e^{rT}}{T} \int_0^\infty \frac{C(K,T)\mathbf{1}_{K>F} + P(K,T)\mathbf{1}_{K \leq F}}{K^2} \, dK$

where $T = 30/365$ and $F = S_0 e^{(r-q)T}$ is the 30-day forward. This formula is model-free: it does not assume log-normal returns or any particular dynamics. It integrates the out-of-the-money option prices across all strikes, weighting by $1/K^2$.

The VIX is the market price of 30-day variance. Buying the VIX (synthetically, via a variance swap or replicating portfolio of OTM options) is equivalent to buying the entire risk-neutral variance strip. The VIX will exceed ATM vol whenever the smile is convex (BF > 0), because the OTM options — which trade above ATM vol — contribute to the integral. The difference between VIX and ATM vol is the butterfly premium of the variance swap over the vanilla ATM straddle.

Implementation

No standalone code is presented in this article. The accompanying notebook demonstrates:

1. Surface construction: building a realistic parameterised equity implied vol surface with negative skew, positive curvature, and upward-sloping term structure.
2. Surface visualisation: contour plot and smile slices by maturity, showing the flattening of skew with time.
3. Breeden-Litzenberger density extraction: numerical second derivative of the call price surface; comparison to the log-normal density at flat vol.
4. 25d risk reversal and butterfly: computation across maturities; verification of negative RR and positive BF.
5. Arbitrage check: comparison of a well-behaved parametric surface against a naively interpolated surface with butterfly violations.
6. VIX-style model-free variance: numerical integration of the OTM option strip; comparison to ATM vol across maturities.

All cells are executable end-to-end in the Pyodide environment.

Validation

Three model-free checks that any fitted surface must pass:

Flat vol benchmark: at flat vol $\sigma(K,T) = \sigma_0$ for all $(K,T)$, the Breeden-Litzenberger density should recover the log-normal density of the underlying under Black-Scholes — verify numerically that the extracted density integrates to 1 and matches $\text{LogNormal}(\ln S_0 + (r-q-\frac{1}{2}\sigma_0^2)T, \sigma_0^2 T)$.

Put-call parity: for any arbitrage-free surface, $C(K,T) - P(K,T) = S_0 e^{-qT} - K e^{-rT}$ must hold at every $(K,T)$. Violations indicate a surface that cannot be simultaneously consistent with calls and puts — a direct arbitrage.

Total variance monotonicity: verify that $w(K,T) = \sigma^2(K,T) \cdot T$ is non-decreasing in $T$ for all $K$ in the fitted surface. Any point where $\partial w / \partial T < 0$ is a calendar spread arbitrage.

Limitations

Static snapshot, not a dynamic model: the vol surface is a representation of today's market prices. It is not a model of how the surface evolves over time. Pricing path-dependent products (barriers, Asian options, cliquets) consistently with the surface requires a full dynamic model — local vol, stochastic vol (Heston), or local stochastic vol (LSV). A vol surface alone cannot price these products without additional modelling assumptions.

Discrete strike grid: the Breeden-Litzenberger formula and the VIX formula require a continuum of strikes. In practice, only a discrete set of liquid strikes is quoted. Interpolation between quoted strikes introduces model dependence: different interpolation schemes give different risk-neutral densities. The choice of interpolation scheme is a modelling decision, not a market observable.

Surface non-stationarity: in crisis periods (e.g., March 2020), the surface shape changes rapidly — the skew steepens, the term structure inverts, and bid-offer spreads widen dramatically. Historical calibrations become stale within hours. A desk that reprices using a surface calibrated at the start of the day is running stale Greeks by the close.

Liquidity and bid-offer: deep OTM options (e.g., 10-delta or lower) may have very wide bid-offer spreads — 2–5 vol points or more. The midpoint implied vol is an average of a wide range and is poorly defined. Mark-to-market for deep OTM positions is effectively mark-to-model. Risk systems that treat the implied vol of illiquid strikes as precisely known are overstating confidence in their Greeks.

The surface is not a probability forecast: the risk-neutral density extracted from the surface is not the physical (real-world) density. The two differ by the market price of risk. A fat-tailed risk-neutral density for $S_T$ does not mean the market expects a crash — it means the market is pricing crash protection expensively, which could reflect either genuine crash probability or excess demand for hedges (risk aversion).

Interview Angle

Junior (L1): "What is the equity volatility skew? Why do OTM puts have higher implied vol than ATM options? What is a risk reversal, and how is it quoted?" Expected answer: post-1987 phenomenon; OTM puts are in higher demand for portfolio protection; RR = $\sigma_{25d\text{-call}} - \sigma_{25d\text{-put}}$, negative for equities.

Senior (L2): "State the Breeden-Litzenberger formula. What are the two arbitrage constraints on a vol surface? How do you check a fitted surface for butterfly arbitrage?" Expected: $p(K) = e^{rT} \partial^2 C / \partial K^2$; calendar spread (total variance monotone in $T$) and butterfly (second derivative of calls positive in $K$); check by computing the density on a fine strike grid and verifying non-negativity everywhere.

Researcher (L3): "Decompose the equity skew into its three structural drivers. Which driver dominates at short maturities vs. long maturities? Design a model risk framework for a desk that prices OTM puts using an interpolated surface." Expected: leverage effect (always present), crash risk premium (demand-supply), stochastic vol / jumps (jumps dominate short-dated, SV dominates long-dated); model risk framework: multiple interpolation schemes, sensitivity of prices and Greeks to interpolation choice, liquidity-weighted confidence intervals on implied vols, daily surface validation against Breeden-Litzenberger and calendar spread checks.