Chapter 12: Implied Volatility¶

This chapter develops the theory of implied volatility---the market's consensus volatility extracted from observed option prices---as both a practical quoting convention and a window into the risk-neutral distribution. Starting from the definition as the inverse of the Black-Scholes pricing map, we establish existence and uniqueness via monotonicity and the intermediate value theorem, characterize the static and dynamic arbitrage constraints on the implied volatility surface, derive the model-free results connecting it to the risk-neutral density, local volatility, and variance swaps, study the asymptotic behavior in extreme regimes of strike and maturity, and analyze the sensitivities and smile dynamics that govern hedging in the presence of a non-flat smile.

Key Concepts¶

Implied Volatility as Inverse Pricing Map¶

The implied volatility \(\sigma_{\text{IV}}(K,T)\) is defined as the unique \(\sigma > 0\) solving \(C_{\text{BS}}(S_0, K, T, r, \sigma) = C_{\text{market}}(K,T)\). The Black-Scholes pricing map \(\mathcal{C}: (0,\infty) \to (C_{\text{intrinsic}}, S)\) is a diffeomorphism with smooth inverse \(\sigma_{\text{IV}} = \mathcal{C}^{-1}(C_{\text{market}})\), established by the strict positivity of vega \(\nu = S\phi(d_1)\sqrt{T} > 0\) (ensuring injectivity), the boundary behavior \(C_{\text{BS}} \to (S - Ke^{-rT})^+\) as \(\sigma \to 0\) and \(C_{\text{BS}} \to S\) as \(\sigma \to \infty\) (characterizing the range), and the intermediate value theorem (guaranteeing existence). Implied volatility serves as a coordinate transformation from price space to volatility space, re-parameterizing option prices through a unit-free quantity that facilitates comparison across strikes, maturities, and underlyings. Newton-Raphson iteration \(\sigma^{(n+1)} = \sigma^{(n)} - (C_{\text{BS}}(\sigma^{(n)}) - C_{\text{mkt}})/\nu(\sigma^{(n)})\) converges quadratically due to the smoothness and strict positivity of vega, while Brent's method provides a robust derivative-free alternative for production systems.

Existence and Uniqueness¶

Existence and uniqueness of the inverse map rest on a rigorous analysis of the pricing functional \(F(\sigma;\theta)\) where \(\theta = (S, K, T, r, q)\). The zero-volatility limit \(\lim_{\sigma \to 0^+} C_{\text{BS}} = \max(Se^{-qT} - Ke^{-rT}, 0)\) and the infinite-volatility limit \(\lim_{\sigma \to \infty} C_{\text{BS}} = Se^{-qT}\) establish the admissible price domain. Implied volatility exists if and only if \(\max(Se^{-qT} - Ke^{-rT}, 0) < C_{\text{market}} < Se^{-qT}\)---precisely the static no-arbitrage bounds. The inverse map is \(C^\infty\) smooth by the Implicit Function Theorem, with derivative \(d\sigma_{\text{IV}}/dC = 1/\mathcal{V}(\sigma_{\text{IV}})\) and second derivative involving the vomma \(d\mathcal{V}/d\sigma = Se^{-qT}\phi(d_1)\sqrt{T} \cdot d_1 d_2 / \sigma\). Stability analysis reveals local Lipschitz continuity with condition number \(\kappa = C/(\sigma_{\text{IV}} \cdot \mathcal{V})\) that diverges near the intrinsic value boundary and at expiry, indicating ill-conditioned regimes for numerical extraction.

Arbitrage Constraints on Implied Volatility¶

The implied volatility surface must satisfy no-arbitrage conditions that restrict its shape independently of any model. The butterfly constraint \(\partial^2 C/\partial K^2 \geq 0\) ensures non-negativity of the risk-neutral density via Breeden-Litzenberger, equivalent to requiring convexity of call prices in strike. The calendar spread constraint \(\partial C/\partial T \geq 0\) requires total variance \(w(K,T) = \sigma_{\text{IV}}^2 T\) to be non-decreasing in maturity. Durrleman's complete characterization establishes that the surface is arbitrage-free if and only if the joint condition \(g(y,T) = (1 - yw_y/(2w))^2 - w_y^2(1/w + 1/4)/4 + w_{yy}/2 \geq 0\) holds together with \(w_T \geq 0\), where \(y = \ln(K/F)\) is log-moneyness. Gatheral's constraint provides a sufficient condition via convexity of total variance in log-moneyness, and Lee's moment formula constrains the wing behavior: \(\lim_{|y|\to\infty} \sigma_{\text{IV}}^2(y,T)\,T/|y| = 2/m_\pm\) where \(m_\pm\) are the maximum finite moments.

The Implied Volatility Surface: Smile, Skew, and Term Structure¶

Plotting \(\sigma_{\text{IV}}(K,T)\) reveals systematic patterns that violate the Black-Scholes constant-volatility assumption. The smile is a U-shaped pattern with \(\sigma_{\text{IV}}(K) > \sigma_{\text{IV}}(K_{\text{ATM}})\) for strikes far from ATM, reflecting fat tails in the risk-neutral distribution; it is typical in FX markets. The skew (or smirk) is a monotonically decreasing pattern \(\partial\sigma_{\text{IV}}/\partial K < 0\) characteristic of equity indices post-1987, reflecting the leverage effect and crash risk premium; the 25-delta risk reversal \(\text{RR}_{25} = \sigma_{25\Delta C} - \sigma_{25\Delta P}\) quantifies the asymmetry. The Taylor expansion around ATM in log-moneyness gives \(\sigma(y,T) = \sigma_0(T) + \sigma_1(T)\,y + \frac{1}{2}\sigma_2(T)\,y^2 + O(y^3)\), where \(\sigma_1\) encodes the skew (related to distributional skewness via \(\sigma_1 \approx -\gamma_3/(6\sigma_0)\)) and \(\sigma_2\) encodes the curvature (related to excess kurtosis via \(\sigma_2 \approx (\gamma_4 - 3)/(24\sigma_0)\)). Parametric smile models include the SVI parametrization \(w(y) = a + b(\rho(y-m) + \sqrt{(y-m)^2 + \sigma^2})\), the SSVI surface extension of Gatheral-Jacquier, and the SABR stochastic volatility model with approximate implied volatility formula involving the backbone \((FK)^{-(1-\beta)/2}\) and correction terms in \(T\).

Term Structure of Implied Volatility¶

The term structure \(\sigma_{\text{ATM}}(T)\) describes how implied volatility varies with maturity. In upward-sloping (normal) regimes, current volatility is below its long-run mean and mean reversion drives expectations higher; in inverted regimes (post-crisis), short-dated vol exceeds long-dated vol. The forward implied volatility \(\sigma_{\text{fwd}}^2(T_1,T_2) = (w(T_2) - w(T_1))/(T_2 - T_1)\) and the instantaneous forward volatility \(\sigma_{\text{inst}}^2(T) = dw/dT = \sigma_{\text{ATM}}^2 + 2T\sigma_{\text{ATM}}\,d\sigma_{\text{ATM}}/dT\) decompose the variance curve into forward-looking components. In the Heston model, \(\sigma_{\text{ATM}}^2(T) \approx v_0 + \kappa(\theta - v_0)T + O(T^2)\), converging to \(\sqrt{\theta + \xi^2/(4\kappa)}\) for large \(T\), with the initial slope \(d\sigma_{\text{ATM}}^2/dT|_{T=0} = \kappa(\theta - v_0)\) governing whether the term structure is upward or downward sloping. The no-arbitrage calendar constraint requires \(dw/dT \geq 0\), which permits a decreasing \(\sigma_{\text{ATM}}(T)\) as long as total variance remains non-decreasing.

Empirical Implied Volatility Smile (SPX)¶

The S&P 500 smile is the canonical example of equity index smile behavior. Pre-1987, the smile was approximately flat; post-Black Monday, a pronounced downside skew emerged permanently, driven by crashophobia and portfolio insurance demand. Modern SPX characteristics include ATM IV of 12--18\% in normal regimes (25--50\%+ in crisis), 25-delta put skew of 4--8 vol points, and 10-delta put skew of 8--15 vol points. The smile exhibits regime dependence (steeper during stress), seasonality (FOMC humps, earnings effects), and cross-market variation (VIX options display an inverted skew with ATM vol of 80--120\%). Put-call parity ensures \(\sigma_{\text{IV}}^{\text{call}}(K,T) = \sigma_{\text{IV}}^{\text{put}}(K,T)\), and deviations signal arbitrage, illiquidity, or dividend effects.

Model-Free Results: Breeden-Litzenberger, Dupire, and Variance Swaps¶

Several powerful results connect implied volatility to the risk-neutral distribution without assuming any model. The Breeden-Litzenberger formula \(q(K) = e^{rT}\,\partial^2 C/\partial K^2\) extracts the full risk-neutral density from the second derivative of the call price surface, with non-negativity of \(q(K)\) equivalent to the no-butterfly-arbitrage condition. The implied volatility surface encodes distributional moments: the ATM level reflects variance, the skew slope encodes skewness via \(\text{Skew}^{\mathbb{Q}} \approx -6F\mathcal{S}\sqrt{T}/\sigma_{\text{ATM}}\), and the curvature encodes kurtosis via \(\text{Kurt}^{\mathbb{Q}} - 3 \approx 12F^2\mathcal{C}\,T\). Dupire's formula \(\sigma_{\text{loc}}^2(K,T) = (\partial_T C + qC + (r-q)K\partial_K C)/(\frac{1}{2}K^2\partial_{KK}C)\) determines the local volatility surface entirely from the market call price surface, establishing a one-to-one map between arbitrage-free call surfaces and local volatility functions. The variance swap fair strike \(K_{\text{var}} = (2e^{rT}/T)\int_0^\infty C(K)/K^2\,dK\) is a model-free integral over the call price surface, providing a direct measure of expected integrated variance; the VIX formula \(\text{VIX}^2 = (2/T)\sum_i (\Delta K_i/K_i^2)\,Q(K_i)\) discretizes this for practical computation. The variance risk premium \(\text{VRP} = \sigma_{\text{IV}}^2 - \mathbb{E}^P[\text{RV}]\) captures the wedge between risk-neutral and physical expectations.

Implied Volatility Asymptotics¶

Option prices and implied volatility exhibit characteristic behavior in extreme regimes. In the short-maturity limit (\(T \to 0\)), the Berestycki-Busca-Florent theorem gives \(\sigma_{\text{IV}}(K,T) \to \sigma_{\text{loc}}(K,0)\): ATM implied volatility converges to the spot volatility \(\sigma_{\text{spot}}\), and the OTM smile steepens as \(\sigma_{\text{IV}}^2 T \sim 2I(K)\) where \(I(K)\) is a rate function from large deviations theory. For the Heston model, the small-time expansion gives \(\sigma_{\text{IV}}(y,T) = \sqrt{v_0} + (\rho\xi/4)\,y + O(T)\), showing that short-dated skew is proportional to correlation times vol-of-vol. Hagan's SABR asymptotic formula provides a widely used closed-form approximation with the backbone \((FK)^{-(1-\beta)/2}\) and first-order time correction. In the ATM expansion, \(\sigma(y) = \sigma_0 + \sigma_1 y + \frac{1}{2}\sigma_2 y^2 + O(y^3)\), the coefficients \(\sigma_i\) connect to risk-neutral moments and to model parameters: in Heston, \(\sigma_0^2 = v_0\), \(\sigma_1 = \rho\xi/(4\sqrt{v_0})\), and market quoting conventions express skew via risk reversals (\(\text{RR} \approx 2\sigma_1 y_\Delta\)) and curvature via butterflies (\(\text{BF} \approx \frac{1}{2}\sigma_2 y_\Delta^2\)). In the wing asymptotics (\(|y| \to \infty\)), Roger Lee's moment formula relates the implied volatility wing slopes to the existence of moments: \(p_\pm = \lim_{|y|\to\infty} \sigma_{\text{IV}}^2(y,T)\,T/|y| = 2/m_\pm\) where \(m_+\) (\(m_-\)) is the supremum of \(p\) such that \(\mathbb{E}[S_T^{1+p}] < \infty\) (\(\mathbb{E}[S_T^{-p}] < \infty\)), with the constraint \(0 < p_\pm \leq 2\) ensuring finite variance. Empirically, equity indices exhibit asymmetric wings (\(p_- < p_+\)) reflecting fat left tails, while FX markets show symmetric wings (\(p_- \approx p_+ \approx 2\)).

Implied Volatility Sensitivities and Hedging¶

Vega \(\mathcal{V} = \partial C/\partial\sigma_{\text{IV}} = Se^{-qT}\phi(d_1)\sqrt{T}\) quantifies the option's sensitivity to implied volatility, is maximized at ATM, and grows as \(\sqrt{T}\) for moderate maturities. Higher-order volatility Greeks include vanna \(\partial^2 P/\partial S\partial\sigma = -\mathcal{V}\,d_2/(S\sigma\sqrt{T})\), which captures the spot-vol cross-sensitivity and is crucial for the leverage effect, and volga (vomma) \(\partial^2 P/\partial\sigma^2 = \mathcal{V}\,d_1 d_2/\sigma\), which measures vol convexity and is approximately zero at ATM. P&L attribution decomposes into \(\text{P\&L} = \Delta\,dS + \frac{1}{2}\Gamma(dS)^2 + \mathcal{V}\,d\sigma + \text{Vanna}\,dS\,d\sigma + \frac{1}{2}\text{Volga}\,(d\sigma)^2 + \Theta\,dt\). Vega bucketing by maturity and strike enables term structure and skew hedging using multiple instruments. The sticky-strike assumption (\(\partial\sigma_{\text{IV}}(K)/\partial S = 0\)) yields \(\Delta_{\text{total}} = \Delta_{\text{BS}}\) and is consistent with local volatility models; the sticky-delta assumption (\(\partial\sigma_{\text{IV}}(\Delta)/\partial S = 0\)) requires the smile-adjusted delta \(\Delta_{\text{smile}} = \Delta_{\text{BS}} + \mathcal{V}\cdot\partial\sigma_{\text{IV}}/\partial S\) and is more consistent with stochastic volatility and FX market conventions. Empirically, neither holds exactly; the skew stickiness ratio (SSR), typically 0.3--0.6 for SPX, measures the actual behavior between these extremes.

Smile Dynamics: Forward Smile and Dynamic Consistency¶

The forward smile \(\sigma_{\text{fwd}}(m, T_1, T_2)\) describes the conditional distribution of future returns implied by today's option prices. Total variance decomposes as \(w(T_2) = w(T_1) + w_{\text{fwd}}(T_1, T_2)\), giving \(\sigma_{\text{fwd}}^2 = (\sigma_{\text{IV}}^2(T_2)T_2 - \sigma_{\text{IV}}^2(T_1)T_1)/(T_2 - T_1)\). Local volatility models predict the forward smile flattens rapidly---contradicting persistent empirical skew---while stochastic volatility models (Heston, Bergomi) produce more realistic persistent forward skew. Dynamic consistency requires that the forward smile implied by today's surface be consistent with the smile observed when that future date arrives; formally, \(\sigma_{\text{fwd}}^{\text{model}}(K,T;X_t,\Theta_0) \stackrel{d}{=} \sigma_{\text{IV}}^{\text{recalibrated}}(K,T)\). Most models violate this: local vol is dynamically inconsistent (forward smile flattens), stochastic vol is moderate, and Bergomi's variance curve framework achieves the best consistency by modeling forward variances directly via \(d\xi_t^T = \xi_t^T\,\omega(T-t)\,dZ_t\).

Empirical Stylized Facts of Smile Dynamics¶

Robust empirical regularities govern smile evolution across asset classes. The leverage effect produces \(\Delta\sigma_{\text{ATM}} \approx \beta\,\Delta S/S\) with \(\beta \approx -2.0\) for SPX, and the response is asymmetric (\(|\beta_{\text{down}}| > |\beta_{\text{up}}|\)). The spot-vol correlation is strongly negative for equities (\(\rho_{S,\sigma} \approx -0.75\) for SPX), near zero for major FX pairs, and moderately negative for gold. Short-dated vol responds more strongly to spot moves than long-dated vol, causing term structure inversion during stress approximately 15--20\% of the time. Implied volatility exhibits strong persistence (daily autocorrelation \(\rho \approx 0.95\)) with mean reversion half-life of 2--4 weeks, and vol-of-vol is approximately 3--5\% annualized. The skew stickiness ratio (SSR) of 0.3--0.6 places empirical behavior between sticky strike and sticky delta. These stylized facts constrain model selection: a realistic model requires negative spot-vol correlation, mean-reverting variance, non-trivial vol-of-vol, and multi-factor term structure dynamics.

Role in the Book

Implied volatility provides the empirical bridge between the Black-Scholes formula (Chapter 6) and the richer models that explain the smile. Dupire's formula connects directly to local volatility (Chapter 13), the smile's shape constrains stochastic volatility model calibration (Chapter 14, Chapter 16), and the VIX/variance swap theory provides model-free benchmarks for calibration (Chapter 17). The wing asymptotics and moment formulas connect to the characteristic function methods of Chapter 9 and the affine theory of Chapter 15. The sensitivity analysis and hedging framework extend the Greeks of Chapter 10 to the smile-adjusted setting essential for practical risk management (Chapter 22).