Vega Hedging and Vol-of-Vol

Vega hedging removes the first-order sensitivity of an option book to a parallel implied-volatility change. It does not remove volatility risk. Implied volatility moves by expiry and strike, while an option's value is nonlinear in volatility and coupled to spot through vanna. Vol-of-vol is the uncertainty of future variance or implied-volatility dynamics; it is what makes a vega-flat book capable of substantial gains or losses.

First-order vega is a local coordinate

For a Black-Scholes option,

Vega = dV/dsigma = S exp(-qT) phi(d1) sqrt(T)

The formula is largest near ATM and grows roughly with the square root of maturity. A simple hedge sells a liquid option with opposite vega:

hedge_quantity = -vega_book / vega_hedge

This only offsets a one-vol-point parallel shift evaluated at today's spot, time, and surface. It should be called “ATM vega-flat,” not “volatility-neutral.”

Risk	Mathematical object	Typical hedge limitation
Vega	dV/dsigma	assumes a parallel move
Volga/vomma	d2V/dsigma2	residual to a large vol move
Vanna	d2V/(dS dsigma)	spot-vol correlation
Skew risk	change in slope by delta	one ATM hedge misses wings
Term risk	change by maturity	one expiry hedge misses curve

The second-order terms

For a finite surface move, a local expansion is:

dV ≈ Delta*dS + Vega*dIV
     + 0.5*Gamma*(dS)^2 + Vanna*dS*dIV
     + 0.5*Volga*(dIV)^2

Volga is often positive for far out-of-the-money and long-dated options. Long-volga positions gain from the magnitude of an IV move in either direction, subject to the limitations of a local Taylor expansion. Vanna matters in equity markets, where falling spot and rising IV are jointly common. A hedge that offsets vega at unchanged spot may acquire substantial exposure after a selloff.

Vol-of-vol in models and markets

In Heston stochastic volatility, xi controls the diffusion scale of variance:

dv_t = kappa(theta - v_t)dt + xi sqrt(v_t)dW_v,t

It contributes to smile curvature, while spot/variance correlation shapes skew. In SABR, nu plays an analogous role for stochastic alpha. Neither calibrated parameter is a directly tradable, model-free vol-of-vol quote; it depends on the model, calibration weights, and market liquidity. Nevertheless, changes in wing curvature, variance-option prices, and the dynamics of VIX-like instruments provide observable proxies.

Build a factor hedge

Replace a single scalar vega with a vector of surface-bucket sensitivities. For example, measure exposure to 1-month ATM level, 1-month 25-delta risk reversal, 1-month butterfly, 3-month ATM level, and so on. Solve a regularized hedge problem:

import numpy as np

def hedge_weights(exposures, hedge_matrix, ridge=1e-6):
    # columns are candidate hedge instruments
    a = hedge_matrix
    return np.linalg.solve(a.T @ a + ridge*np.eye(a.shape[1]),
                           -a.T @ exposures)

hedge_matrix should contain bumped-and-recalibrated risks, not only Black-Scholes vegas. Add constraints for liquidity, position limits, maturity, and transaction cost. An unconstrained least-squares solution may neutralize a surface factor with illiquid wings and unacceptable jump risk.

Dynamic hedge error

A vega hedge must be recalibrated after spot, time, and surface moves. Rebalancing too rarely leaves factor exposure; too frequently turns a noisy model estimate into costs. Test hedge performance under historical coherent surface shocks, not independently bumped nodes. A parallel 1-vol-point move is useful for reporting, but it is not a realistic stress for a crash.

Scenario	Vega-flat book can lose because
Spot down, IV up	vanna and skew repricing
Vol spike	volga and unstable hedge ratios
Front IV rises only	term-structure mismatch
Wings richen	curvature exposure
Liquidity shock	hedges cannot be executed at marks

Governance and attribution

Report raw market P&L alongside model attribution. When a surface calibration changes, distinguish quote moves from interpolation and parameter moves. Treat xi, nu, or similar model parameters as risk factors only after validating that their bumps correspond to stable observable surface changes. A beautiful factor hedge can fail if it maps to quantities no market maker actually quotes.

Key takeaways

• Vega neutrality protects only against a small parallel implied-volatility shift.
• Volga, vanna, skew, and term structure create material residual volatility risk.
• Vol-of-vol parameters explain model dynamics but are not automatically tradable quotes.
• Hedge a vector of coherent surface factors with liquidity and cost constraints.
• Validate hedge error under spot-vol shocks and full surface revaluation.