Implied Volatility

Motivation: why this matters in quant finance

Volatility is the only input to the Black-Scholes formula that is not directly observable. Spot price, strike, time to maturity, and interest rate all come from the market or the contract. Volatility is different: you cannot look up "the volatility of IBM" on a Bloomberg terminal — you observe option prices, and volatility is what you back out of them.

Implied volatility is the number

\sigma_{\text{imp}}

that, when plugged into Black-Scholes, reproduces an observed market option price:

C_{\text{market}} = C_{\text{BS}}(S_0, K, T, r, \sigma_{\text{imp}})

Three reasons this single concept dominates the practical quant workflow:

Quoting convention. Options traders quote in vol, not price — "sell the 3-month 25-delta put at 18 vol" is the natural unit. Vol is stable across strikes and maturities in a way that price isn't.
The vol surface. Plotting $\sigma_{\text{imp}}(K, T)$ across strikes and maturities reveals the shapes — smiles, skews, term structures — that falsify the Black-Scholes model and motivate stochastic volatility and local volatility extensions.
Calibration. Every pricing model for exotic options calibrates by matching the implied-vol surface of liquid vanillas. The implied vol is the market's summary statistic of the risk-neutral distribution of returns.

This note defines implied volatility precisely, proves it is well-defined (existence and uniqueness), shows why the volatility smile is not an artefact of bad data but rather a direct falsification of the log-normal assumption, and sketches how practitioners compute implied vol numerically using Newton's method on vega. The reader should leave with a clear understanding that implied vol is a model-dependent transform of price — not a "true" underlying volatility.

The informal idea

Given a European call with observed market price $C_{\text{market}}$ , rearrange the Black-Scholes formula to solve for $\sigma$ :

C_{\text{BS}}(\sigma) = C_{\text{market}} \quad\Longrightarrow\quad \sigma = \sigma_{\text{imp}}

Unlike the other Black-Scholes inputs,

\sigma

appears only in

d_1

and

d_2

and cannot be isolated algebraically — inversion is numerical. But as a function of

\sigma

, the call price

C_{\text{BS}}(\sigma)

is strictly increasing and continuous, so for any arbitrage-free market price there is a unique

\sigma_{\text{imp}}

— solving the equation is well-posed.

What implied vol is not. It is not the "true" volatility of the stock. The Black-Scholes model assumes constant vol; real stocks have time-varying, stochastic, mean-reverting vol. Implied vol is the single number you'd have to pretend was constant in Black-Scholes to match the market. When the real-world dynamics deviate from log-normal, implied vol varies with

K

and

T

— not because the "true vol" varies, but because the Black-Scholes inverse lens is a lossy projection of a richer reality.

The smile and the skew. Plot

\sigma_{\text{imp}}(K)

for a single maturity

T

. For equity indices, the curve slopes down — out-of-the-money puts trade at higher implied vols than OTM calls (the equity skew). For FX, the curve is often symmetric and U-shaped (the smile). Both patterns violate Black-Scholes, which would produce a flat line. The shape is the market's verdict that tail risk is larger than log-normal, especially on the downside for equities.

Formal definition

The inversion problem

Given observed market inputs

S_0, K, T, r

and an observed call price

C_{\text{market}}

, the implied volatility

\sigma_{\text{imp}}

is defined by:

C_{\text{BS}}(S_0, K, T, r, \sigma_{\text{imp}}) = C_{\text{market}}

where

C_{\text{BS}}(\sigma) = S_0\Phi(d_1) - Ke^{-rT}\Phi(d_2), \qquad d_{1,2} = \frac{\ln(S_0/K) + (r \pm \sigma^2/2)T}{\sigma\sqrt{T}}

Analogous definitions hold for puts (via put-call parity, a call's implied vol equals its paired put's implied vol when strike and maturity match).

Existence and uniqueness

Proposition. For any strike $K > 0$ , maturity $T > 0$ , and market price $C_{\text{market}}$ satisfying the arbitrage bounds

$\max(S_0 - Ke^{-rT},\, 0) < C_{\text{market}} < S_0,$

there exists a unique $\sigma_{\text{imp}} \in (0, \infty)$ solving $C_{\text{BS}}(\sigma_{\text{imp}}) = C_{\text{market}}$ .

Proof sketch. On

(0, \infty)

$C_{\text{BS}}(\sigma)$ is continuous in $\sigma$ .
$\lim_{\sigma \to 0^+} C_{\text{BS}}(\sigma) = \max(S_0 - Ke^{-rT}, 0)$ — the zero-vol (deterministic) intrinsic value.
$\lim_{\sigma \to \infty} C_{\text{BS}}(\sigma) = S_0$ — the zero-strike asymptote (with infinite vol, $\Phi(d_1) \to 1$ , $\Phi(d_2) \to 0$ ).
Vega $\partial C_{\text{BS}}/\partial\sigma = S_0\phi(d_1)\sqrt{T} > 0$ for all $\sigma > 0$ .

Since $C_{\text{BS}}(\sigma)$ is continuous and strictly increasing from the intrinsic value to $S_0$ , it is a bijection onto $(\max(S_0 - Ke^{-rT}, 0), S_0)$ . For any market price in this open interval, the inverse $\sigma_{\text{imp}}$ exists and is unique. $\square$

What the arbitrage bounds mean. A call must cost strictly more than its discounted intrinsic value (otherwise buying the call and holding beats buying the stock outright at

t = 0

via some arbitrage), and strictly less than the stock itself (the payoff is capped at

S_T

). Market prices violating these bounds indicate stale quotes or arbitrage — no implied vol can reproduce them.

Numerical inversion: Newton's method

No closed-form inverse exists, but the monotonicity + smoothness of

C_{\text{BS}}

makes Newton's method fast and robust:

\sigma^{(n+1)} = \sigma^{(n)} - \frac{C_{\text{BS}}(\sigma^{(n)}) - C_{\text{market}}}{\text{vega}(\sigma^{(n)})}

Starting from $\sigma^{(0)} \approx 0.2$ , convergence to $10^{-6}$ tolerance typically requires 3–6 iterations for liquid options. For deep OTM or ITM strikes, vega vanishes and convergence is slow — industry-standard libraries use hybrid bisection/Newton schemes in those regimes. Brenner-Subrahmanyam (1988) gives a famous closed-form approximation for ATM options:

\sigma_{\text{imp}}^{\text{ATM}} \approx \frac{C_{\text{market}}}{S_0}\,\sqrt{\frac{2\pi}{T}}

Useful as an initial guess or a back-of-envelope sanity check.

Key properties

Put-call parity implies equal implied vols. For a call and put with the same strike and maturity, put-call parity

C - P = S_0 - Ke^{-rT}

holds under any model (it is a static hedging argument, not Black-Scholes-specific). Since Black-Scholes satisfies parity, call-implied and put-implied vols must agree — any deviation is an arbitrage signal, not a real effect.

The implied-vol surface.

\sigma_{\text{imp}}(K, T)

is typically plotted as a 2D surface. Its features:

Skew — the cross-section at fixed $T$ , showing $\sigma_{\text{imp}}$ vs $K$ (or often vs log-moneyness $\ln(K/S_0)$ or vs delta). Equity indices display a steep negative skew; FX crosses show milder smiles.
Term structure — the cross-section at fixed moneyness, showing $\sigma_{\text{imp}}$ vs $T$ . Typically upward-sloping in calm markets (longer maturities carry more vol-of-vol uncertainty) and inverted in crises (short-dated vol spikes).
Smile dynamics — how the skew shape shifts as spot moves (sticky strike, sticky delta, sticky local vol are three standard assumptions used by traders).

Vega-implied-vol duality. Vega

\partial C/\partial\sigma

is the sensitivity of the Black-Scholes price to implied vol. Equivalently, it is the derivative of the inversion map

C \mapsto \sigma_{\text{imp}}

inverted. A trader hedging an option against vol risk is hedging implied vol, not realised vol — they buy/sell other options with matching vega to offset the sensitivity to shifts in the implied-vol surface.

Implied vol is not additive. The implied vol of a portfolio of options is generally not a simple combination of the individual implied vols — unlike delta, which is additive, or gamma, which is additive. Portfolio vol risk is handled via the vega of each constituent summed up, not by any combined "portfolio implied vol."

Worked examples

Example 1: the inversion is monotone — numerical illustration

Take $S_0 = 100$ , $K = 100$ , $T = 0.5$ , $r = 0$ . Compute Black-Scholes call prices across a grid of $\sigma$ values:

$\sigma$	$C_{\text{BS}}$
$0.05$	$1.41$
$0.10$	$2.82$
$0.20$	$5.64$
$0.30$	$8.46$
$0.50$	$14.09$

Strictly increasing. Given any observed market price $C_{\text{market}}$ , linear (or Newton) interpolation between adjacent entries recovers $\sigma_{\text{imp}}$ to the required precision. If $C_{\text{market}} = 7.00$ , then $\sigma_{\text{imp}}$ lies between $0.20$ and $0.30$ — Newton's method starting from $\sigma^{(0)} = 0.25$ converges to $\sigma_{\text{imp}} \approx 0.248$ in a handful of steps.

Example 2: the equity skew

For an equity index like the S&P 500, a representative 3-month implied-vol skew is:

Strike (% of spot)	$\sigma_{\text{imp}}$
$80\%$	$28\%$
$90\%$	$22\%$
$100\%$ (ATM)	$18\%$
$110\%$	$16\%$
$120\%$	$15\%$

The skew is steep and monotone: OTM puts (low strikes) trade at $\sigma_{\text{imp}} = 28\%$ , far above ATM's $18\%$ . Interpretation: the market is pricing the left tail of the return distribution as fatter than log-normal — large down-moves happen more often than Black-Scholes would suggest. This is empirically confirmed by historical return data and economically understood through the leverage effect (firms become more levered as equity drops, increasing vol) and through crash risk (investors demand a premium for hedging against jumps).

A flat implied-vol surface would imply the Black-Scholes model is exactly right. Every non-flat surface is a falsification.

Example 3: computing implied vol in Python

import numpy as np
from scipy.stats import norm

def bs_call(S0, K, T, r, sigma):
    d1 = (np.log(S0 / K) + (r + 0.5 * sigma ** 2) * T) / (sigma * np.sqrt(T))
    d2 = d1 - sigma * np.sqrt(T)
    return S0 * norm.cdf(d1) - K * np.exp(-r * T) * norm.cdf(d2)

def bs_vega(S0, K, T, r, sigma):
    d1 = (np.log(S0 / K) + (r + 0.5 * sigma ** 2) * T) / (sigma * np.sqrt(T))
    return S0 * norm.pdf(d1) * np.sqrt(T)

def implied_vol(C_market, S0, K, T, r, tol=1e-8, max_iter=100):
    sigma = 0.2
    for _ in range(max_iter):
        price = bs_call(S0, K, T, r, sigma)
        vega = bs_vega(S0, K, T, r, sigma)
        if vega < 1e-12:
            break
        diff = price - C_market
        if abs(diff) < tol:
            return sigma
        sigma -= diff / vega
        sigma = max(sigma, 1e-6)
    return sigma

print(implied_vol(7.00, 100, 100, 0.5, 0.0))
# 0.24795367612...

Six iterations to converge from $\sigma^{(0)} = 0.2$ to $\sigma_{\text{imp}} \approx 0.2480$ . The bound $\sigma \geq 10^{-6}$ prevents negative iterates if a bad initial guess produces overshoot.

Common confusions and pitfalls

"Implied vol is the 'true' volatility of the underlying." No. Implied vol is the number that makes Black-Scholes match the market price — nothing more. Realised vol (the sample volatility of returns) is the true quantity in the real-world sense. Implied vol is a risk-neutral projection: it reflects both the market's estimate of future realised vol and a risk premium for holding volatility risk. The two rarely agree.

"If implied vol is 20%, the stock will move 20% per year." Implied vol is an annualised standard deviation under the risk-neutral measure. Expected realised moves under the real-world measure are generally smaller — the variance risk premium is persistently negative (realised vol under-delivers relative to implied vol), which is why selling straddles tends to be a historically profitable (but risky) strategy.

"The implied-vol smile means Black-Scholes is wrong." The smile means the market does not price options as if returns were log-normal. The Black-Scholes formula itself is an identity — given log-normal dynamics and no arbitrage, it gives the price. The failure is in the assumption of log-normal dynamics, not in the formula. Stochastic volatility and local volatility models attempt to reproduce the observed smile under richer dynamics.

"Implied vol is an input to pricing." For vanilla options quoted in price, implied vol is an output of the inversion. For exotic options priced in a calibrated model, implied vol is a target — the model's parameters are calibrated until it reproduces the observed vanilla implied-vol surface, and only then is the exotic priced. The direction depends on which side of the calibration pipeline you stand on.

"Out-of-the-money implied vols are unreliable because vega is small." The computation of implied vol from a given price is numerically delicate for deep-OTM/ITM options (small vega means the inversion is ill-conditioned). But the implied vol itself is still defined. What's actually unreliable is the quoted price — wide bid-ask spreads and low volume produce noisy prices, and the noise amplifies through the inversion. Most vol surfaces are fit with smoothness penalties or parametric (SVI, SABR) forms specifically to regularise the wings.

Where this goes next

Derivation of the Black-Scholes Formula: The formula being inverted to extract implied vol. A thorough understanding of implied vol requires understanding what Black-Scholes assumes about the risk-neutral distribution.
Risk-Neutral Measure: Implied vol is the standard deviation of log-returns under $\mathbb{Q}$ . Its difference from realised vol (under $\mathbb{P}$ ) is the variance risk premium, a direct measurement of the $\mathbb{Q}$ -vs- $\mathbb{P}$ gap.
Stochastic Volatility Models: Models (Heston, SABR, rBergomi) that endogenise the vol surface. These are the industry-standard tools for pricing exotics consistently with the vanilla vol surface.
Quantitative Research: Variance risk premium strategies, vol-of-vol estimation, and vol trading are heavily studied research directions — all grounded in implied vs realised vol analysis.
Algorithmic Trading: Market-makers quote in implied vol, manage a vol book, and hedge vega across strikes. The vol surface dynamics are the core state variable.