Existence and Uniqueness of Implied Volatility¶
Introduction¶
The well-posedness of implied volatility—existence and uniqueness of the inverse pricing map—is fundamental to using volatility as a quoting convention. This section rigorously establishes conditions under which the equation \(C_{\text{BS}}(\sigma) = C_{\text{market}}\) admits a unique solution, examines boundary behavior, and explores failure modes when no-arbitrage conditions are violated.
General Framework for Existence and Uniqueness¶
1. Abstract Formulation¶
Consider a pricing functional:
where \(\theta = (S, K, T, r, q)\) represents market parameters (including dividend yield \(q\) for generality).
Definition 4.2.1 (Implied Parameter)
For a given market price \(P_{\text{market}}\), the implied parameter \(\sigma_*\) is the solution to:
2. Conditions for Well-Posedness¶
Theorem 4.2.1 (Existence and Uniqueness via Monotonicity)
Let \(F: (a, b) \to \mathbb{R}\) be continuous and strictly monotone. Then for any \(P \in (\inf_{\sigma \in (a,b)} F(\sigma), \sup_{\sigma \in (a,b)} F(\sigma))\), there exists a unique \(\sigma_* \in (a, b)\) such that \(F(\sigma_*) = P\).
Proof. - Existence: By the Intermediate Value Theorem, since \(F\) is continuous and \(P\) lies in the range of \(F\) - Uniqueness: Strict monotonicity prevents multiple solutions □
3. Application to Black-Scholes¶
For the Black-Scholes call pricing function:
We establish:
- Domain: \((0, \infty)\)
- Monotonicity: \(\partial C_{\text{BS}}/\partial \sigma > 0\) (vega is positive)
- Range: \((C_{\text{intrinsic}}, C_{\text{max}})\) where:
- \(C_{\text{intrinsic}} = \max(S e^{-qT} - K e^{-rT}, 0)\)
- \(C_{\text{max}} = S e^{-qT}\) (undiscounted spot for non-dividend case)
Detailed Analysis of Range and Limits¶
1. Lower Bound: Intrinsic Value Limit¶
Theorem 4.2.2 (Zero Volatility Limit)
Proof. Consider the forward moneyness \(m = \frac{S e^{-qT}}{K e^{-rT}} = \frac{S}{K} e^{(r-q)T}\).
As \(\sigma \to 0^+\):
Case 1: \(m > 1\) (in-the-money forward) - \(\ln m > 0 \Rightarrow d_1, d_2 \to +\infty\) - \(\Phi(d_1), \Phi(d_2) \to 1\) - \(C_{\text{BS}} \to S e^{-qT} - K e^{-rT}\)
Case 2: \(m < 1\) (out-of-the-money forward) - \(\ln m < 0 \Rightarrow d_1, d_2 \to -\infty\) - \(\Phi(d_1), \Phi(d_2) \to 0\) - \(C_{\text{BS}} \to 0\)
Case 3: \(m = 1\) (at-the-money forward) - Requires more careful analysis using \(\Phi(x) = \frac{1}{2} + \frac{1}{\sqrt{2\pi}} x + O(x^3)\) for small \(x\) - Limit gives \(0\) (zero time value at zero volatility) □
2. Upper Bound: Infinite Volatility Limit¶
Theorem 4.2.3 (Infinite Volatility Limit)
Proof. As \(\sigma \to \infty\):
(dominated by the \(+\sigma\sqrt{T}/2\) term)
Therefore: - \(\Phi(d_1) \to 1\) - \(\Phi(d_2) \to 0\) - \(C_{\text{BS}} \to S e^{-qT} \cdot 1 - K e^{-rT} \cdot 0 = S e^{-qT}\) □
Economic interpretation: At infinite volatility, the option has value equal to the discounted spot (assuming immediate exercise), as the optionality dominates completely.
3. Characterization of Admissible Prices¶
Corollary 4.2.1 (Necessary and Sufficient Conditions for Existence)
Implied volatility exists for a market price \(C_{\text{market}}\) if and only if:
These are precisely the static no-arbitrage bounds on European call prices.
Strict Monotonicity: Vega Analysis¶
1. Vega Formula¶
The derivative of the Black-Scholes price with respect to volatility:
where \(\phi(x) = \frac{1}{\sqrt{2\pi}} e^{-x^2/2}\) is the standard normal density.
2. Positivity and Implications¶
Proposition 4.2.1 (Strict Positivity of Vega)
For all \((S, K, T, r, q, \sigma)\) with \(S > 0\), \(T > 0\), \(\sigma > 0\):
Proof. Each factor is strictly positive: - \(S e^{-qT} > 0\) (discounted spot) - \(\phi(d_1) > 0\) (Gaussian density is positive everywhere) - \(\sqrt{T} > 0\) □
Corollary 4.2.2 (Strict Monotonicity)
The map \(\sigma \mapsto C_{\text{BS}}(\sigma)\) is strictly increasing on \((0, \infty)\).
3. Uniform Lower Bound on Vega¶
While vega is always positive, its magnitude varies with moneyness and maturity.
Proposition 4.2.2 (Vega Bounds)
For \(\sigma \in [\sigma_{\min}, \sigma_{\max}]\) and fixed \((S, K, T)\):
Proof. On any compact interval \([\sigma_{\min}, \sigma_{\max}]\), the continuous positive function \(\mathcal{V}\) attains its infimum, which is strictly positive by Proposition 4.2.1. □
This ensures that the inverse map \(C \mapsto \sigma_{\text{IV}}\) has bounded derivative:
Continuity and Differentiability of the Inverse Map¶
1. Continuous Dependence on Price¶
Theorem 4.2.4 (Continuity of Implied Volatility)
The implied volatility map:
is continuous.
Proof. Let \(C_n \to C\) with \(C_n, C \in (C_{\text{intrinsic}}, S e^{-qT})\). Let \(\sigma_n = \mathcal{I}(C_n)\).
By definition: \(C_{\text{BS}}(\sigma_n) = C_n\)
Since \(\{\sigma_n\}\) is bounded (contained in preimage of bounded set under continuous \(C_{\text{BS}}\)), extract convergent subsequence \(\sigma_{n_k} \to \sigma_*\).
By continuity of \(C_{\text{BS}}\):
By uniqueness, \(\sigma_* = \mathcal{I}(C)\). Full sequence converges by uniqueness of limit. □
2. Smoothness via Implicit Function Theorem¶
Theorem 4.2.5 (Differentiability of Implied Volatility)
The implied volatility is \(C^\infty\) smooth with:
Proof. Apply the Implicit Function Theorem to:
We have: - \(\frac{\partial G}{\partial \sigma} = \mathcal{V}(\sigma) \neq 0\) (non-degeneracy condition satisfied) - \(C_{\text{BS}}\) is \(C^\infty\) in \(\sigma\)
Therefore, the implicit function \(\sigma_{\text{IV}}(C)\) is \(C^\infty\) with:
Higher derivatives follow by differentiation. □
3. Second Derivative: Curvature of Inverse Map¶
Differentiating the relation \(C_{\text{BS}}(\sigma_{\text{IV}}(C)) = C\):
Differentiate again:
where \(d\mathcal{V}/d\sigma\) is the vomma (or volga):
Boundary Behavior and Non-Existence Cases¶
1. Behavior Near Intrinsic Value¶
As \(C \to C_{\text{intrinsic}}^+\):
The rate of convergence depends on moneyness:
Proposition 4.2.3 (Rate of Convergence to Zero)
For deep ITM options with \(m = S e^{-qT}/(K e^{-rT}) \gg 1\):
as \(C \to C_{\text{intrinsic}}^+\).
2. Behavior Near Upper Bound¶
As \(C \to (S e^{-qT})^-\):
Proposition 4.2.4 (Divergence Rate)
As \(C \to S e^{-qT}\):
3. Non-Existence: Arbitrage Violations¶
Implied volatility fails to exist when:
Case 1: \(C \leq C_{\text{intrinsic}}\)
Price below intrinsic value violates static arbitrage (buy option, exercise immediately for profit)
Case 2: \(C \geq S e^{-qT}\)
Price at or above discounted spot (short option, buy stock, guaranteed profit)
In practice, small violations occur due to: - Bid-ask spreads - Illiquidity - Stale quotes - Microstructure noise
These cases are handled by: - Clipping prices to admissible range - Rejecting quotes as non-tradable - Using robust estimation methods
Stability Analysis¶
1. Lipschitz Continuity¶
Theorem 4.2.6 (Local Lipschitz Continuity)
On any compact interval \([C_1, C_2] \subset (C_{\text{intrinsic}}, S e^{-qT})\), the implied volatility map is Lipschitz continuous:
where:
Proof. By Mean Value Theorem:
for some \(\xi \in (C, C')\). The Lipschitz constant is the supremum of the derivative. □
2. Condition Number Analysis¶
The condition number for the inversion problem:
measures sensitivity of implied volatility to relative errors in price.
Proposition 4.2.5 (Ill-Conditioning Regimes)
The condition number \(\kappa \to \infty\) as:
1. \(C \to C_{\text{intrinsic}}^+\) (low volatility regime)
2. \(T \to 0\) (expiry approach)
This indicates that implied volatility extraction becomes increasingly sensitive to price errors in these regimes.
Extension to Other Instruments¶
1. Put Options¶
For European puts, by put-call parity:
Since put-call parity is model-independent:
for the same \((K, T)\). The existence and uniqueness analysis is identical, with modified intrinsic value bounds.
2. Binary Options¶
For digital (binary) calls with payoff \(\mathbb{1}_{S_T > K}\):
Monotonicity fails for binary options:
This can be positive or negative depending on moneyness, destroying uniqueness of implied volatility for digitals.
Summary¶
The existence and uniqueness of implied volatility rests on:
- Monotonicity: Strict positivity of vega ensures injectivity
- Range characterization: Limits \(\sigma \to 0, \infty\) match arbitrage bounds
- Continuity: Smooth dependence via Implicit Function Theorem
- Stability: Locally Lipschitz with condition number analysis
Key results: - Implied volatility exists \(\iff\) \(C \in (C_{\text{intrinsic}}, S e^{-qT})\) - The inverse map is \(C^\infty\) smooth - Numerical inversion is well-conditioned except near boundaries - Extension to non-vanilla payoffs requires careful monotonicity verification
This rigorous foundation justifies using implied volatility as a fundamental coordinate system for option pricing.
Exercises¶
Exercise 1. Consider a European call with \(S = 50\), \(K = 55\), \(T = 0.25\), \(r = 3\%\), and \(q = 0\). Compute the admissible price interval \((C_{\text{intrinsic}}, S e^{-qT})\) for implied volatility to exist. If \(C_{\text{market}} = 0.80\), verify that implied volatility exists.
Solution to Exercise 1
With \(S = 50\), \(K = 55\), \(T = 0.25\), \(r = 0.03\), and \(q = 0\), compute the admissible interval.
The lower bound (intrinsic value) is:
Computing \(K e^{-rT} = 55 \cdot e^{-0.0075} \approx 55 \cdot 0.99252 = 54.589\).
Since \(50 - 54.589 = -4.589 < 0\), the intrinsic value is:
The upper bound is:
Therefore the admissible interval is \((0, 50)\).
Since \(C_{\text{market}} = 0.80\) satisfies \(0 < 0.80 < 50\), implied volatility exists. This is an out-of-the-money option (\(S < K e^{-rT}\)), and the positive market price reflects pure time value, consistent with a well-defined positive implied volatility.
Exercise 2. Prove that the vega of a Black-Scholes call option satisfies \(\mathcal{V} > 0\) for all \(\sigma > 0\) directly from the formula
State explicitly which properties of the Gaussian density \(\phi\) and the parameters \(S\), \(T\) are used.
Solution to Exercise 2
The vega formula is:
We show each factor is strictly positive:
-
\(S e^{-qT} > 0\): The spot price \(S > 0\) by assumption (a traded asset has positive price), and \(e^{-qT} > 0\) since the exponential function is always positive.
-
\(\phi(d_1) > 0\): The standard normal density is \(\phi(x) = \frac{1}{\sqrt{2\pi}} e^{-x^2/2}\). The prefactor \(\frac{1}{\sqrt{2\pi}} > 0\) and \(e^{-x^2/2} > 0\) for all \(x \in \mathbb{R}\) (the exponential function is everywhere positive). Therefore \(\phi(d_1) > 0\) regardless of the value of \(d_1\).
-
\(\sqrt{T} > 0\): The time to maturity \(T > 0\) by assumption (the option has not yet expired), so \(\sqrt{T} > 0\).
Since all three factors are strictly positive, their product is strictly positive:
for all admissible parameter values with \(S > 0\), \(T > 0\), and \(\sigma > 0\).
Exercise 3. Using the Implicit Function Theorem applied to \(G(C, \sigma) = C_{\text{BS}}(\sigma) - C = 0\), derive the formula for the second derivative of implied volatility with respect to price:
Evaluate the sign of this expression at ATM (where \(d_1 d_2 \approx 0\)) and deep OTM (where \(d_1 d_2 > 0\)).
Solution to Exercise 3
Starting from \(G(C, \sigma) = C_{\text{BS}}(\sigma) - C = 0\) and differentiating implicitly with respect to \(C\):
Since \(\frac{\partial G}{\partial C} = -1\) and \(\frac{\partial G}{\partial \sigma} = \mathcal{V}\), we get:
Differentiating again with respect to \(C\):
Therefore:
The vomma (derivative of vega with respect to \(\sigma\)) is:
At ATM (\(S = K e^{-rT}\)): We have \(d_1 = \sigma\sqrt{T}/2\) and \(d_2 = -\sigma\sqrt{T}/2\), so \(d_1 d_2 = -\sigma^2 T/4 < 0\). Therefore \(d\mathcal{V}/d\sigma < 0\), and:
The inverse map is convex at ATM, meaning the implied volatility curve bows upward as a function of price.
Deep OTM (\(d_1, d_2\) both large and negative, or both positive for deep ITM): When the option is sufficiently out-of-the-money, \(d_1\) and \(d_2\) are both negative and large in magnitude, so \(d_1 d_2 > 0\). Then \(d\mathcal{V}/d\sigma > 0\) and:
The inverse map is concave for deep OTM options.
Exercise 4. The condition number for implied volatility extraction is \(\kappa = \frac{C}{\sigma_{\text{IV}} \mathcal{V}(\sigma_{\text{IV}})}\). For a near-ATM option with \(S = K = 100\), \(T = 1\), \(r = 0\), and \(\sigma_{\text{IV}} = 0.20\), compute \(\kappa\) numerically. Then compute \(\kappa\) for a deep OTM option with \(K = 130\) (same other parameters) and compare. Explain why implied volatility extraction is harder for OTM options.
Solution to Exercise 4
With \(S = K = 100\), \(T = 1\), \(r = 0\), and \(\sigma_{\text{IV}} = 0.20\):
The vega is:
Computing \(\phi(0.10) = \frac{1}{\sqrt{2\pi}} e^{-0.01/2} = 0.39695 \cdot e^{-0.005} \approx 0.39695 \cdot 0.99501 \approx 0.39497\).
So \(\mathcal{V}_{\text{ATM}} = 100 \times 0.39497 = 39.497\).
The ATM call price is approximately \(C_{\text{ATM}} = 100 \cdot \Phi(0.10) - 100 \cdot \Phi(-0.10) \approx 100(0.53983 - 0.46017) = 7.966\).
The condition number is:
Now for the deep OTM option with \(K = 130\):
Computing \(\phi(-1.2118) = \phi(1.2118) \approx 0.19069\).
The vega is \(\mathcal{V}_{\text{OTM}} = 100 \times 0.19069 = 19.069\).
The OTM call price: \(C = 100 \cdot \Phi(-1.2118) - 130 \cdot \Phi(-1.4118) \approx 100(0.1128) - 130(0.0790) \approx 11.28 - 10.27 = 1.01\).
The condition number is:
In this case, the OTM condition number is actually smaller. However, the key issue for OTM options is not the relative condition number but the absolute sensitivity: vega is much smaller for deep OTM options (\(\mathcal{V} = 19.07\) vs. \(39.50\)), so a fixed dollar error in price produces a larger absolute error in implied volatility via \(\Delta \sigma \approx \Delta C / \mathcal{V}\). As options move even further OTM, vega shrinks toward zero while prices also shrink, and the ratio \(1/\mathcal{V}\) grows, making extraction increasingly ill-conditioned.
Exercise 5. The zero-volatility limit of the Black-Scholes price requires a case analysis on the forward moneyness \(m = S e^{-qT}/(K e^{-rT})\). For the ATM-forward case \(m = 1\), show carefully using the Taylor expansion \(\Phi(x) \approx \frac{1}{2} + \frac{x}{\sqrt{2\pi}}\) that \(\lim_{\sigma \to 0^+} C_{\text{BS}} = 0\).
Solution to Exercise 5
For the ATM-forward case \(m = 1\), we have \(\ln m = 0\), so:
As \(\sigma \to 0^+\), both \(d_1 \to 0^+\) and \(d_2 \to 0^-\). Using the Taylor expansion \(\Phi(x) \approx \frac{1}{2} + \frac{x}{\sqrt{2\pi}}\) for small \(x\):
With \(m = 1\), the ATM-forward condition gives \(S e^{-qT} = K e^{-rT}\). Substituting into the Black-Scholes formula:
Using the approximations:
Therefore:
The call price vanishes linearly in \(\sigma\), confirming that the ATM-forward option has zero intrinsic value and its time value disappears as volatility goes to zero.
Exercise 6. For a binary (digital) call option with payoff \(\mathbb{1}_{S_T > K}\), the Black-Scholes price is \(D_{\text{BS}}(\sigma) = e^{-rT} \Phi(d_2)\). Show that \(\partial D_{\text{BS}} / \partial \sigma\) can be negative for certain values of moneyness and volatility. Provide a specific numerical example where uniqueness of implied volatility fails for a digital option.
Solution to Exercise 6
The digital call price under Black-Scholes is \(D_{\text{BS}}(\sigma) = e^{-rT} \Phi(d_2)\), where:
Computing the derivative with respect to \(\sigma\):
We compute:
More directly, one can write:
using the relation \(d_1 = d_2 + \sigma\sqrt{T}\). The sign of \(\partial D_{\text{BS}}/\partial\sigma\) is determined by \(-d_1\):
- If \(d_1 > 0\) (roughly ITM): \(\partial D_{\text{BS}}/\partial\sigma < 0\) (price decreases with \(\sigma\))
- If \(d_1 < 0\) (roughly OTM): \(\partial D_{\text{BS}}/\partial\sigma > 0\) (price increases with \(\sigma\))
Numerical example of non-uniqueness: Take \(S = 100\), \(K = 100\), \(T = 1\), \(r = 0.05\), \(q = 0\).
At \(\sigma = 0.10\): \(d_1 = \frac{0.05 + 0.005}{0.10} = 0.55\), \(d_2 = 0.45\), \(D = e^{-0.05} \Phi(0.45) \approx 0.9512 \times 0.6736 \approx 0.6408\).
At \(\sigma = 0.50\): \(d_1 = \frac{0.05 + 0.125}{0.50} = 0.35\), \(d_2 = -0.15\), \(D = e^{-0.05} \Phi(-0.15) \approx 0.9512 \times 0.4404 \approx 0.4189\).
At \(\sigma = 2.00\): \(d_1 = \frac{0.05 + 2.0}{2.0} = 1.025\), \(d_2 = -0.975\), \(D = e^{-0.05} \Phi(-0.975) \approx 0.9512 \times 0.1648 \approx 0.1568\).
Since \(D\) is not monotone in \(\sigma\) (it first decreases in this ITM-forward example), a price such as \(D = 0.40\) could correspond to two different volatility values. The pricing function first moves in one direction and then reverses, so the inverse map is not unique.
Exercise 7. Implement the Newton-Raphson iteration for implied volatility:
Starting from \(\sigma_0 = 0.50\), compute three iterations for a call with \(S = 100\), \(K = 100\), \(T = 0.5\), \(r = 0.02\), \(q = 0\), and \(C_{\text{market}} = 8.00\). Verify quadratic convergence by examining the ratio of successive errors.
Solution to Exercise 7
With \(S = 100\), \(K = 100\), \(T = 0.5\), \(r = 0.02\), \(q = 0\), and \(C_{\text{market}} = 8.00\).
Iteration 0: \(\sigma_0 = 0.50\)
Iteration 1: \(\sigma_1 \approx 0.2662\)
Iteration 2: \(\sigma_2 \approx 0.2673\)
The method has converged to \(\sigma_{\text{IV}} \approx 0.2673\) after three iterations.
Quadratic convergence check: Let \(\sigma^* \approx 0.2673\).
- Error after iteration 0: \(|e_0| = |0.500 - 0.2673| = 0.2327\)
- Error after iteration 1: \(|e_1| = |0.2662 - 0.2673| = 0.0011\)
- Error after iteration 2: \(|e_2| \approx 0.0000\) (essentially converged)
The ratio \(|e_1|/|e_0|^2 \approx 0.0011/0.0541 \approx 0.020\), and \(|e_2|/|e_1|^2\) is of similar order. The dramatic reduction in error (from \(0.23\) to \(0.001\) to effectively \(0\)) is characteristic of quadratic convergence, where the number of correct digits roughly doubles with each iteration.