Implicit Differentiation

Motivation: why this matters in quant finance

In most textbook calculus problems, you have an explicit function

y = f(x)

and you differentiate directly. But in quant finance, the most important quantities are often defined implicitly — through an equation you cannot solve in closed form. The canonical example is implied volatility: given a market price

C_{\text{mkt}}

, implied vol

\sigma_{\text{impl}}

is defined by:

C_{\text{BS}}(S, K, T, r, \sigma_{\text{impl}}) = C_{\text{mkt}}

You cannot write $\sigma_{\text{impl}} = g(S, K, T, r, C_{\text{mkt}})$ in closed form — no such explicit formula exists. But you still need the sensitivities: how does implied vol change when the spot moves ( $\partial\sigma_{\text{impl}}/\partial S$ )? When time passes ( $\partial\sigma_{\text{impl}}/\partial T$ )? When the market price changes? Implicit differentiation answers these questions without requiring an explicit solution.

Beyond implied volatility, implicit differentiation appears whenever a model is calibrated to market data (calibrated parameters depend implicitly on observables), whenever an equilibrium condition defines a quantity (yield-to-maturity, breakeven inflation, internal rate of return), and whenever you work with constraint equations in optimisation (Lagrange multipliers).

Definition and method

The idea

Suppose

x

and

y

are related by an equation

F(x, y) = 0

, where you cannot (or do not want to) solve for

y

as an explicit function of

x

. To find

dy/dx

, differentiate both sides with respect to $x$ , treating

y

as a function of

x

and using the chain rule wherever

y

appears:

\frac{d}{dx}F(x, y(x)) = 0 \implies F_x + F_y \cdot \frac{dy}{dx} = 0

Solving:

\frac{dy}{dx} = -\frac{F_x}{F_y}

provided

F_y \neq 0

. This is the implicit function theorem in its simplest form: if

F(x_0, y_0) = 0

and

F_y(x_0, y_0) \neq 0

, then near

(x_0, y_0)

there exists a smooth function

y = y(x)

satisfying

F(x, y(x)) = 0

, and its derivative is

-F_x / F_y

Multivariable extension

If $F(x_1, x_2, \ldots, x_n, y) = 0$ defines $y$ implicitly as a function of $(x_1, \ldots, x_n)$ :

\frac{\partial y}{\partial x_i} = -\frac{\partial F / \partial x_i}{\partial F / \partial y} = -\frac{F_{x_i}}{F_y}

Each partial derivative of

y

is the ratio of two partial derivatives of

F

, with a sign flip. This is the formula used repeatedly for implied volatility sensitivities.

Key results

The Implicit Function Theorem

If $F(x_0, y_0) = 0$ , $F$ is continuously differentiable near $(x_0, y_0)$ , and $F_y(x_0, y_0) \neq 0$ , then:

There exists a unique smooth function $y(x)$ near $x_0$ with $F(x, y(x)) = 0$ .

$y'(x) = -F_x / F_y$ .

The condition $F_y \neq 0$ is essential. Geometrically, it means the curve $F = 0$ is not vertical at the point — you can locally express $y$ as a function of $x$ . If $F_y = 0$ , the curve has a vertical tangent and $dy/dx$ is undefined (or infinite).

Finance interpretation: For implied vol,

F = C_{\text{BS}}(\sigma) - C_{\text{mkt}} = 0

, so

F_\sigma = \mathcal{V}

(vega). The condition

F_\sigma \neq 0

means vega must be nonzero — the Black-Scholes price must be sensitive to volatility at the solution point. This is true for any non-degenerate option (

T > 0

, option not deep ITM/OTM), which is why implied vol is well-defined for practically all traded options.

Second-order implicit derivatives

To find

d^2y/dx^2

, differentiate the first-order result

dy/dx = -F_x/F_y

with respect to

x

using the quotient rule and the chain rule (remembering

y

depends on

x

\frac{d^2y}{dx^2} = -\frac{F_{xx}F_y^2 - 2F_{xy}F_xF_y + F_{yy}F_x^2}{F_y^3}

This gives the curvature of the implicit curve and is used when computing second-order sensitivities of calibrated parameters (e.g., the curvature of the implied vol smile with respect to strike).

Examples and applications

Example 1: implied volatility sensitivities

The defining equation is $F(S, K, T, r, \sigma) = C_{\text{BS}}(S, K, T, r, \sigma) - C_{\text{mkt}} = 0$ , where $C_{\text{mkt}}$ is treated as fixed (or as another variable).

Sensitivity of implied vol to spot:

\frac{\partial\sigma_{\text{impl}}}{\partial S} = -\frac{\partial C_{\text{BS}}/\partial S}{\partial C_{\text{BS}}/\partial \sigma} = -\frac{\Delta}{\mathcal{V}}

This is the "sticky-strike" implied vol sensitivity. When the spot moves by $\Delta S$ , the implied vol adjusts by approximately $-(\Delta/\mathcal{V})\Delta S$ to keep the market price matched.

Sensitivity of implied vol to time:

\frac{\partial\sigma_{\text{impl}}}{\partial T} = -\frac{\Theta}{\mathcal{V}}

This tells you how the implied vol surface shifts as time passes, holding everything else fixed.

Sensitivity of implied vol to the market price:

\frac{\partial\sigma_{\text{impl}}}{\partial C_{\text{mkt}}} = -\frac{-1}{\mathcal{V}} = \frac{1}{\mathcal{V}}

A $1 increase in the market price raises implied vol by $1/\mathcal{V}$ . Since vega is in dollars per vol-point, the inverse is in vol-points per dollar. This is why options with small vega (deep OTM, near expiry) have implied vols that are very sensitive to small price changes — the "vega instability" of short-dated options.

Example 2: yield to maturity

The yield to maturity $y$ of a bond with price $P$ , coupon $c$ , and maturity $T$ is defined implicitly by:

\sum_{i=1}^{n} \frac{c}{(1+y)^{t_i}} + \frac{F}{(1+y)^T} = P

This cannot be solved for $y$ in closed form (for $n > 4$ ). But implicit differentiation gives:

\frac{\partial y}{\partial P} = -\frac{1}{\partial P_{\text{formula}}/\partial y} = \frac{1}{D \cdot P}

where $D$ is the (modified) duration. A dollar increase in the bond price implies a yield decrease of $1/(D \cdot P)$ .

Example 3: internal rate of return

The IRR $r^*$ of a cash flow stream $(C_0, C_1, \ldots, C_n)$ is defined by:

\sum_{i=0}^{n} \frac{C_i}{(1 + r^*)^i} = 0

Implicit differentiation with respect to any cash flow $C_k$ gives the sensitivity of the IRR to that specific cash flow — useful in project finance and capital budgeting when estimating how changes in projected cash flows affect the return.

Example 4: the unit circle (textbook illustration)

The equation $x^2 + y^2 = 1$ defines $y$ implicitly as a function of $x$ (near any point where $y \neq 0$ ). Differentiating both sides:

2x + 2y\frac{dy}{dx} = 0 \implies \frac{dy}{dx} = -\frac{x}{y}

This is the classic textbook example: the slope of the circle at $(x, y)$ is $-x/y$ , which is undefined at $y = 0$ (the horizontal tangent points $(1, 0)$ and $(-1, 0)$ ... actually there $x/y$ diverges, corresponding to vertical tangents). The structure $dy/dx = -F_x/F_y$ is the same in every implicit differentiation problem.

Common confusions and pitfalls

Forgetting the chain rule when differentiating $y$ . In

F(x, y) = 0

y

depends on

x

, so every occurrence of

y

must be differentiated using the chain rule:

\frac{d}{dx}g(y) = g'(y) \cdot dy/dx

. Treating

y

as a constant is the most common error.

Dividing by zero ( $F_y = 0$ ). The formula

dy/dx = -F_x/F_y

fails when

F_y = 0

. In the implied vol context,

F_\sigma = \mathcal{V} = 0

occurs for options with zero vega — typically deep ITM/OTM options near expiry. At these points, the implied vol is ill-defined or infinitely sensitive to price changes.

Confusing "implicit" with "numerical." Implicit differentiation gives exact analytical formulas for derivatives (

dy/dx = -\Delta/\mathcal{V}

) even when the function

y(x)

has no closed form. You do not need to solve for

y

first. This is the power of the method: it bypasses the explicit solution entirely.

Assuming the implicit function exists globally. The implicit function theorem is local: it guarantees

y(x)

exists in a neighbourhood of the point. Globally, the implicit curve may fold, branch, or self-intersect. For implied vol, global uniqueness follows from the monotonicity of Black-Scholes in

\sigma

(vega is always positive), so the local result extends globally — but this requires a separate argument.

Where this goes next

Implicit differentiation connects to:

Partial Derivatives: The formula $\partial y/\partial x_i = -F_{x_i}/F_y$ is a ratio of partials. All implied vol sensitivities are computed this way.
Chain Rule: The method is powered by the chain rule applied to $y(x)$ inside $F$ .
Change of Variables: Implicit relationships between variables arise when performing substitutions in integrals and when changing probability measures.
Numerical Integration: Finding the implied value itself (e.g., $\sigma_{\text{impl}}$ ) requires a numerical root-finding algorithm (Newton-Raphson, bisection). Implicit differentiation tells you the sensitivity analytically; the root-finder gives you the level.

References

Stewart, J. (2008). Single Variable Calculus: Early Transcendentals (6th ed.). Thomson Brooks/Cole. Ch. 3 Section 3.5 (Implicit Differentiation) for the method and circle/tangent examples.