Call-price density recovery (source code)

= Call-price density recovery
{title2=$C''(K)=Df_Q(K)$}

= Breeden-Litzenberger density formula
{c}
{synonym}

For an integrable terminal <stock> price and a deterministic positive <discount factor> $D$, a <European call option> curve has the form $C(K)=D\mathbb E_Q[(S_T-K)^+]$. Bounded difference quotients and <dominated convergence> give its one-sided <derivatives>:
$$
C'_+(K)=-D Q(S_T>K),\qquad C'_-(K)=-D Q(S_T\geq K).
$$
Thus the slope jump at a strike is $D Q(S_T=K)$. If the terminal <probability distribution> has a continuous <probability density function> $f_Q$, differentiation gives
$$
C''(K)=D f_Q(K).
$$
The call curve's <convexity> makes this curvature nonnegative. The original construction obtains <Arrow state prices> from increasingly narrow call spreads.<Prices of State-Contingent Claims Implicit in Option Prices by Douglas T. Breeden and Robert H. Litzenberger>[sectionII, printed pp625–627, equation(2)]{ref}

A continuous distribution need not have a density: a singular continuous law is a counterexample. In general the distributional second <derivative> is the measure $D Q(S_T\in dK)$. For a deterministic terminal value $s_0$, the curve $D(s_0-K)^+$ has a kink and second derivative $D\delta_{s_0}$, rather than an ordinary density function. The entire strike curve therefore records atoms as well as any density. Finitely many observed strikes usually leave many compatible terminal laws.