Mathematical Foundations of Quantum Pricing Theory

Abstract: Let $M$ be a von Neumann algebra and let $(N_t)<em>{t\in[0,T]}$ be an increasing family of abelian von Neumann subalgebras encoding a (classical) information flow. Fix a faithful normal state $\varphi</em>ρ$ on $M$ and assume a filtration of normal $\varphi_ρ$-preserving conditional expectations $E_t:M\to N_t$ satisfying the tower property. For self-adjoint observables affiliated with $M$, we introduce a truncation-stable notion of $(N_t,E_t)$-martingales via bounded functional-calculus cutoffs $f_n$, and formulate a \emph{Local Informational Efficiency Principle} requiring symmetrically discounted traded prices to be martingales in this localized sense. Assuming the existence of a pricing state $\varphi<sup>\star$ and a compatible family of normal $\varphi<sup>\star$-preserving conditional expectations $(E_t<sup>\star)$, we define for bounded terminal payoffs $X\in M_T$ the dynamic pricing operator [ Πt(X):=B_t^{1/2}\,E_t^\star!\bigl(B_T^{-1/2}XB_T^{-1/2}\bigr)\,B_t^{1/2}, ] where $(B_t)$ is a strictly positive numéraire adapted to $(N_t)$. We prove that $(Π_t)</em>{t\in[0,T]}$ is normal, completely positive, unital, $N_t$-bimodular, and time-consistent; in the commutative reduction it coincides with classical risk-neutral valuation by conditional expectation. We further develop an $L<sup>2(M,\varphi_ρ)$-prediction theory: $E_t$ acts as the $L<sup>2$-optimal predictor and yields a canonical innovation decomposition. For differentiable parametric families of normal states, we introduce an operator-valued Fisher information relative to $(N_t)$ and derive a noncommutative Cramér--Rao inequality giving a quantitative lower bound on conditional mean-square prediction error under information constraints; the bound is computed explicitly for compound Poisson lattice-jump models under the risk-neutral constraint $\sum_αγ_α(e<sup>{αΔx}-1)=r$.