3 Pricing and Hedging
Bracket invariance, the two-parameter VRNC, the hedging decomposition, and the scaling law for fractional kernels. The central message: the kernel \(K\) enters the pricing structure itself—continuous and jump risk premia cannot be chosen independently [1, Theorem 5.1].
3.1 Bracket Invariance
BracketData def:VolterraKernel The continuous bracket of the partial process: \(\langle V_T^c \rangle _t = \sigma _L^2 \int _0^t K(T,s)^2\, ds\). Carries \(\sigma _L\), the kernel-squared integral, and the bracket function. The key property: the bracket is deterministic (depends only on \(K\) and \(\sigma _L\), not on \(\omega \)) [1, Theorem 5.3(a)].
CombinedEnergySpace.bracket_deterministic def:BracketData \(\exists c,\; \langle V_T^c \rangle _t = c\). The bracket is a real number at each \(t\), not a random variable. The witness is \(\sigma _L^2 \cdot \int _0^t K(T,s)^2\, ds\) from bracket_eq. ALGEBRAIC.
Witness from bracket_eq.
CombinedEnergySpace.bracket_measure_invariant def:BracketData \(\langle V_T^c \rangle _t = \sigma _L^2 \cdot \int _0^t K(T,s)^2\, ds\). The RHS has no \(\omega \)-dependence: deterministic, hence identical under \(\mathbb {P}\) and \(\mathbb {Q}\). Neither \(\sigma _L\) nor \(K\) changes under an absolutely continuous measure change [1, Theorem 5.3]. ALGEBRAIC.
Immediate from bracket_eq.
CombinedEnergySpace.roughness_invariant thm:bracket_measure_invariant Both time points \(t_1, t_2\) have the deterministic formula. The Hurst parameter \(H\)—extracted from the log-log slope of \(\langle V_T^c \rangle _t\)—is therefore the same under any equivalent measure. The roughness estimated from historical data is the roughness that governs risk-neutral dynamics. This is a structural feature of the energy space, not a modelling assumption [1, Corollary 5.4]. ALGEBRAIC.
Two applications of bracket_eq.
3.2 The Two-Parameter VRNC
GirsanovData The market prices of risk: continuous \(\theta ^W(t)\) (Girsanov drift removal) and jump \(\theta ^J(t,z)\) (Lévy–Girsanov exponential tilting [2, Theorem 9.12]), diffusion coefficient \(\sigma _L\), and jump drift contribution \(\Phi ^J(t) = \int z(e^{\theta ^J(t,z)} - 1)\, \nu (dz)\).
CombinedEnergySpace.effectiveDriftRate def:GirsanovData \(\Psi _t := -\sigma _L \theta _t^W + \Phi _t^J\). The effective drift rate aggregates the Brownian and jump risk premia into a single quantity. The kernel \(K\) couples \(\Psi \) across time via a Volterra integral equation [1, eq. (5.3)]. Defined, not assumed.
VRNCData def:GirsanovData, def:VolterraKernel Extends Girsanov data with the kernel, drift rate \(\mu _t\), risk-free rate \(r\), direct Lévy exposure \(\lambda _t\), memory factor loading \(\beta _t\), and the drift functional \(B_t^\theta = \int _0^t K(t,s)\Psi (s)\, ds\). The VRNC requires solving a Volterra integral equation: \(\mu _t - r = -[\lambda _t + \beta _t K(t,t)]\Psi _t - \beta _t \int _0^t \partial _t K(t,s)\Psi _s\, ds\) [1, Theorem 5.1].
CombinedEnergySpace.VRNC_determines_Psi def:effectiveDriftRate \(\Psi _t = -\sigma _L \theta _t^W + \Phi _t^J\). ALGEBRAIC: unfolds the definition.
rfl.
CombinedEnergySpace.VRNC_continuous_recovery def:effectiveDriftRate When \(\nu = 0\): \(\Psi = -\sigma _L \theta ^W\). The jump drift vanishes and the constraint reduces to the continuous Volterra risk-neutral condition of [1, Corollary 5.2]. ALGEBRAIC: substitutes \(\Phi ^J = 0\).
simp [effectiveDriftRate, hno_jumps].
CombinedEnergySpace.continuous_from_jump_choice def:effectiveDriftRate \(\theta ^W_t = (\Phi ^J_t - \Psi ) / \sigma _L\). Given a choice of jump pricing (\(\Phi ^J\)), the continuous market price of risk is determined. This is the measure-selection degree of freedom: one equation in two unknowns [1, Proposition 5.5]. ALGEBRAIC: field_simp + linarith.
Unfold, field_simp, linarith.
3.3 Scaling Law for Fractional Kernels
DefectScalingData def:NuMomentStructure Hurst parameter \(H {\gt} 0\), terminal time \(T {\gt} 0\), moment structure \(\nu \), and the convergence condition \(2p(H - 1/2) + 1 {\gt} 0\) (i.e., \(H {\gt} (p-1)/(2p)\)). Under these conditions, the fractional kernel integral \(\int _0^T (T-t)^{2p(H-1/2)}\, dt\) converges [1, Assumption HP].
CombinedEnergySpace.scalingExponent def:DefectScalingData \(2H - 1 + 1/p\). This exponent governs how the defect \(L^p\)-norm scales with terminal time \(T\). For \(H = 1/2\) (Markov), the exponent is \(1/p\) (standard Lévy defect); for \(H {\lt} 1/2\) (rough), the exponent is smaller, concentrating the defect on recent jumps [1, Theorem 6.1].
CombinedEnergySpace.scalingBound_pos def:DefectScalingData, def:scalingExponent \(0 {\lt} T^{2H-1+1/p} \cdot (c_\gamma /(2p-\gamma ))^{1/p}\). The scaling bound is positive: roughness and jumps interact multiplicatively in the defect norm [1, eq. (6.4)]. CONCRETE: from rpow_pos_of_pos and div_pos.
mul_pos with rpow_pos_of_pos.
CombinedEnergySpace.markov_scaling_exponent def:scalingExponent \(H = 1/2 \implies \) exponent \(= 1/p\). Recovers the standard Lévy defect scaling (no kernel memory effect) [1, Corollary 6.2]. ALGEBRAIC.
simp [scalingExponent, hH].
CombinedEnergySpace.rough_exponent_less def:scalingExponent \(H {\lt} 1/2 \implies \) exponent \({\lt} 1/p\). The defect concentrates on recent jumps for rough kernels, but the kernel singularity amplifies each jump—the two effects interact multiplicatively [1, Corollary 6.3]. ALGEBRAIC.
simp [scalingExponent]; linarith.
CombinedEnergySpace.convergence_condition_2p The kernel integral \(\int _0^T (T-t)^{2p(H-1/2)}\, dt\) converges iff \(H {\gt} (p-1)/(2p)\) [1, Assumption HP]. CONCRETE: via nlinarith.
Unfold, split into two directions, each closed by nlinarith.
CombinedEnergySpace.quadratic_defect_exact_volterra \((K \cdot z)^2 = K^2 \cdot z^2\). For \(f(x) = x^2\), the defect integrand is exact and deterministic: \(g_f^K(t,z) = K(T,t)^2 z^2\) (no Taylor approximation—the identity \((x+y)^2 - x^2 - 2xy = y^2\) is algebraic). This gives two-sided BDG bounds and sharpness of the scaling law [1, Proposition 6.4]. ALGEBRAIC: ring.
ring.
CombinedEnergySpace.varianceSwapIsometry_pos \(0 {\lt} T^{4H-1}/(4H-1) \cdot \nu _4\). For the quadratic payoff, the Itô isometry gives the exact \(L^2\)-norm of the defect: \(\mathbb {E}[|\Gamma _V^{x^2}|^2] = \int _0^T K(T,t)^4\, dt \cdot \nu _4\) [1, Proposition 6.4, eq. (6.8)]. CONCRETE.
mul_pos with div_pos and rpow_pos_of_pos.
3.4 Hedging Decomposition
CombinedEnergySpace.HedgingDecomposition def:CombinedEnergySpace The three-level hedge [1, Proposition 7.2]: discounted claim \(= \mathbb {E}^\mathbb {Q}[\text{claim}] + \delta _W(\text{brownian}) + \tilde{\delta }_J(\nu \text{-singular}) + \tilde{\delta }_J(\text{defect})\). Level 1: stock + variance swap absorb all continuous risk. Level 2: a Lévy driver proxy absorbs the \(\nu \)-singular (first-order) jump exposure. Level 3: options on jump variation absorb the \(\nu \)-regular defect. The residual after all three levels is zero—the boundary of hedgeability is the defect \(\Gamma _V^f\) itself.
CombinedEnergySpace.continuous_hedge_absorbs_brownian def:HedgingDecomposition \(\text{claim} - \text{fair\_ value} - \delta _W(\text{brownian}) = \tilde{\delta }_J(\text{singular}) + \tilde{\delta }_J(\text{defect})\). After delta-hedging with the stock and vega-hedging with a variance swap, the residual is purely jump-driven. The continuous hedge absorbs ALL Brownian risk; the remaining risk is entirely in the Poisson integrals [1, Proposition 7.2(a)]. Same algebraic pattern as product_rule_with_jump_defect in BanachLevyComplete.lean [4]. ALGEBRAIC.
Rewrite with decomp_identity, close with abel.
CombinedEnergySpace.perfect_hedging_without_jumps def:HedgingDecomposition When the defect integrand is zero: \(\tilde{\delta }_J(\text{defect}) = 0\). In pure rough volatility models (\(\nu = 0\)), the defect vanishes and hedging is perfect—recovering the El Euch–Rosenbaum (2019) result as a special case [1, Corollary 7.3]. ALGEBRAIC: from map_zero.
Rewrite with zero hypothesis, apply map_zero.
3.5 Master Theorem
CombinedEnergySpace.complete_volterra_levy_framework thm:D_VL_split, thm:two_regime_decomposition, thm:defect_vanishes_without_jumps, thm:VRNC_continuous_recovery, thm:markov_scaling_exponent, thm:rough_exponent_less, thm:quadratic_defect_exact, thm:perfect_hedging_without_jumps Eight-conjunct master theorem packaging the full pipeline of [1]: (1) the operator derivative splits, (2) the Poisson integral decomposes into \(\nu \)-singular \(+\) \(\nu \)-regular, (3) the defect vanishes without jumps, (4) the VRNC reduces to the continuous case when \(\nu = 0\), (5) the Markov case \(H = 1/2\) gives exponent \(1/p\), (6) roughness amplifies the defect for \(H {\lt} 1/2\), (7) the quadratic defect is algebraically exact, (8) perfect hedging holds in pure rough vol. Each conjunct is an algebraic consequence of the structure fields; the Volterra–Lévy Itô decomposition (from which they all flow) is itself a theorem.
Each conjunct delegates to its individual theorem.