This paper develops a path-first theory using signatures as universal coordinates for deterministic paths, rough paths, jumps, and path-valued random variables. It introduces the geometricity defect principle, characterizes Itô vs Marcus lifts, and delivers finite-dimensional Hawkes closures for affine and scalar cases, along with a directional cross-area statistic. The results link shuffle, Marcus–Itô, and signature kernels to the geometry of the free nilpotent group, and establish identifiability and large-deviation implications for path-law representations.
速读内容
Geometricity defect and shuffle multiplicativity
The shuffle identity is the geometricity anchor; deviations come in two canonical second-order forms: bracket (convention defect) and covariance (averaging defect) .
Geometricity defect formalizes why non-geometric lifts (Itô) and averaged signatures typically fail to be group-like except in deterministic reduced paths .
Jump lifts: forward Itô vs Marcus
For pure-jump finite-variation paths, forward Itô lift equals iterated-sums; Marcus lift is Hoffman’s exponential image of it. This establishes a single Hopf-algebraic transition between discrete and continuous signatures .
Jump convention and discrete-to-continuous conversion are algebraically tied via the Hopf square: SigI(X) vs SigM(X) via Hoffman’s exponential Φ .
Finite-dimensional Hawkes closures: affine and scalar cases
Affine Hawkes clocks admit linear closures for truncated expected signatures after augmenting the state with weights; level-two Hawkes matrix A2 provides a closed ODE system for E[S^(N)] (Theorem 9.3, 9.4) .
For scalar exponential Hawkes, explicit state-weighted coordinates yield a finite-dimensional linear system, enabling local identifiability of (µ, α, β) from derivatives of E[NT] at small horizons .
Directional cross-area: two-channel Hawkes and lead–lag
The antisymmetric second-level cross-area captures the direction of cross-excitation and is leading-order sensitive to the two-channel structure; reversal of excitation flips the cross-area sign, serving as a signature-native lead–lag statistic .
Reversal, antipodes, and cross-area sign
Reversal is the antipode on the tensor Hopf algebra; the cross-area sign changes under time reversal when excitation roles swap, aligning with reversal experiments .
Moment thresholds, normalization, and large deviations
Raw expected signatures exhibit a stable-law moment threshold; when moments fail, normalized expected signatures (kernel-embedding) remain well-defined and injective for law characterization via MMD .
Large deviations of signatures follow by contraction from rough-path LDPs under suitable hypotheses .
思维导图
深度阅读
以下是对“ A General Theory of Paths: Signatures, Jump Lifts, and Expected Signatures of Self-Exciting Processes”一份极其详尽的分析性解读。文章将路径理论的核心工具(签名、乱动(rough paths)框架、跳跃处理、以及自激 Hawkes 过程的期望签名)系统性地结合,构建一个面向金融事件驱动建模的通用坐标体系。分析结构遵循原文的章节划分,力求逐段解读每个论点、数据点、假设与结论,并对文中的图表与表格给出深度解读与溯源标注。
元数据与概览
标题、作者与机构、日期与主题
标题:A General Theory of Paths: Signatures, Jump Lifts, and Expected Signatures of Self-Exciting Processes
作者与机构:Miquel Noguer i Alonso, Artificial Intelligence Finance Institute
2.12 Heavy-tailed 驱动下的矩阈值(12 Moment threshold for heavy-tailed drivers)
核心要点:在对称 γ-stable Levy 驱动的一维情形下,Geometric Marcus 签名的每一级签名存在严格的 moment 阈值:对任意整数 k≥1,签名的 k 级坐标存在有限的期望只有在 k < γ 的情形。也就是说,存在“完全的期望签名”的严格等级上界,超过此等级将不再存在全局矩方法的稳定性描述。该定理给出了整个框架在重尾驱动下的天然边界 。
论证要点:一维情形下,几何签名等同于总增量的张量指数;γ-stable 的随机变量在 k 阶矩存在的条件下才成立,这一性质直接决定了“全期望签名的存在性”与否。作者将其作为对框架的边界描述,明确指出“重尾驱动”下的局限性及需要的归一化策略(见 Proposition 12.2 的归一化期望签名与核的性质) 。
道路延伸: Proposition 12.2 给出归一化期望签名的存在性与核的特性,在重尾情形下仍可通过归一化后的特征 map 保证其可辨识性(Kernel 的特征映射为有界),从而实现对分布的统计表征,即使 raw 坐标不再具有良好矩性质。该结果为在极端分布情形下的辨识提供了“替代路径” 。