Back
2026ECO001.pdf
Doctoral thesis

On applications of kernel methods to financial econometrics

  • 2026

PhD: Università della Svizzera italiana

Kernel methods
Financial econometrics
Portfolio optimization
Stochastic discount factor
Options
Positivity/monotonicity/convexity tests
Mean-variance optimization
Statistical learning
English This doctoral dissertation develops kernel-based, nonparametric statistical learning methods for financial econometrics, with emphasis on high-dimensional, nonlinear, and dependent data. It centers on scenario extraction and mean-variance decision rules with economically motivated shape properties. The study proposes two algorithms that summarize large panel datasets into representative scenarios that retain moment information of the empirical samples. One algorithm builds covariance scenarios for use in modeling, while the other formulates scenario selection as a sparse compressive-sensing problem, employing orthogonal matching pursuit combined with a pivoted Cholesky procedure. These approaches substantially reduce the number of required scenarios for modeling purposes, thereby easing computational burden and improving interpretability. In addition, the dissertation introduces a kernel-based framework for nonparametric mean-variance optimization and for testing shape constraints on optimal decision rules. It establishes both asymptotic and finite-sample statistical guarantees and develops a joint Wald-type test for shape features, together with a computationally scalable method suitable for large datasets. The proposed tools are applied in three main domains: portfolio optimization using Fama--French excess returns, explainable artificial intelligence via sparse kernel-based surrogate models, and empirical asset pricing through nonparametric estimation of the stochastic discount factor. Broadly, the dissertation provides effective solutions to central problems in financial econometrics and opens new directions in portfolio analytics, interpretability, and nonparametric asset pricing.
Collections
Language
  • English
Classification
Economics
License
License undefined
Open access status
green
Identifiers
Persistent URL
https://n2t.net/ark:/12658/srd1334463
Statistics
Document views: 39 File downloads:
  • 2026ECO001.pdf: 54