The pricing of barrier options and the computation of their sensitivities (Greeks) represent a challenging problem in quantitative finance due to payoff discontinuities and numerical instabilities near the barrier. Classical numerical approaches based on finite differences or Monte Carlo methods require repeated solutions of partial differential equations, leading to significant computational costs in practical applications. This thesis investigates the use of Neural Operators, with a particular focus on Physics-Informed Fourier Neural Operators (PINO), for the pricing of barrier options and the simultaneous computation of Delta and Vega. By learning the pricing operator directly in function space, neural operator models offer discretization-invariant representations and enable efficient inference once trained. The work presents the theoretical foundations of Fourier-based neural operators and their physics-informed extensions, together with an end-to-end implementation framework including dataset construction, model architectures, and training strategies. An experimental evaluation is conducted on synthetic datasets designed to reflect realistic option pricing scenarios, with the aim of assessing accuracy, robustness, and computational performance. Overall, this thesis explores the applicability of physics-informed neural operators as an alternative tool for barrier option pricing and Greek computation in quantitative finance.
La determinazione del prezzo delle opzioni barriera e il calcolo delle loro sensibilità (Greeks) rappresentano un problema complesso nella finanza quantitativa a causa delle discontinuità nel payoff e delle instabilità numeriche in prossimità della barriera. Gli approcci numerici classici basati su differenze finite o su metodi Monte Carlo richiedono la risoluzione ripetuta di equazioni alle derivate parziali, con conseguenti costi computazionali significativi nelle applicazioni pratiche. Questa tesi indaga l’utilizzo dei Neural Operators, con particolare attenzione ai Physics-Informed Fourier Neural Operators (PINO), per il pricing delle opzioni barriera e il calcolo simultaneo di Delta e Vega. Apprendendo direttamente l’operatore di pricing nello spazio delle funzioni, i modelli di neural operator offrono rappresentazioni invarianti rispetto alla discretizzazione e consentono un’inferenza efficiente una volta completato l’addestramento. Il lavoro presenta i fondamenti teorici dei neural operator basati su Fourier e delle loro estensioni physics-informed, insieme a un framework di implementazione end-to-end che include la costruzione del dataset, le architetture dei modelli e le strategie di training. Viene inoltre condotta una valutazione sperimentale su dataset sintetici progettati per riflettere scenari realistici di pricing delle opzioni, con l’obiettivo di valutare accuratezza, robustezza e prestazioni computazionali. Nel complesso, questa tesi esplora l’applicabilità dei neural operator informati dalla fisica come strumento alternativo per il pricing delle opzioni barriera e il calcolo delle Greeks nella finanza quantitativa.
Fast Pricing and Greeks Computation for Barrier Options via Physics-Informed Fourier Neural Operators
TIBONI, GABRIELE
2025/2026
Abstract
The pricing of barrier options and the computation of their sensitivities (Greeks) represent a challenging problem in quantitative finance due to payoff discontinuities and numerical instabilities near the barrier. Classical numerical approaches based on finite differences or Monte Carlo methods require repeated solutions of partial differential equations, leading to significant computational costs in practical applications. This thesis investigates the use of Neural Operators, with a particular focus on Physics-Informed Fourier Neural Operators (PINO), for the pricing of barrier options and the simultaneous computation of Delta and Vega. By learning the pricing operator directly in function space, neural operator models offer discretization-invariant representations and enable efficient inference once trained. The work presents the theoretical foundations of Fourier-based neural operators and their physics-informed extensions, together with an end-to-end implementation framework including dataset construction, model architectures, and training strategies. An experimental evaluation is conducted on synthetic datasets designed to reflect realistic option pricing scenarios, with the aim of assessing accuracy, robustness, and computational performance. Overall, this thesis explores the applicability of physics-informed neural operators as an alternative tool for barrier option pricing and Greek computation in quantitative finance.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tiboni_Gabriele.pdf
accesso aperto
Dimensione
1.69 MB
Formato
Adobe PDF
|
1.69 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/106575