Paper: arXiv 2511.05519
Authors: Sina Kazemian, Ghazal Farhani, Amirhessam Yazdi
Abstract
We present an uncertainty-aware, physics-informed neural network (PINN) for option pricing that solves the Black–Scholes (BS) partial differential equation (PDE) as a mesh-free, global surrogate over \((S,t)\). The model embeds the BS operator and boundary/terminal conditions in a residual-based objective and requires no labeled prices. For American options, early exercise is handled via an obstacle-style relaxation while retaining the BS residual in the continuation region. To quantify \emph{“epistemic”} uncertainty, we introduce an anchored-ensemble fine-tuning stage (AT–PINN) that regularizes each model toward a sampled anchor and yields prediction bands alongside point estimates. On European calls/puts, the approach attains low errors (e.g., MAE \(\sim 5\times10^{"-2"}\), RMSE \(\sim 7\times10^{"-2"}\), explained variance \(\approx 0.999\) in representative settings) and tracks ground truth closely across strikes and maturities. For American puts, the method remains accurate (MAE/RMSE on the order of \(10^{"-1"}\) with EV \(\approx 0.999\)) and does not exhibit the error accumulation associated with time-marching schemes. Against data-driven baselines (ANN, RNN) and a Kolmogorov–Arnold FINN variant (KAN), our PINN matches or outperforms on accuracy while training more stably; anchored ensembles provide uncertainty bands that align with observed error scales. We discuss design choices (loss balancing, sampling near the payoff kink), limitations, and extensions to higher-dimensional BS settings and alternative dynamics.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper is highly mathematical, incorporating advanced PDE theory, residual-based neural network training, and anchored ensembles for uncertainty quantification; while it reports strong empirical results (low MAE/RMSE, explained variance) and benchmarks against ANN/RNN/KAN baselines, it lacks real market data and focuses on synthetic/simulated settings, limiting its practical backtest readiness.
Research Flowchart
flowchart TD
A["Research Goal<br>Mesh-free solution to Black-Scholes PDE<br>with epistemic uncertainty quantification"] --> B["Methodology: Physics-Informed PINN"]
B --> C["Input: Unsupervised data<br>Random sample from domain S,t"]
B --> D["Core Computational Process<br>Residual minimization of BS PDE + boundary conditions"]
D --> E["Adaptation for American Options<br>Obstacle-style relaxation"]
D --> F["UQ: Anchored-Ensemble Fine-tuning<br>AT-PINN for prediction bands"]
C & E & F --> G["Outcomes"]
subgraph G ["Results"]
G1["High Accuracy<br>MAE ~5e-2, EV ~0.999"]
G2["Stable Performance<br>No error accumulation vs time-marching"]
G3["Reliable Uncertainty<br>Bands align with error scales"]
end
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes an uncertainty‑aware physics‑informed neural network (PINN) with an anchored‑ensemble fine‑tuning stage for solving the Black‑Scholes PDE and pricing options.
Source sentence
We present an uncertainty-aware, physics-informed neural network (PINN) for option pricing that solves the Black–Scholes (BS) partial differential equation (PDE) as a mesh-free, global surrogate over (S,t).
Assumes 3
- The Black‑Scholes dynamics with constant risk‑free rate and volatility (no‑dividend) are assumed for the pricing PDE. model
Source sentence
To enforce no-arbitrage under a non-dividend assumption, labels y are clipped to instrument-specific bounds—American put: L = max(K −S, 0), U = K; European call: L = max(S −Ke^{−rτ}, 0), U = S.
- Early‑exercise feature of American options is handled via an obstacle‑style relaxation while keeping the BS residual in the continuation region. model
Source sentence
For American options, we handle early exercise via an obstacle‑style relaxation while retaining the BS residual in the continuation region.
- The PINN is trained without any labeled option price data, relying solely on the PDE residual and boundary/terminal conditions. data
Source sentence
The model embeds the BS operator and boundary/terminal conditions in a residual‑based objective and requires no labeled prices.
Data
- Yahoo Finance option chain dataset options public
Source sentence
We build supervised datasets for American puts and European calls on the Yahoo Finance tickers AAPL, MSFT, GOOGL, AMZN, META, F, T, NOK, CCL, SOFI, SNAP, NIO, AAL, PBR, RIVN.
Costs
Unclear - sources disagree
Source sentence (the authors' words)
... transaction costs, stochastic volatility, and jumps.
Finds 4
- The anchored PINN achieves low pricing errors for European options (MAE ≈5×10⁻², RMSE ≈7×10⁻²) with explained variance ≈0.999. predictive-accuracy
Source sentence
On European calls/puts, the approach attains low errors (e.g., MAE
- For American puts the method remains accurate (MAE/RMSE on the order of 10⁻¹ with EV ≈0.999) and avoids error accumulation of time‑marching schemes. predictive-accuracy
Source sentence
with EV ≈0.999) and avoids the error accumulation typical of time-marching sc
- Across four parameter sets, the PINN‑anchor outperforms ANN and RNN and matches or slightly exceeds the KAN baseline, while training more stably. performance
Source sentence
Consistent with the tables, PINN‑anchor and KAN generally outperform ANN/RNN, with visibly smaller deviations near the strike. Between PINN‑anchor and KAN, accuracy is similar across most parameter sets; however, KAN trains markedly slower (about 10×) and exhibits
- The model provides calibrated epistemic uncertainty bands that respect no‑arbitrage bounds and reflect model uncertainty. predictive-accuracy
Source sentence
predictions are returned with calibrated uncertainty bands that (i) respect no-arbitrage bounds through our bounded-output mapping and (ii) reflect model uncertainty arising from model behaviour.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 47 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (58)
- Google photos labeled black people’gorillas’ (2015)
- PhD thesis: Uncertainty in deep learning (2016)
- A practical Bayesian framework for backpropagation networks (1992)
- Uncertainty in neural networks: Approximately bayesian ensembling (2020)
- Invariance of weight distributions in rectified MLPs (2018)
- Option Pricing: A Simplified Approach (1979) — link
- Options: A Monte Carlo Approach (1977) — link
- A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (1993) — link
and 50 more
- Option Pricing When Underlying Stock Returns Are Discontinuous (1976) — link
- The Pricing of Options and Corporate Liabilities (1973) — link
- Theory of Rational Option Pricing (1973) — link
- Options, Futures, and Other Derivatives (2022)
- Arbitrage Theory in Continuous Time (2020) — link
- The Volatility Surface: A Practitioner’s Guide (2012) — link
- Moment Explosions and Long-Term Behavior of Affine Stochastic Volatility Models (2011) — link
- Multilevel Monte Carlo Methods (2015) — link
- A Jump-Diffusion Model for Option Pricing (2002) — link
- Financial Modelling with Jump Processes (2004)
- Binomial Tree Method for Option Pricing: Discrete Cosine Transform Approach (2022) — link
- Black–Scholes 50 Years Later: Has the Outperformance of Passive Option Strategies Finally Faded? (2024) — link
- Binomial Models for Option Valuation—Examining and Improving Convergence (1996) — link
- Pricing Financial Instruments: The Finite Difference Method (2000)
- Finite Difference Methods in Financial Engineering: A Partial Differential Equation Approach (2006) — link
- Convergence Remedies for Non-Smooth Payoffs in Option Pricing (2003) — link
- Multigrid for American Option Pricing with Stochastic Volatility (1999) — link
- Convergence of Numerical Methods for Valuing Path-Dependent Options Using Interpolation (2002) — link
- Momentum diminishes the effect of spectral bias in physics-informed neural networks (2022) — link
- A simple remedy for failure modes in physics informed neural networks (2025)
- The application of improved physics-informed neural network (IPINN) method in finance (2022)
- Physics-informed neural networks (pinns) in finance (2023)
- Option pricing in the Heston model with physics inspired neural networks (2024)
- Predicting option prices: From the Black-Scholes model to machine learning methods (2025)
- A comparative analysis of option pricing models: Black–Scholes, Bachelier, and artificial neural networks (2025)
- ANN model for Call Options Pricing using S&P 100 Market Data (2024)
- The AI Black-Scholes: Finance-Informed Neural Network — → 2412.12213
- Kolmogorov–Arnold Finance-Informed Neural Network in Option Pricing (2024)
- A Black–Scholes user’s guide to the Bachelier model (2022)
- Pricing options with a new hybrid neural network model (2024)
- Deep Learning in Financial Modeling: Predicting European Put Option Prices with Neural Networks (2025)
- Option pricing with deep learning: a long short-term memory approach (2025)
- Option pricing using lstm: A perspective of realized skewness (2023)
- Option hedging using LSTM-RNN: an empirical analysis (2021)
- Merged LSTM-MLP for option valuation (2025)
- Physics-informed neural networks: Minimizing residual loss with wide networks and effective activations (2024) — link
- Physics informed neural network for option pricing (2023) — link
- Valuing American Options by Simulation: A Simple Least-Squares Approach (2001) — link
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019) — link
- Physics-informed machine learning (2021) — link
- On Calibration of Modern Neural Networks (2017)
- Can You Trust Your Model’s Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift (2019)
- What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision? (2017)
- A Bayesian neural network approach for tropospheric temperature retrievals from a lidar instrument (2023)
- Optimal estimation method retrievals of stratospheric ozone profiles from a DIAL (2019)
- Applied optimal estimation (1974)
- Inverse problem theory and methods for model parameter estimation (2005)
- Influence of higher-order electron-phonon interaction on the electron-related lattice thermal properties of two-dimensional Dirac crystals (2024)
- Dynamic dielectric function and phonon self-energy from electrons strongly correlated with acoustic phonons in 2D Dirac crystals (2023)
- Thermal conductivity of thin film-substrate systems from two-side scanning photothermal deflection measurements: Theoretical model and validation (2019)
1 of 58 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.