Paper: arXiv 2411.02375

Authors: Kentaro Hoshisashi, Carolyn E. Phelan, Paolo Barucca

Abstract

Calibrating the time-dependent Implied Volatility Surface (IVS) using sparse market data is an essential challenge in computational finance, particularly for real-time applications. This task requires not only fitting market data but also satisfying a specified partial differential equation (PDE) and no-arbitrage conditions modelled by differential inequalities. This paper proposes a novel Physics-Informed Neural Networks (PINNs) approach called Whack-a-mole Online Learning (WamOL) to address this multi-objective optimisation problem. WamOL integrates self-adaptive and auto-balancing processes for each loss term, efficiently reweighting objective functions to ensure smooth surface fitting while adhering to PDE and no-arbitrage constraints and updating for intraday predictions. In our experiments, WamOL demonstrates superior performance in calibrating intraday IVS from uneven and sparse market data, effectively capturing the dynamic evolution of option prices and associated risk profiles. This approach offers an efficient solution for intraday IVS calibration, extending PINNs applications and providing a method for real-time financial modelling.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents advanced mathematical formalisms including PDEs, differential inequalities, and specialized loss functions (PINNs/DCPINNs) for a complex calibration problem. It demonstrates empirical rigor through backtesting on real market data and evaluating computational efficiency, though the absence of publicly available code or datasets slightly limits full replicability.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Intraday IVS Calibration<br>from Sparse Market Data"] --> B["Methodology: WamOL Framework"]
  
  B --> C{"Data & Inputs:<br>Market Data +<br>PDE/No-Arbitrage Constraints"}
  C --> D["Computational Process:<br>Self-Adaptive & Auto-Balancing<br>Multi-Objective Loss Optimization"]
  D --> E["Key Findings:<br>Superior IVS Calibration<br>Dynamic Risk Profiling<br>Real-time Applicability"]

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 Whack-a-mole Online Learning (WamOL), a calibration algorithm that combines self‑adaptive and loss‑balancing techniques within PINNs to calibrate time‑dependent implied volatility surfaces in real‑time.

Source sentence
We introduce the Whack-a-mole Online Learning (WamOL) algorithm, which combines self-adaptive and loss-balancing learning techniques to calibrate time-dependent IVS using PINNs.

Assumes 4

  • Option prices must satisfy no‑arbitrage inequalities on first and second derivatives. model
    Source sentence
    These principles state that market prices should not allow for guaranteed returns exceeding the risk‑free rate. We consider the necessary and sufficient conditions for no‑arbitrage in [2, 11], which are represented as non‑strict inequalities involving several first and second derivatives, −1
  • Market data are sparse, uneven, and arrive intraday, requiring online learning updates. data
    Source sentence
    WamOL leverages online learning techniques to facilitate real‑time intraday calibration, enabling the model to adapt swiftly to evolving market conditions.
  • Only out‑of‑the‑money (OTM) S&P 500 options are used, with forward prices derived via put‑call parity linear regression. market
    Source sentence
    The data source provided spot prices, selecting out‑of‑the‑money (OTM) options, and we derived forward prices using linear regression based on put‑call parity.
  • Training and validation errors are measured on 15‑minute intervals throughout the trading day. execution
    Source sentence
    calibration time (solid lines) and validated using the calibrated model on data from the rest of the day (dotted lines) at 15‑minute intervals

Data

  • Intraday S&P 500 options dataset · CBOE DataShop options commercial
    2022‑10‑01 to 2023‑09‑30 (248 business d · S&P 500 options (OTM)
    Source sentence
    We conducted backtesting on extensive intraday options data for WamOL using a dataset of S&P 500 options. The historical data comprised intraday traded prices of S&P 500 options for 248 business days from October 1, 2022, to September 30, 2023. This dataset contains approximately 500,000 data points per day with 1 ms p

Recognised sources: Cboe

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • WamOL consistently achieves lower calibration and validation errors than competing models (MLP, standard PINNs) across training and prediction phases. performance
    Source sentence
    WamOL consistently achieves lower calibration and validation errors than the other models overall in both training and predictions.
  • WamOL exhibits faster and more scalable computation times than baseline models, with sub‑linear growth as dataset size increases. performance
    Source sentence
    Table 3 presents the computation times required by WamOL and the baseline models for calibrating the IVS on datasets of varying sizes. The computational efficiency of WamOL is evident from its ability to handle these complex optimisation challenges without significant overhead

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 28 related papers). Coverage is partial, so this is neither replication nor refutation.

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 (44)

  1. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019)
  2. Masterclass with deutsche bank. the vol smile problem (2002)
  3. The volatility surface: a practitioner’s guide (2011)
  4. The application of improved physics-informed neural network (IPINN) method in finance (2022)
  5. No-Arbitrage Deep Calibration for Volatility Smile and Skewness — → 2310.16703
  6. Volatility is (mostly) path-dependent (2023)
  7. An expert’s guide to training physics-informed neural networks (2023) — link
  8. Implied volatility: Market models (2010)
and 36 more
  1. Pricing with a smile (1994)
  2. A note on sufficient conditions for no arbitrage (2005)
  3. Nonparametric option pricing under shape restrictions (2003)
  4. Arbitrage free implied volatility surfaces (2010)
  5. Semi-nonparametric estimation of the call-option price surface under strike and time-to-expiry no-arbitrage constraints (2015)
  6. Learning representations by back-propagating errors (1986)
  7. Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks (1990)
  8. Managing smile risk (2002)
  9. Experiment Tracking with Weights and Biases (2020) — link
  10. Respecting causality is all you need for training physics-informed neural networks (2022) — link
  11. Self-adaptive physics-informed neural networks using a soft attention mechanism (2020) — link
  12. The pricing of options and corporate liabilities (1973)
  13. A neural network-based framework for financial model calibration (2019)
  14. FuNVol: A Multi-Asset Implied Volatility Market Simulator using Functional Principal Components and Neural SDEs — → 2303.00859
  15. Variational autoencoders: A hands-off approach to volatility (2022)
  16. Option valuation under no-arbitrage constraints with neural networks (2021)
  17. Simulation of arbitrage-free implied volatility surfaces (2022)
  18. Deep learning calibration of option pricing models: some pitfalls and solutions (2019) — link
  19. Deep smoothing of the implied volatility surface (2020)
  20. On deep calibration of (rough) stochastic volatility models (2019) — link
  21. An artificial neural network representation of the SABR stochastic volatility model (2020)
  22. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
  23. JAX : composable transformations of P ython+ N um P y programs (2018) — link
  24. F lax: A neural network library and ecosystem for JAX (2023) — link
  25. The pricing of commodity contracts (1976)
  26. Adam: A method for stochastic optimization (2014) — link
  27. Physics-informed neural networks with hard constraints for inverse design (2021)
  28. Volatility interpolation (2011)
  29. Dynamics ofimplied volatility surfaces (2002)
  30. Deterministicimplied volatility models (2002)
  31. Arbitrage-free neural-SDE market models (2023)
  32. Arbitrage-free implied volatility surface generation with variational autoencoders (2023)
  33. Implied volatility: Statics, dynamics, and probabilistic interpretation (2005)
  34. Whack-a-Mole Learning: Physics-Informed Deep Calibration for Implied Volatility Surface (EasyChair, 2024)
  35. DARE: The Deep Adaptive Regulator for Control of Uncertain Continuous-Time Systems
  36. Physics Informed Neural Network for Option Pricing (2023) — link

2 of 44 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.