Paper: arXiv 2307.07657

Authors: Serena Della Corte, Laurens Van Mieghem, Antonis Papapantoleon, Jonas Papazoglou-Hennig

Abstract

We consider the supervised learning problem of learning the price of an option or the implied volatility given appropriate input data (model parameters) and corresponding output data (option prices or implied volatilities). The majority of articles in this literature considers a (plain) feed forward neural network architecture in order to connect the neurons used for learning the function mapping inputs to outputs. In this article, motivated by methods in image classification and recent advances in machine learning methods for PDEs, we investigate empirically whether and how the choice of network architecture affects the accuracy and training time of a machine learning algorithm. We find that the generalized highway network architecture achieves the best performance, when considering the mean squared error and the training time as criteria, within the considered parameter budgets for the Black-Scholes and Heston option pricing problems. Considering the transformed implied volatility problem, a simplified DGM variant achieves the lowest error among the tested architectures. We also carry out a capacity-normalised comparison for completeness, where all architectures are evaluated with an equal number of parameters. Finally, for the implied volatility problem, we additionally include experiments using real market data.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced neural network architectures (highway, DGM) and references PDE methods, indicating high math complexity. It conducts structured experiments with real market data and performance metrics like MSE and training time, demonstrating high empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Investigate impact of <br>neural network architecture <br>on option pricing"] --> B["Methodology: <br>Train & compare architectures <br>under fixed parameter budget"]
  B --> C["Data: <br>Black-Scholes & Heston models <br>+ Real Market Data"]
  C --> D["Computational Process: <br>Supervised learning of <br>Price/Implied Volatility"]
  D --> E["Key Findings:"]
  E --> F["Generalized Highway <br>Network: Best for <br>Option Pricing"]
  E --> G["Simplified DGM: <br>Lowest error for <br>Implied Volatility"]
  E --> H["Standard Feed-Forward: <br>Baseline comparison"]

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 empirically investigates how different neural network architectures affect option pricing and implied volatility estimation, identifying the generalized highway network as best for pricing and a simplified DGM variant as best for implied volatility.

Source sentence
We find that the generalized highway network architecture achieves the best performance, when considering the mean squared error and the training time as criteria, within the considered parameter budgets for the Black‑Scholes and Heston option pricing problems.

Assumes 6

  • The asset price follows the Black‑Scholes dynamics with constant volatility. model
    Source sentence
    where r denotes the risk‑free interest rate, σ the volatility and W denotes a Q‑Brownian motion.
  • The asset price follows the Heston stochastic volatility model. model
    Source sentence
    A popular model that captures the behavior of market data is the Heston stochastic volatility model, cf. Heston [19].
  • Traded option prices are observable in the market for various strikes and maturities. market
    Source sentence
    Assume that we can observe in a financial market the traded prices πM t (C; K, T) for a call option C, for various strikes K and maturities T.
  • Implied volatility is defined as the volatility that makes the Black‑Scholes price match the market price. other
    Source sentence
    The implied volatility σiv(K, T) is the volatility that should be inserted into the Black‑Scholes equation (2.2) such that the model price πt(C; K, T, σ) matches the traded market price πM t (C; K, T).
Show 2 more
  • Synthetic training data are generated from the analytical Black‑Scholes and Heston formulas. data
    Source sentence
    ... for the Black‑Scholes and Heston option pricing problems.
  • Real implied‑volatility surfaces are taken from the OptionMetrics database on a fixed grid of maturities and deltas. data
    Source sentence
    We evaluated the architectures on real implied volatility data obtained from the OptionMetrics database. Each daily implied volatility surface is given on a fixed grid of maturities {10, 30, 60, 91, 122, 152, 182, 273, 365} days, and put deltas {−90, −80, −70, −60, −50, −40, −30, −20, −10}.

Data

  • Synthetic Black‑Scholes and Heston option price data synthetic unknown
    Source sentence
    ... for the Black‑Scholes and Heston option pricing problems.
  • OptionMetrics implied‑volatility surface dataset · OptionMetrics options commercial
    Source sentence
    We evaluated the architectures on real implied volatility data obtained from the OptionMetrics database.

Recognised sources: OptionMetrics IvyDB

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Generalized highway networks achieve the lowest mean‑squared error and fastest training time for Black‑Scholes and Heston pricing within the tested parameter budgets. performance
    Source sentence
    highway‑type network architectures outperform all other variants, when considering the mean squared error and the computational time as criteria.
  • A simplified, no‑recurrence DGM variant yields the lowest error for the transformed implied‑volatility problem. predictive-accuracy
    Source sentence
    a simplified DGM variant achieves the lowest error among the tested architectures.
  • When models are capacity‑normalized (equal number of parameters), the performance ordering remains largely unchanged. performance
    Source sentence
    even when all models have the same number of parameters, the differences in performance did not disappear.
  • Real‑world implied‑volatility experiments confirm synthetic results, with MSE in the range 10⁻³–10⁻⁴ and similar architecture rankings. predictive-accuracy
    Source sentence
    The performance on real implied volatility data is highly consistent with the synthetic experiments
  • Transforming the implied‑volatility dataset (e.g., scaling inputs) dramatically improves prediction accuracy, while more complex architectures do not help on the raw data. negative
    Source sentence
    When training networks on the implied volatility dataset, we found poor performance for all networks

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 4 papers in the archive (4 by independent authors)

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.