Paper: arXiv 2506.17511

Authors: Alessio Brini, David A. Hsieh, Patrick Kuiper, Sean Moushegian, David Ye

Abstract

The key objective of this paper is to develop an empirical model for pricing SPX options that can be simulated over future paths of the SPX. To accomplish this, we formulate and rigorously evaluate several statistical models, including neural network, random forest, and linear regression. These models use the observed characteristics of the options as inputs – their price, moneyness and time-to-maturity, as well as a small set of external inputs, such as the SPX and its past history, dividend yield, and the risk-free rate. Model evaluation is performed on historical options data, spanning 30 years of daily observations. Significant effort is given to understanding the data and ensuring explainability for the neural network. A neural network model with two hidden layers and four neurons per layer, trained with minimal hyperparameter tuning, performs well against the theoretical Black-Scholes-Merton model for European options, as well as two other empirical models based on the random forest and the linear regression. It delivers arbitrage-free option prices without requiring these conditions to be imposed.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper combines advanced statistical methods like GARCH and neural networks with a rigorous evaluation on 30 years of historical options data, showing practical backtest readiness. While the math includes stochastic modeling and machine learning, the heavy empirical focus on real-world data validation places it in the Holy Grail quadrant.

Research Flowchart

  flowchart TD
  A["Research Goal: Develop empirical model<br>for SPX options pricing"] --> B["Methodology: Train ML models<br>NN, Random Forest, Linear Regression"]
  B --> C["Data & Inputs:<br>SPX options price, moneyness,<br>time-to-maturity, SPX history,<br>dividend yield, risk-free rate"]
  C --> D["Computational Process:<br>Train models on 30 years<br>of historical daily data"]
  D --> E["Key Outcomes:<br>NN model (2 layers, 4 neurons) delivers<br>arbitrage-free prices, outperforms<br>Black-Scholes-Merton, Random Forest,<br>and Linear Regression"]

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 a hybrid framework that combines a GARCH volatility model with a neural network to price SPX options.

Source sentence
we propose a hybrid framework that integrates econometric volatility modeling with data-driven option pricing via a non‑parametric model. Specifically, we employ a Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model of Bollerslev [1986] to capture the time‑varying volatility

Assumes 6

  • Only contracts with at least one month and at most 18 months to expiry are kept. data
    Source sentence
    We retain only contracts with at least one month until expiry to ensure they have some remaining time to expiration. We also exclude options with more than 18 months to expiration.
  • Options must have a positive bid price. data
    Source sentence
    Options must have a positive bid price.
  • The Black‑Scholes price may be used as an optional regressor. model
    Source sentence
    We optionally include the Black‑Scholes price computed from these variables as a regressor.
  • Regressors include SPX, moneyness, time to maturity, dividend yield, risk‑free rate, predicted GARCH volatility and interaction terms. model
    Source sentence
    The set of regressors includes SPX, moneyness, time to maturity, dividend yield, the risk‑free rate, the predicted GARCH volatility, and squared terms and interaction terms among these variables.
Show 2 more
  • The options studied are European‑style put options on the S&P 500 index. market
    Source sentence
    The dataset comprises European‑style put options on the S&P 500 index
  • GARCH‑normal model adequately captures time‑varying volatility of SPX returns. statistical
    Source sentence
    We employ a Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model of Bollerslev [1986] to capture the time‑varying volatility inherent in asset returns

Data

  • S&P 500 Put Options Dataset · OptionMetrics options commercial
    January 1996 to December 2022 · European‑style put options on the S&P 500 index
    Source sentence
    The dataset comprises European‑style put options on the S&P 500 index, with daily data sourced from OptionMetrics, covering the period from January 1996 to December 2022.

Recognised sources: OptionMetrics IvyDB Cboe CRSP

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Neural network and random forest consistently outperform linear regression for OTM put pricing. performance
    Source sentence
    NN and RF consistently outperform LR.
  • Including the Black‑Scholes price as a feature improves predictive accuracy for all models except the neural network, which remains competitive without it. predictive-accuracy
    Source sentence
    Including the BS price as a feature improves model performance across both training methodologies, namely the expanding and rolling window approaches.
  • Random forest loses predictive power and often performs worse than the Black‑Scholes baseline when the BS feature is omitted. performance
    Source sentence
    When BS is removed, RF loses predictive power and often performs worse than the BS baseline.
  • Neural network retains strong performance even when the Black‑Scholes price is excluded. predictive-accuracy
    Source sentence
    NN retains stronger performance even when BS is excluded.
  • A neural network with two hidden layers and four neurons each consistently outperforms linear regression and random forest, especially for deep OTM options and during high volatility periods. performance
    Source sentence
    a neural network with two hidden layers and four neurons each consistently outperforms both linear regression and random forest benchmarks, especially in pricing deep OTM options and during periods of elevated volatility.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 51 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 (74)

  1. A Nonparametric Approach to Pricing and Hedging Derivative Securities Via Learning Networks (1994) — link
  2. A factor model for option returns (2022) — link
  3. Common Factors Affecting Bond Returns (1991) — link
  4. A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
  5. The pricing of options and corporate liabilities (1973)
  6. A conditionally heteroskedastic time series model for speculative prices and rates of return (1987)
  7. Managing smile risk (2002)
  8. Pricing with a smile (1994)
and 66 more
  1. Riding on a smile (1994)
  2. The constant elasticity of variance option pricing model (1996)
  3. Jumps and stochastic volatility: Exchange rate processes implicit in deutsche mark options (1996)
  4. A jump-diffusion model for option pricing (2002)
  5. The variance gamma process and option pricing (1998)
  6. Option pricing using machine learning (2021)
  7. Application of Machine Learning in Option Pricing: A Review (2022)
  8. Machine learning for quantitative finance: fast derivative pricing, hedging and fitting (2018)
  9. Machine learning for pricing American options in high-dimensional Markovian and non-Markovian models (2020)
  10. Can a machine correct option pricing models? (2023)
  11. Deep learning calibration of option pricing models: some pitfalls and solutions (2019) — link
  12. Option return predictability with machine learning and big data (2023)
  13. DeepOption: A novel option pricing framework based on deep learning with fused distilled data from multiple parametric methods (2021)
  14. Black–Scholes option pricing using machine learning (2023)
  15. Parametric models and non-parametric machine learning models for predicting option prices: Empirical comparison study over KOSPI 200 Index options (2014)
  16. Stock options pricing via machine learning methods combined with firm characteristics (2023)
  17. Deep learning based hybrid computational intelligence models for options pricing (2022)
  18. Pricing cryptocurrency options with machine learning regression for handling market volatility (2024)
  19. The deep parametric PDE method and applications to option pricing (2022)
  20. Machine learning for option pricing: an empirical investigation of network architectures — → 2307.07657
  21. Gated neural networks for option pricing: Rationality by design (2017)
  22. Time-sequencing European options and pricing with deep learning–Analyzing based on interpretable ALE method (2022)
  23. Option pricing with machine learning (2019)
  24. Artificial neural network for option pricing with and without asymptotic correction (2021)
  25. An intelligent learning and ensembling framework for predicting option prices (2021)
  26. Differential machine learning (2020) — link
  27. Option Pricing Based on the Residual Neural Network (2024)
  28. Neural network models for bitcoin option pricing (2019)
  29. Deep hedging (2019)
  30. Deep learning for exotic option valuation (2021) — link
  31. Option Pricing Models: From Black-Scholes-Merton to Present. (2022)
  32. A neural network approach to understanding implied volatility movements (2020)
  33. Machine Learning Methods for Pricing Financial Derivatives — → 2406.00459
  34. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
  35. The US Treasury yield curve: 1961 to the present (2007)
    1. A value for n-person games (2016)
  36. A unified approach to interpreting model predictions (2017)
  37. Teaching economics to the machines (2023)
  38. rugarch: Univariate GARCH models. (2024)
  39. Empirical performance of alternative option pricing models (1997)
  40. Empirical option pricing models (2022)
  41. Empirical option pricing: A retrospection (2003)
  42. Nonlinear option pricing (2013)
  43. Random forests (2001)
  44. The elements of statistical learning: data mining, inference, and prediction (2009)
  45. A machine learning approach to volatility forecasting (2023)
  46. Deep learning (2016)
  47. Efficient backprop (2002)
  48. A comprehensive survey of regression-based loss functions for time series forecasting (2024)
  49. Human-level control through deep reinforcement learning (2015)
  50. Option pricing when underlying stock returns are discontinuous (1976)
  51. Generalized Autoregressive Conditional Heteroskedasticity (1986) — link
  52. The volatility surface: a practitioner’s guide (2011)
  53. Dynamics of implied volatility surfaces (2002)
  54. Implied volatility and future portfolio returns (2007)
  55. Liquidity, implied volatility and tail risk: A comparison of liquidity measures (2020)
  56. Dynamic estimation of volatility risk premia and investor risk aversion from option-implied and realized volatilities (2011)
  57. Factors explaining movements in the implied volatility surface (2002)
  58. On the seasonality in the implied volatility of electricity options (2019)
  59. Pricing and hedging with smiles (1997)
  60. Volatility surfaces: theory, rules of thumb, and empirical evidence (2007)
  61. Dynamics of the implied volatility surface: Theory and empirical evidence (2010)
  62. Arbitrage-free smoothing of the implied volatility surface (2009)
  63. Revisiting unreasonable effectiveness of data in deep learning era (2017)
  64. Scaling laws for neural language models (2020) — link
  65. Tail risk hedging: Contrasting put and trend strategies (2021)

2 of 74 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.