Paper: arXiv 2510.01446
Authors: Georgy Milyushkov
Abstract
This study investigates the application of machine learning techniques, specifically Neural Networks, Random Forests, and CatBoost for option pricing, in comparison to traditional models such as Black-Scholes and Heston Model. Using both synthetically generated data and real market option data, each model is evaluated in predicting the option price. The results show that machine learning models can capture complex, non-linear relationships in option prices and, in several cases, outperform both Black-Scholes and Heston models. These findings highlight the potential of data-driven methods to improve pricing accuracy and better reflect market dynamics.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced stochastic calculus (Itô’s Lemma, coupled SDEs) and numerical methods (semi-analytical solutions, characteristic functions), yet it also demonstrates strong empirical rigor through synthetic and real market data, model implementation (Python/QuantLib), GitHub code availability, and standard error metrics.
Research Flowchart
flowchart TD
A["Research Goal<br>Can ML Outperform Traditional<br>Option Pricing Models?"] --> B["Methodology<br>Compare ML vs. Black-Scholes & Heston"]
B --> C{"Data Sources"}
C --> D["Synthetic Data"]
C --> E["Real Market Data"]
D --> F["Computation<br>Train Neural Networks, Random Forests, CatBoost"]
E --> F
F --> G["Analysis<br>Evaluate Prediction Accuracy"]
G --> H["Key Findings<br>ML models capture non-linear relationships<br>ML often outperforms traditional models"]
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 evaluates neural networks, random forests, and CatBoost for option pricing against Black‑Scholes and Heston models using synthetic and real market data.
Source sentence
This study investigates the application of machine learning techniques, specifically Neural Networks, Random Forests, and CatBoost for option pricing, in comparison to traditional models such as Black‑Scholes and Heston Model.
Assumes 4
- Black‑Scholes assumes constant volatility, continuous trading and frictionless markets. model
Source sentence
offer closed‑form solutions for European options under highly strict assumptions, including constant volatility, continuous trading, and frictionless markets.
- Risk‑free rate is taken from U.S. Treasury Par Yield Curve and linearly interpolated to each option’s maturity. market
Source sentence
linear interpolation was employed to estimate the appropriate yield for each option’s specific maturity.
- Synthetic data are generated using the Black‑Scholes formula with linear distortions. data
Source sentence
Synthetic data was generated using the Black‑Scholes formula for Euro…
- Midpoint between bid and ask prices is a reliable proxy for the true market price. execution
Source sentence
The midpoint between bid and ask prices was used as a proxy for the observed market price, due to its stability and ability to reflect a more accurate valuation than last‑traded prices.
Data
- Synthetic option data (linear distortion) synthetic unknown
Source sentence
Synthetic data was generated using the Black‑Scholes formula for Euro…
- Apple (AAPL) call option chain options public
Source sentence
Option chain data for Apple Inc. (AAPL) was extracted, including bid, ask, implied volatility, strike, and expiration dates.
Costs
Not applicable
Finds 4
- CatBoost achieved the lowest MSE (0.00262) and highest R2 (0.99998) under linear distortion, outperforming other ML models and Black‑Scholes. performance
Source sentence
CatBoost, trained using Bayesian hyperparameter optimization, outperformed all other models in this setting, attaining the lowest MSE (0.00262) and highest R2 score (0.99998).
- MLP after hyperparameter tuning reached R2 of 0.9999755, surpassing the Black‑Scholes benchmark. performance
Source sentence
reducing both MSE and MAE and achieving an R2 of 0.9999755 - outperforming the Black‑Scholes benchmark.
- Random Forests with engineered features achieved R2 of 0.99961, improving over the baseline without features. performance
Source sentence
Random Forests improved substantially, achieving an R2 of 0.99961.
- Synthetic linear‑distortion results showed MSE 0.05655 and R2 0.99952 for the baseline model, indicating strong fit despite distortion. statistical
Source sentence
The results were surprisingly strong, with an MSE of 0.05655 and an R2 of 0.99952.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 39 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 (10)
- The Pricing of Options and Corporate Liabilities (1973)
- A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (1993) — link
- Option pricing when underlying stock returns are discontinuous (1976) — link
- A Nonparametric Approach to Pricing and Hedging Derivative Securities Via Learning Networks (1994)
- Machine learning for option pricing: an empirical investigation of network architectures — → 2307.07657
- Machine learning for pricing American options in high-dimensional Markovian and non-Markovian models (2020) — link
- American Option Pricing using Self-Attention GRU and Shapley Value Interpretation (2023) — link
- The Shape and Term Structure of the Index Option Smirk: Why Multifactor Stochastic Volatility Models Work So Well (2009) — link
and 2 more
1 of 10 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.