Paper: arXiv 2504.20088
Authors: Joao Felipe Gueiros, Hemanth Chandravamsi, Steven H. Frankel
Abstract
This paper explores the use of deep residual networks for pricing European options on Petrobras, one of the world’s largest oil and gas producers, and compares its performance with the Black-Scholes (BS) model. Using eight years of historical data from B3 (Brazilian Stock Exchange) collected via web scraping, a deep learning model was trained using a custom built hybrid loss function that incorporates market data and analytical pricing. The data for training and testing were drawn between the period spanning November 2016 to January 2025, using an 80-20 train-test split. The test set consisted of data from the final three months: November, December, and January 2025. The deep residual network model achieved a 64.3% reduction in the mean absolute error for the 3-19 BRL (Brazilian Real) range when compared to the Black-Scholes model on the test set. Furthermore, unlike the Black-Scholes solution, which tends to decrease its accuracy for longer periods of time, the deep learning model performed accurately for longer expiration periods. These findings highlight the potential of deep learning in financial modeling, with future work focusing on specialized models for different price ranges.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper features significant mathematical content, including partial differential equations and neural network architectures, while also demonstrating strong empirical rigor through multi-year Brazilian market data collection, hybrid loss functions, and clear performance metrics.
Research Flowchart
flowchart TD
A["Research Goal:<br>Compare Deep Learning vs<br>Black-Scholes on Petrobras Options"] --> B
subgraph B ["Methodology & Data"]
direction LR
B1["Web Scraping<br>B3 Historical Data"] --> B2["Preprocessing<br>Nov 2016 - Jan 2025"]
B2 --> B3["Split: 80% Train / 20% Test<br>Test: Nov-Jan 2025"]
end
B --> C
subgraph C ["Computational Processes"]
direction LR
C1["Black-Scholes<br>Standard Analytical Model"] --- C2["Deep Learning<br>ResNet + Hybrid Loss Function"]
end
C --> D
subgraph D ["Key Findings"]
D1["64.3% reduction in MAE<br>for 3-19 BRL price range"]
D2["DL accuracy maintained<br>for longer expiration periods"]
D3["BS model accuracy<br>decreases over time"]
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 a deep residual neural network model, trained with a custom hybrid loss, to price European Petrobras options and compares it to Black‑Scholes.
Source sentence
A deep neural network was designed, inspired by residual network architectures commonly used in computer vision. The model was trained using a custom hybrid loss function that dynamically weights empirical market data and theoretical Black‑Scholes prices.
Assumes 5
- Stock prices follow a random walk and can be modeled using a Gaussian distribution. model
Source sentence
The BS equation is based on the idea that stock prices follow a random walk and can be modeled using a Gaussian distribution
- Only European‑style contracts are considered; American options are excluded. market
Source sentence
the focus of this analysis was limited to Petrobras (PETR) European‑style contracts
- Volatility is computed as a 21‑day rolling standard deviation of daily returns annualised by √252. data
Source sentence
we computed the volatility of PETR4 by taking the daily percentage changes in the stock price and applying a 21‑day rolling standard deviation. This rolling standard deviation was then multiplied by √252
- If an exact date match for stock price, volatility or SELIC rate is missing, the closest preceding date is used. execution
Source sentence
If an exact date match was missing for any of these fields, we used the closest preceding date instead
Show 1 more
- Options with expiration between one and three months are retained; longer or shorter expirations are excluded. data
Source sentence
retaining only contracts set to expire within one to three months
Data
2016-11 – 2025-01 · Petrobras European options · options
- Petrobras options data (PETR) from B3 options public
Source sentence
The dataset contained information on all listed options, but the focus of this analysis was limited to Petrobras (PETR) European‑style contracts.
- PETR4 stock price and volatility market-prices public
Source sentence
To create a single dataset suitable for model training, we needed to combine the Petrobras option data, the PETR4 stock price (including its volatility), and the SELIC rate.
- SELIC rate macro public
Source sentence
To create a single dataset suitable for model training, we needed to combine the Petrobras option data, the PETR4 stock price (including its volatility), and the SELIC rate.
Costs
Not applicable
Finds 4
- The deep residual network reduces mean absolute error by 64.3 % for options priced between 3 and 19 BRL, covering 43.41 % of Petrobras option transactions on B3. performance
Source sentence
For this price range, the model achieves a significant improvement of a 64.3% reduction in MAE across all months for around 43.41% of all Petrobras option transactions on B3.
- The model maintains accuracy for longer expiration periods, contrary to Black‑Scholes which degrades over time. predictive-accuracy
Source sentence
the deep learning model performed accurately for longer expiration periods.
- The model overestimates option prices outside the 3‑19 BRL range. negative
Source sentence
the deep learning model also has limitations, particularly a tendency to overestimate option prices outside the 3-19 BRL range.
- The model successfully values contracts around 5 BRL that Black‑Scholes severely undervalues. performance
Source sentence
it successfully values contracts priced around 5 BRL, such as PETRL36, PETRL410, PETRD397, and PETR409, whereas the Black‑Scholes model severely undervalues them.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 25 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 (30)
- Théorie de la spéculation (1900)
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019) — link
- CS230: Deep Learning - Option Pricing with Deep Learning (2019) — link
- Neural Network Learning of Black-Scholes Equation for Option Pricing — → 2405.05780
- Deep Residual Learning for Image Recognition (2016)
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015)
- Optuna: A Next-generation Hyperparameter Optimization Framework (2019)
- Adam: A Method for Stochastic Optimization (2015) — link
and 22 more
- Option pricing based on the residual neural network (2024)
- On the numerical solution of nonlinear Black–Scholes equations (2008)
- A Deep Learning Approach for Option Pricing in Emerging Markets (2021) — link
- Deep Learning Artificial Neural Network for Pricing Multi-Asset European Options (2023)
- Accelerated American Option Pricing with Deep Neural Networks (2023)
- Stable Multilevel Deep Neural Networks for Option Pricing and xVAs Using Forward-Backward Stochastic Differential Equations (2022)
- Deep Neural Network Framework Based on Backward Stochastic Differential Equations for Pricing and Hedging American Options in High Dimensions (2018)
- Option pricing using deep learning approach based on LSTM-GRU neural networks: Case of London stock exchange (2022)
- A neural network model for estimating option prices (1993) — link
- Proof that properly anticipated prices fluctuate randomly (2016)
- The pricing of options and corporate liabilities (1973)
- Theory of rational option pricing (1971)
- Market and industry factors in stock price behavior (1966)
- A non-random walk down Wall Street (2011)
- The valuation of American put options (1977)
- Deep optimal stopping (2019)
- A nonparametric approach to pricing and hedging derivative securities via learning networks (1994)
- Pricing and hedging derivative securities with neural networks and a homogeneity hint (2000)
- Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
- A neural network model for estimating option prices (1993)
- Option pricing model combining ensemble learning methods and network learning structure (2022)
- Neural networks for option pricing and hedging: a literature review (2019) — link
1 of 30 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.