Paper: arXiv 2409.19706
Authors: Ananya Unnikrishnan
Abstract
An accurate valuation of American call options is critical in most financial decision making environments. However, traditional models like the Barone-Adesi Whaley (B-AW) and Binomial Option Pricing (BOP) methods fall short in handling the complexities of early exercise and market dynamics present in American options. This paper proposes a Modular Neural Network (MNN) model which aims to capture the key aspects of American options pricing. By dividing the prediction process into specialized modules, the MNN effectively models the non-linear interactions that drive American call options pricing. Experimental results indicate that the MNN model outperform both traditional models as well as a simpler Feed-forward Neural Network (FNN) across multiple stocks (AAPL, NVDA, QQQ), with significantly lower RMSE and nRMSE (by mean). These findings highlight the potential of MNNs as a powerful tool to improve the accuracy of predicting option prices.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced neural network architectures (Modular Neural Networks) with complex non-linear modeling, warranting a high math score, and includes empirical backtesting on real stocks (AAPL, NVDA, QQQ) with reported error metrics (RMSE, nRMSE), justifying a moderate-to-high empirical rigor.
Research Flowchart
flowchart TD A["Research Goal<br>American Call Option Pricing"] --> B["Data Preparation<br>AAPL, NVDA, QQQ Data"] B --> C["Model Architecture<br>Modular Neural Network"] C --> D["Computational Process<br>1. Feature Engineering<br>2. Training/Validation<br>3. Early Exercise Modeling"] D --> E["Traditional Baseline<br>B-AW & Binomial Models"] D --> F["Feed-forward Neural Network"] E & F --> G["Performance Comparison"] G --> H["Key Findings<br>MNN outperforms all benchmarks<br>Lower RMSE & nRMSE<br>Superior for complex early exercise"] style A fill:#e1f5fe style H fill:#e8f5e8 style C fill:#fff3e0 style D fill:#f3e5f5
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 Modular Neural Network (MNN) model to price American call options by dividing the prediction process into specialized modules.
Source sentence
This paper proposes a Modular Neural Network (MNN) model which aims to capture the key aspects of American options pricing.
Assumes 3
- Risk‑free rate is proxied by matching U.S. Treasury maturities to the option’s days‑to‑expiration. model
Source sentence
interest rates were derived from FRED by matching U.S. Treasury maturities to the option’s DTE. This served as a proxy for the risk‑free rate
- Data covers only call options for AAPL, NVDA and QQQ for the calendar year 2023. data
Source sentence
Data was originally collected for AAPL, NVDA and QQQ call options for the year 2023
- Hyper‑parameter tuning was performed only on the AAPL dataset and the resulting best model was applied to NVDA and QQQ. execution
Source sentence
We only conducted hyper‑parameter tuning on the AAPL dataset to find the best model, as performing extensive tuning on each dataset would have been computationally expensive
Data
- AAPL, NVDA and QQQ call options 2023 options public
Source sentence
Data was originally collected for AAPL, NVDA and QQQ call options for the year 2023, which included relevant features like quote date, expiration date, underlying price, days to expiration (DTE), option Greeks (delta, gamma, vega, theta, rho), implied volatility, volume, and strike price.
- Macro‑economic indicators macro public
Source sentence
Key macroeconomic indicators like GDP growth, inflation, and unemployment rates were also sourced from the Federal Reserve Economic Data (FRED).
FRED Yahoo Finance
Costs
Not applicable
Finds 3
- For AAPL options the MNN achieved RMSE 3.335 and nRMSE 0.0849, outperforming B‑AW (RMSE 8.632, nRMSE 0.2191), BOP (RMSE 8.639, nRMSE 0.2192) and FNN (RMSE 3.551, nRMSE 0.0904). predictive-accuracy
Source sentence
the FNN model also performed worse than the MNN, with an RMSE of 3.551 and an nRMSE of 0.0904.
- For NVDA options the MNN recorded RMSE 17.516 and nRMSE 0.1701, better than B‑AW (RMSE 41.737, nRMSE 0.4060), BOP (RMSE 41.748, nRMSE 0.4061) and FNN (RMSE 19.563, nRMSE 0.1900). predictive-accuracy
Source sentence
the B‑AW method resulted in an RMSE of 41.737 and an nRMSE of 0.4060
- For QQQ options the MNN achieved RMSE 8.052 and nRMSE 0.1878, surpassing B‑AW (RMSE 14.881, nRMSE 0.3489), BOP (RMSE 14.889, nRMSE 0.3491) and FNN (RMSE 8.699, nRMSE 0.2029). predictive-accuracy
Source sentence
The MNN model achieved an RMSE of 8.052 and an nRMSE of 0.1878, outperforming both the traditional methods and the FNN model
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 10 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.