Paper: arXiv 2409.06724

Authors: Boris Ter-Avanesov, Homayoon Beigi

Abstract

We explore the performance of various artificial neural network architectures, including a multilayer perceptron (MLP), Kolmogorov-Arnold network (KAN), LSTM-GRU hybrid recursive neural network (RNN) models, and a time-delay neural network (TDNN) for pricing European call options. In this study, we attempt to leverage the ability of supervised learning methods, such as ANNs, KANs, and gradient-boosted decision trees, to approximate complex multivariate functions in order to calibrate option prices based on past market data. The motivation for using ANNs and KANs is the Universal Approximation Theorem and Kolmogorov-Arnold Representation Theorem, respectively. Specifically, we use S&P 500 (SPX) and NASDAQ 100 (NDX) index options traded during 2015-2023 with times to maturity ranging from 15 days to over 4 years (OptionMetrics IvyDB US dataset). Black & Scholes’s (BS) PDE \cite{“Black1973”} model’s performance in pricing the same options compared to real data is used as a benchmark. This model relies on strong assumptions, and it has been observed and discussed in the literature that real data does not match its predictions. Supervised learning methods are widely used as an alternative for calibrating option prices due to some of the limitations of this model. In our experiments, the BS model underperforms compared to all of the others. Also, the best TDNN model outperforms the best MLP model on all error metrics. We implement a simple self-attention mechanism to enhance the RNN models, significantly improving their performance. The best-performing model overall is the LSTM-GRU hybrid RNN model with attention. Also, the KAN model outperforms the TDNN and MLP models. We analyze the performance of all models by ticker, moneyness category, and over/under/correctly-priced percentage.

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 involves advanced neural network architectures (LSTM-GRU hybrid with attention) and theoretical foundations (Universal Approximation, Kolmogorov-Arnold Representation Theorem), but it is heavily grounded in real-world application using a large dataset (OptionMetrics IvyDB), rigorous backtesting against a standard benchmark (Black-Scholes), and detailed performance analysis by moneyness category.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Price SPX/NDX European Calls<br>using ANNs vs Black-Scholes"] --> B["Data: S&P 500 & NASDAQ 100<br>2015-2023 (OptionMetrics IvyDB)"]
  
  B --> C["Methodology:<br>Supervised Learning Models"]
  
  subgraph C ["Models Implemented"]
      C1["MLP"]
      C2["XGBoost"]
      C3["KAN"]
      C4["TDNN"]
      C5["LSTM-GRU Hybrid<br>+ Self-Attention"]
  end
  
  C --> D["Computational Process:<br>Train Models to Approximate<br>Complex Multivariate Pricing Functions"]
  D --> E["Benchmark:<br>Black-Scholes PDE Model"]
  
  E --> F["Performance Analysis:<br>By Ticker, Moneyness,<br>Over/Under Pricing"]
  
  F --> G["Key Findings/Outcomes"]
  
  subgraph G ["Results"]
      G1["BS Model Underperforms<br>vs. All ML Models"]
      G2["TDNN Outperforms<br>MLP (All Metrics)"]
      G3["KAN Outperforms<br>TDNN & MLP"]
      G4["LSTM-GRU Hybrid<br>with Attention = Best Overall<br>(Significant Improvement)"]
  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 using several artificial neural network architectures, including MLP, KAN, TDNN, and an LSTM‑GRU hybrid RNN with attention, to price European call options.

Source sentence
We explore the performance of various artificial neural network architectures, including a multilayer perceptron (MLP), Kolmogorov‑Arnold network (KAN), LSTM‑GRU hybrid recursive neural network (RNN) models, and a time‑delay neural network (TDNN) for pricing European call options.

Assumes 6

  • No transaction costs or taxes market
    Source sentence
    1. No transaction costs or taxes
  • No riskless arbitrage can exist market
    Source sentence
    2. No riskless arbitrage can exist
  • No dividends or other cash flows are paid during the lifetime of the security market
    Source sentence
    3. No dividends or other cash flows are paid during the lifetime of the security
  • Trading and hedging of security is continuous, with the asset’s liquidity being guaranteed execution
    Source sentence
    4. Trading and hedging of security is continuous, with the asset’s liquidity being guaranteed
Show 2 more
  • Risk‑free rate is constant for all maturities model
    Source sentence
    5. Risk‑free rate is constant for all maturities
  • Stock price follows Geometric Brownian motion with constant volatility over the option’s life model
    Source sentence
    7. Stock price follows Geometric Brownian motion with volatility of the underlying asset being constant over the lifetime of the option.

Data

  • OptionMetrics IvyDB US (subset df) · OptionMetrics via Wharton Research Data Services options commercial
    2015/05/13 to 2023/02/28 · European call options on SPX and NDX indices
    Source sentence
    The data for this study comes from the Wharton Research Data Services as a subset of the (optionm_all) IvyDB US by OptionMetrics data set

Recognised sources: OptionMetrics IvyDB WRDS Yahoo Finance

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Not applicable

Paper text extraction: Not modeled

Source sentence (the authors' words)
1. No transaction costs or taxes

Finds 5

  • The LSTM‑GRU hybrid RNN model with attention achieves the lowest error among all models, with MSE 0.000021, RMSE 0.004583, MAE 0.003301. performance
    Source sentence
    (1) Best RNN 0.000021 0.004583 0.003301
  • The KAN model outperforms the TDNN and MLP models. performance
    Source sentence
    Also, the KAN model outperforms the TDNN and MLP models.
  • The Black‑Scholes model performs far worse than the machine‑learning models, with MSE 0.003142, RMSE 0.056056, MAE 0.018677. performance
    Source sentence
    (6) Best BS 0.003142 0.056056 0.018677
  • Best XGBoost model achieves MSE 0.000025, RMSE 0.005021, MAE 0.003660, better than MLP but worse than RNN and KAN. performance
    Source sentence
    (2) Best XGBoost 0.000025 0.005021 0.003660
  • Best MLP model achieves MSE 0.000056, RMSE 0.007449, MAE 0.005483. performance
    Source sentence
    (4) Best MLP 0.000056 0.007449 0.005483

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 50 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.