Artificial Neural Networks Performance in WIG20 Index Options Pricing

Area of Science:

• Quantitative Finance
• Computational Finance
• Machine Learning Applications

Background:

• Option pricing models are crucial for financial markets.
• Artificial neural networks (ANNs) offer data-driven alternatives to traditional models.
• Emerging markets present unique challenges like low liquidity and high volatility.

Purpose of the Study:

• To compare the performance of ANNs against the Black-Scholes-Merton (BSM) model for option pricing.
• To evaluate the accuracy and robustness of ANNs using real-world market data.
• To investigate ANN performance in the context of emerging markets.

Main Methods:

• Trained and tested ANNs on historical options data (2009-2019) from the Warsaw Stock Exchange.
• Utilized a market data-driven approach for neural network training.
• Compared ANN results with the BSM model using various error metrics across different moneyness ratios.

Main Results:

• The BSM model demonstrated superior precision and robustness compared to the ANNs.
• ANNs did not yield more accurate option prices despite hyperparameter tuning.
• Significant differences in forecast bias were observed for ANNs across various moneyness states.

Conclusions:

• The BSM model remains a more reliable tool for option pricing than ANNs in this context.
• ANNs may require further development for effective application in low-liquidity, high-volatility emerging markets.
• This study offers initial insights into deep learning for option pricing in challenging market conditions.