Risk-neutral density extraction from option prices: improved pricing with mixture density networks.

Area of Science:

• Quantitative Finance
• Financial Econometrics
• Computational Finance

Background:

• Accurate pricing and hedging of financial derivatives are crucial in finance.
• Existing models like Black-Scholes and GARCH have limitations in capturing stylized facts.

Purpose of the Study:

• To develop a novel semi-nonparametric approach for risk-neutral density (RND) extraction.
• To enhance the flexibility and accuracy of option pricing models.
• To provide better tools for risk management.

Main Methods:

• Utilizing an extension of mixture density networks for RND estimation.
• Modeling RND shape non-linearly as a function of the time horizon.
• Applying the method to a large dataset of FTSE 100 options data.

Main Results:

• The proposed method successfully captures stylized facts like negative skewness and excess kurtosis.
• Demonstrated significantly superior out-of-sample pricing accuracy compared to Black-Scholes and GARCH models.
• Extracted RNDs offer valuable insights for Value-at-Risk (VaR) estimations.

Conclusions:

• The new semi-nonparametric approach offers a more flexible and accurate method for RND extraction.
• This model provides a significant improvement over traditional option pricing models.
• The findings have practical implications for risk management and derivative pricing.