Financial Engineering David Araujo

Using Neural Networks to Improve Financial Option Pricing

A backpropagation neural network trained on Black-Scholes values achieves improved pricing accuracy for in- and out-of-the-money options by capturing real-world market imperfections.

neural networks Black-Scholes option pricing machine learning backpropagation financial engineering

This research explores an alternative approach to traditional Black-Scholes pricing by leveraging neural networks for financial options valuation. The methodology involves training a backpropagation neural network using Black-Scholes values as training data inputs, then evaluating the trained model against market prices.

Motivation

The Black-Scholes model rests on assumptions — constant volatility, lognormal returns, frictionless markets — that are routinely violated in practice. These violations produce systematic mispricing, particularly for deep in-the-money and out-of-the-money contracts.

Methodology

A backpropagation neural network is trained on Black-Scholes generated prices, then evaluated against actual market data. The neural network learns to adjust for systematic deviations — effectively learning the correction to Black-Scholes implied by observed market prices.

Key Findings

  • Improved predictive accuracy versus standard Black-Scholes for in-the-money contracts
  • Improved accuracy for out-of-the-money contracts
  • Neural networks can adjust for real-world market imperfections that Black-Scholes cannot

Significance

The research positions this neural network approach as a promising direction for option valuation, offering practitioners an enhanced tool for more accurate market-based pricing predictions.