Machine learning methods for American-style path-dependent contracts

Pallavicini Andrea, Intesa Sanpaolo

In the present work, we introduce and compare state-of-the-art algorithms, that are now classified under the name of machine learning, to price Asian and look-back products with early-termination features. These include randomized feed-forward neural networks, randomized recurrent neural networks, and a novel method based on signatures of the underlying price process. Additionally, we explore potential applications on callable certificates. Furthermore, we present an innovative approach for calculating sensitivities, specifically Delta and Gamma, leveraging Chebyshev interpolation techniques.

Area: CS49 - Analytical and numerical methods for energy transition (Tiziano Vargiolu and Athena Picarelli)

Keywords: Pricing American-style options, Pricing path-dependent options, Random networks, Signature methods, Least-square Monte Carlo, Chebyshev Greeks

Please Login in order to download this file