Deep Learning in Asset Liability Management

Specialeforsvar: Christian Hjerrild Blom

Titel: Deep Learning in Asset Liability Management

Abstract: This thesis explores the application of deep learning techniques to optimize asset liability management (ALM) for banking institutions. Banks employ ALM to manage interest rate risks arising from mismatched maturities, reference rates, and embedded optionalities in their portfolios. This research integrates stochastic models for balance sheet simulation and a deep learning framework for decision-making under regulatory constraints. The study builds on existing methodologies but introduces
enhancements in simulation engines, deep learning architectures, and constraint modeling. The results demonstrate the potential of deep learning to derive optimal strategies that align with both profitability and regulatory compliance, marking advancement in ALM practices. We have found that deep learning can improve profitability, however for pactical purposes risk appetite needs to be fully specified for optimal decision making.

Vejleder: David Skovmand
Censor:   Mads Stenbo Nielsen,  CBS