Application of Deep Learning to Balance Sheet Optimisation
The solution proposed, called the Deep Balance Sheet Optimizer, applies artificial neural networks as the mathematical optimization tool to find the optimal and dynamic asset allocation strategy, considering balance sheet limits and regulatory constraints (a constrained optimization problem).
What is the development and the rationale behind it?
The Deep Balance Sheet Optimizer is to our knowledge the first practical application of deep learning to the challenging problems of Balance Sheet Management and Optimisation, which is based on academic research from world-leading universities. The natural alignment of deep learning algorithms and big data helps to tackle performance challenges and opens up possibilities that were previously unachievable in the field of Asset and Liability Management (ALM), e.g. the self-learning optimization algorithms can be used to:
- find optimal strategic asset and liability allocations considering possible future economic scenarios under regulatory and strategic constraints,
- optimize liquidity management in the presence of ring-fencing and uncertainty,
- find optimal product pricing.
Consequently, the offered functionality is of interest to Financial Institutions (FIs), which want to find, apply, and monitor an optimal asset allocation and balance sheet structure at the global and/or subsidiary/sub-portfolio level. Some of the advantages are summarized in the below-given table and treated in more detail in the remaining part of this document.