Integrating LLM-based Transformers into portfolio optimization and asset allocation

Develop portfolio optimization and asset allocation frameworks that leverage large language model-based Transformers applied to asset pricing and factor investing, and evaluate their incremental benefits relative to forecasting-only applications, following techniques akin to those in transformer-based allocation approaches such as Ma (2023).

Background

Beyond return prediction, the practical value of Transformer models in portfolio construction and allocation remains to be established. The authors recommend exploring LLM-based portfolio optimization or asset allocation techniques, referencing recent transformer-based allocation work.

This points to extending Transformer usage from forecasting to decision-making, assessing risk-adjusted performance improvements and robustness in allocation.