Transformer-based robot portfolio optimization by integrating user adoption willingness

Abstract

In recent years, when traditional portfolio optimization (PO) models deal with highly volatile and irrational market environments, they often face problems of deviation in return prediction and lag in risk response. This makes it difficult for investment decisions to consider dynamic market changes and differences in investor behavior. To solve this problem, this study constructs an intelligent PO model that integrates investors' adoption willingness—Transformer-based Adaptive Willingness Portfolio (TransAWP). By integrating market price sequences, macroeconomic indicators, asset risk characteristics, and investor willingness signals, the model realizes behavior-driven adaptive adjustment of investment weights. Empirical research based on multi-source financial data from 2018 to 2023 indicates that this model shows a relatively high level of performance in both prediction accuracy and robustness; meanwhile, it achieves relatively leading results on key evaluation indicators. Specifically, TransAWP shows lower prediction errors, stronger explanatory power, and more stable output features across multiple evaluation indicators; this reflects its strong ability to capture market trends and adapt to risk adjustments. This study aims to provide institutional investors and intelligent investment advisory systems with a decision-making framework that balances return stability and investors' psychological preferences.