Data-Driven Portfolio Optimization Using a Predict-Then-Optimize Framework

Yi Wang, Takashi Hasuike

Abstract

Effective portfolio diversification remains a central challenge in quantitative asset management. In this study, we propose a data-driven framework based on the predict-then-optimize (PO) paradigm, which combines return forecasting with portfolio allocation in a sequential manner. The forecasting module employs DLinear, a lightweight deep learning model, to capture temporal patterns in historical asset returns. Based on the predicted returns, a portfolio allocation strategy is constructed by optimizing the Sharpe ratio, allowing the model to generate adaptive portfolio weights under changing market conditions. This PO-based framework provides a practical way to connect predictive modeling with downstream decision-making. Empirical results across three asset universes demonstrate that the proposed approach achieves competitive performance under different settings, indicating its potential applicability in real-world portfolio management.

Source: semanticscholar · PDF

Read the AI summary, key takeaways and discussion on WOBR Quant Research.