Momentum-integrated Multi-task Stock Recommendation with Converge-based Optimization

TL;DR

This work addresses the misalignment between traditional forecasting objectives and profit-oriented stock investment. It proposes MiM-StocR, a multi-task framework that adds a momentum line auxiliary task, optimizes a diffusion of ranking with Adaptive-$k$ ApproxNDCG, and stabilizes training via Converge-based Quad-Balancing to delay overfitting under distribution shifts. Across SEE50, CSI100, and CSI300 benchmarks with multiple backbones, MiM-StocR consistently improves ranking metrics and profitability, with notable gains such as approximately $11.6\%$ over the CSI300 index in backtests. The approach offers practical benefits for portfolio construction by focusing learning on top-ranked stocks and by providing robust optimization under volatile market conditions.

Abstract

Stock recommendation is critical in Fintech applications, which leverage price series and alternative information to estimate future stock performance. Traditional time-series forecasting training often fails to capture stock trends and rankings simultaneously, which are essential factors for investors. To tackle this issue, we introduce a Multi-Task Learning (MTL) framework for stock recommendation, \textbf{M}omentum-\textbf{i}ntegrated \textbf{M}ulti-task \textbf{Stoc}k \textbf{R}ecommendation with Converge-based Optimization (\textbf{MiM-StocR}). To improve the model's ability to capture short-term trends, we incorporate a momentum line indicator in model training. To prioritize top-performing stocks and optimize investment allocation, we propose a listwise ranking loss function called Adaptive-k ApproxNDCG. Moreover, due to the volatility and uncertainty of the stock market, existing MTL frameworks face overfitting issues when applied to stock time series. To mitigate this issue, we introduce the Converge-based Quad-Balancing (CQB) method. We conducted extensive experiments on three stock benchmarks: SEE50, CSI 100, and CSI 300. MiM-StocR outperforms state-of-the-art MTL baselines across both ranking and profitability evaluations.

Figures (7)

• Figure 1: Neither classification nor regression provides actionable guidance for profit-oriented stock selection.
• Figure 2: Overview of the MiM-StocR framework. It integrates momentum-based multi-task learning with Adaptive-k ApproxNDCG ranking loss and CQB optimization for robust stock recommendation.
• Figure 3: Training and validation losses of regression and classification tasks under single-task learning, showing clear overfitting.
• Figure 4: Cumulative returns on CSI300. MiM-StocR achieves the highest profit, outperforming the index and multi-task baselines.
• Figure 5: Train and test losses under different multi-task optimizers. CQB delay the overfitting compared with existing methods.