DeltaLag: Learning Dynamic Lead-Lag Patterns in Financial Markets

TL;DR

DeltaLag addresses the dynamic lead-lag problem in financial markets by learning pair-specific, time-varying lags using an end-to-end neural architecture with sparsified cross-attention. The model identifies leader stocks and their optimal lags daily, then uses raw leader features at those lags to predict the lagger's next-day return via a simple MLP, trained with a ranking-oriented loss for improved portfolio performance. Empirical results show DeltaLag outperforms fixed-lag and self-lead-lag baselines and surpasses precomputed statistical graphs, while offering enhanced interpretability through explicit leader selections and lag values. The work demonstrates the practical value of dynamic, cross-asset lead-lag signals for trading and suggests avenues for integrating richer data and intraday dynamics.

Abstract

The lead-lag effect, where the price movement of one asset systematically precedes that of another, has been widely observed in financial markets and conveys valuable predictive signals for trading. However, traditional lead-lag detection methods are limited by their reliance on statistical analysis methods and by the assumption of persistent lead-lag patterns, which are often invalid in dynamic market conditions. In this paper, we propose \textbf{DeltaLag}, the first end-to-end deep learning method that discovers and exploits dynamic lead-lag structures with pair-specific lag values in financial markets for portfolio construction. Specifically, DeltaLag employs a sparsified cross-attention mechanism to identify relevant lead-lag pairs. These lead-lag signals are then leveraged to extract lag-aligned raw features from the leading stocks for predicting the lagger stock's future return. Empirical evaluations show that DeltaLag substantially outperforms both fixed-lag and self-lead-lag baselines. In addition, its adaptive mechanism for identifying lead-lag relationships consistently surpasses precomputed lead-lag graphs based on statistical methods. Furthermore, DeltaLag outperforms a wide range of temporal and spatio-temporal deep learning models designed for stock prediction or time series forecasting, offering both better trading performance and enhanced interpretability.

Figures (3)

• Figure 1: Architecture overview
• Figure 2: Cumulative returns comparison of our lead-lag model against various temporal and spatio-temporal models.
• Figure 3: Distribution of Lead-Lag pairs by lag value