This work proposes Graph Layer for Inference in Dynamic En- vironments (GLIDE), a novel GNN layer enhanced by two theoretically grounded design mechanisms that significantly improve learning under dynamic topology while preserving robustness in static scenarios.
Abstract
Modeling multivariate time series by representing them as graphs, where individual series act as nodes and pairwise temporal corre- lations serve as edges, has gained significant traction. Recent advances in Graph Neural Networks (GNNs) have demonstrated strong perfor- mance by assuming a static graph topology and aggregating information from neighboring series. In this work, we investigate the representa- tional power of GNNs for forecasting under both static and dynamic settings (i.e., when pairwise correlations evolve drastically over time) and identify critical limitations in current architectures. To formalize this, we first propose Temporal Correlation Volatility (TCV), a model- agnostic metric designed to quantify the distributional evolution of these latent structures. We establish a clear connection between TCV and performance degradation, demonstrating that many popular models, including Transformers, generalize poorly in high-TCV settings and are often outperformed by simple structure-agnostic baselines. To address these limitations, we propose Graph Layer for Inference in Dynamic En- vironments (GLIDE), a novel GNN layer enhanced by two theoretically grounded design mechanisms: (D1) Path-based Message Passing, which captures path-based neighborhoods and (D2) Static and Dynamic Propagation Separation, which identifies optimal dynamics via local static approximation. These components significantly improve learning under dynamic topology while preserving robustness in static scenarios. Ex- tensive experiments on synthetic and real-world benchmarks show that GLIDE improves average performance by up to 45.6% across static and dynamic settings, with the largest gain reaching 85.7%. The source code is available at https://github.com/ChenS676/GLIDE.
The extreme non-stationarity, high noise levels, and multi-timescale coupling in financial futures markets pose major challenges for time series forecasting. Existing models often struggle to disentangle localized shocks from global trends due to incompatible inductive biases. To address this issue, we propose a Decomposition-Enhanced Network (DENet). Following a divide-and-conquer paradigm, DENet adopts a multi-stream architecture: the main path extracts stable trends via moving averages and dual-path linear projections, while Auxiliary Stream I captures seasonal and local cyclical patterns using depthwise separable convolutions, and Auxiliary Stream II models high-frequency dynamics through a nonlinear autoregressive-style mapping. These components are integrated via an adaptive fusion mechanism, balancing global robustness and local structural sensitivity. Experiments on real-world futures data demonstrate that DENet outperforms a wide range of state-of-the-art benchmarks. Compared with seven state-of-the-art baselines, DENet achieves an average reduction of 10.15% in RMSE for daily forecasting on iron ore futures, and an average reduction of 22.57% in MAE under the 5-min, 12-step forecasting horizon. Furthermore, we integrate DENet’s dual-scale predictions into the R-Breaker strategy with parameter switching and dynamic position sizing. Backtesting results reveal that the annualized return of iron ore futures outperforms the baseline strategy by an average of 7.6 percentage points across five backtesting windows. Ultimately, DENet effectively bridges advanced structural modeling and actionable algorithmic trading.
This work proposes CvLoss, a plug-in structural regularizer that constrains forecast residuals on a cross-variable graph and shows that CvLoss consistently improves competitive forecasting models, outperforms representative learning objectives, and is compatible with a variety of forecasting backbones.
Kuiye Ding, Yifan Hu, Hanchen Wang et al.· 0 citations
Understanding how links form and predicting future link states is of great importance in social, traffic, and many other complex temporal networks. These temporal networks are typically governed by multiple evolutionary mechanisms. However, existing dynamic graph representation methods, especially dynamic graph neural networks (DGNNs), are constrained by their single-evolutionary-path architecture, where spatial and temporal dynamics are either sequentially stacked or integrated into a monolithic module in a fixed manner. This design not only restricts the exploration of diverse evolutionary paths arising from rich evolutionary mechanisms but also confines spatiotemporal interactions to passive and implicit modeling, resulting in compromised performance and weak interpretability. To address these issues, we propose a novel theoretical model, namely the multiway autoregressive (MARS) model, which characterizes multiple evolutionary paths by capturing dependencies within and across two core factors underlying diverse evolutionary mechanisms. Based on this theoretical foundation, we develop a general DGNN framework, a multiway autoregressive network (MAN), by transforming the network architecture into a 2-D diagram that characterizes evolutionary dependencies in the spatiotemporal domain. Each node encodes an evolving state of the dynamic graph representation, while each edge denotes an evolutionary transition from one state to another. Moreover, three elementary evolutionary operators are incorporated into edges along distinct directions, capturing spatialwise, temporalwise, and cross-spatiotemporal evolutionary dynamics, respectively. This enables researchers to develop a variety of DGNNs by configuring the evolutionary operators in different ways. To validate the effectiveness of this framework, we design a new DGNN, which employs a graph convolutional network (GCN), a gated recurrent unit (GRU), and our proposed time-delayed GCN (TD-GCN) as core components. Promising experimental results demonstrate that the proposed approach achieves state-of-the-art temporal link prediction performance on both synthetic and real-world temporal networks across diverse domains.
Ping He, Xiao-hua Xu· IEEE Transactions on Neural...· 0 citations
Understanding temporal dynamics in complex systems often requires identifying abrupt structural changes, known as change points in multivariate time series. Traditional vector autoregressive (VAR) models have been widely used for modeling dependencies across time, yet their parameter space grows quadratically with the number of variables, leading to computational and estimation challenges in high‐dimensional settings. The recently proposed network autoregressive (NAR) modeling framework offers a computationally efficient alternative by reducing parameter complexity through a network‐based representation. However, existing NAR models either assume homogeneous temporal behavior across all nodes, overlooking node‐specific dynamics that frequently arise in environmental and socio‐economic systems, or do not allow for structural breaks. In this work, we propose
NLDNAR‐CP
, a novel change point detection method within a node‐specific NAR framework that accommodates heterogeneous temporal dependencies across variables. The proposed approach efficiently detects multiple structural breaks while preserving scalability to high‐dimensional networks. We demonstrate the method's superior empirical performance through extensive simulations and a real‐world environmental application.
