RAVEN: A Regime-Aware Variable-context Expert Network for Financial Time Series Forecasting

2026-06-22

AI Summary

Financial time series forecasting is challenged by non-stationarity, low signal-to-noise ratios, and state-dependent temporal dynamics, which hinder fixed-window models from adaptively capturing the optimal lookback horizon. To address this, this work proposes the RAVEN framework, which innovatively integrates a Cumulative Importance Threshold (CIT) mechanism to dynamically construct nested variable-length prefix windows and employs a Mixture-of-Experts architecture to route sequences to scale-specific experts. Additionally, RAVEN incorporates a Global Compressed Representation (GCR) to preserve long-term consistency and introduces a Correlation-Aware Weighting (CAW) strategy to fuse multi-scale predictions. Experiments demonstrate that RAVEN improves Pearson correlation coefficients by 9.2% and 20.2% on cumulative log-return prediction for HS300 and S&P500, respectively, reduces MSE by 18.2% in fund sales forecasting, and achieves state-of-the-art results on 14 out of 16 traffic prediction benchmarks.

Abstract

Financial time series forecasting presents structural challenges absent from standard benchmarks. Log-returns are non-stationary, exhibit exceptionally low signal-to-noise (SNR) ratios, and are governed by regime-dependent temporal dependencies. We identify a key limitation of state-of-the-art (SOTA) time series models in financial settings. A fixed context window is mismatched to the time-varying optimal look-back of non-stationary price processes. We propose the Regime-Aware Variable-context Expert Network (RAVEN), a Mixture-of-Experts framework designed to adaptively determine the temporal context for each input sample. Instead of relying on a fixed look-back horizon, RAVEN constructs a hierarchy of nested contiguous windows whose lengths are determined by the data itself. Specifically, RAVEN scores patches by learned importance in reverse chronological order and applies the Cumulative Importance Thresholding (CIT) mechanism to derive nested prefix windows, each routed to a scale-specialized expert. A Global Compressed Representation (GCR) branch runs in parallel over the full context, preserving global temporal coherence that local experts cannot guarantee. Because the nested routing induces structured overlap among expert inputs, we introduce a Correlation-Aware Weighting (CAW) to align variable-length expert outputs and penalize pairwise cosine similarity prior to aggregation. Experiments on cumulative log-return prediction (HS300, S&P500) and fund sales forecasting demonstrate that RAVEN achieves SOTA performances, improves Pearson correlation by 9.2% on HS300 and 20.2% on S&P500, and reduces MSE by 18.2% on fund sales forecasting, while achieving the best results in 14 of 16 metrics on four PEMS traffic benchmarks.

Problem

• financial time series forecasting
• non-stationarity
• regime-dependent dependencies
• variable context
• low signal-to-noise ratio

Innovation

• Variable-context
• Mixture-of-Experts
• Cumulative Importance Thresholding
• Regime-aware
• Correlation-Aware Weighting