Paper: arXiv 2508.02686
Authors: Diego Vallarino
Abstract
This study develops and empirically validates a Mixture of Experts (MoE) framework for stock price prediction across heterogeneous volatility regimes using real market data. The proposed model combines a Recurrent Neural Network (RNN) optimized for high-volatility stocks with a linear regression model tailored to stable equities. A volatility-aware gating mechanism dynamically weights the contributions of each expert based on asset classification. Using a dataset of 30 publicly traded U.S. stocks spanning diverse sectors, the MoE approach consistently outperforms both standalone models. Specifically, it achieves up to 33% improvement in MSE for volatile assets and 28% for stable assets relative to their respective baselines. Stratified evaluation across volatility classes demonstrates the model’s ability to adapt complexity to underlying market dynamics. These results confirm that no single model suffices across market regimes and highlight the advantage of adaptive architectures in financial prediction. Future work should explore real-time gate learning, dynamic volatility segmentation, and applications to portfolio optimization.
Complexity vs Empirical Score
- Math Complexity: 4.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper uses relatively straightforward mathematical frameworks (Mixture of Experts, LSTM equations, GARCH) without deep derivations, placing it on the lower end of math complexity. However, it demonstrates strong empirical rigor through backtesting on real market data (30 U.S. stocks), explicit performance metrics (MSE improvements), and a commitment to reproducibility via a GitHub repository.
Research Flowchart
flowchart TD
A["Research Goal: Adaptive Stock Forecasting<br>for Volatile vs. Stable Regimes"] --> B["Methodology: Mixture of Experts (MoE)<br>Volatility-Aware Gating"]
B --> C["Data: 30 U.S. Stocks<br>Diverse Sectors"]
C --> D{"Dynamic Classification:<br>Volatility Regime?"}
D -- High Volatility --> E["RNN Expert<br>Deep Temporal Patterns"]
D -- Stable/Low Volatility --> F["Linear Regression Expert<br>Baseline Simplicity"]
E --> G["Output: Weighted Prediction"]
F --> G
G --> H["Findings: Adaptive MoE Outperforms Baselines<br>+33% MSE (Volatile), +28% MSE (Stable)"]
```mermaid
flowchart TD
A["Research Goal: Adaptive Stock Forecasting<br>for Volatile vs. Stable Regimes"] --> B["Methodology: Mixture of Experts (MoE)<br>Volatility-Aware Gating"]
B --> C["Data: 30 U.S. Stocks<br>Diverse Sectors"]
C --> D{"Dynamic Classification:<br>Volatility Regime?"}
D -- High Volatility --> E["RNN Expert<br>Deep Temporal Patterns"]
D -- Stable/Low Volatility --> F["Linear Regression Expert<br>Baseline Simplicity"]
E --> G["Output: Weighted Prediction"]
F --> G
G --> H["Findings: Adaptive MoE Outperforms Baselines<br>+33% MSE (Volatile), +28% MSE (Stable)"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a Mixture of Experts (MoE) framework that combines a Recurrent Neural Network for volatile stocks with a linear regression model for stable stocks, using a volatility‑aware gating mechanism.
Source sentence
This study develops and empirically validates a Mixture of Experts (MoE) framework for stock price prediction across heterogeneous volatility regimes using real market data.
Assumes 3
- The gating function uses ex‑ante asset volatility classification to weight expert contributions. model
Source sentence
The gating function uses ex‑ante asset volatility classification to weight the contributions of each expert in the final prediction.
- The static gating mechanism assigns fixed weights (70% RNN for volatile, 70% linear for stable) throughout the test period. model
Source sentence
The Mixture of Experts model—implemented with a static gating mechanism assigning 70% weight to the RNN for volatile assets and 70% to the linear model for stable assets—
- Training and validation use early stopping based on validation MAE with a patience of five epochs. model
Source sentence
To prevent overfitting, RNN training uses early stopping based on validation MAE, with a patience threshold of five epochs.
Data
- Daily stock prices for 30 publicly traded U.S. firms market-prices public
Source sentence
The dataset used in this study comprises daily stock prices for 30 publicly traded companies from the S&P 500 index, retrieved from Yahoo Finance using the yfinance Python library. The time period spans from January 1, 2015 to December 31, 2024, providing approximately 500 trading days for each stock.
Costs
Not applicable
Finds 2
- The MoE model reduces MSE for volatile firms by 32.99% (to 0.001105) and improves MAE by 18.62% compared to the best individual model. predictive-accuracy
Source sentence
For volatile firms, the MoE reduced the MSE to 0.001105 and the MAE to 0.026333, yielding performance gains of 32.99% and 18.62%, respectively, over the best individual model.
- In the stable regime, the MoE achieves lower errors than either expert alone, with up to 28% MSE improvement relative to the linear baseline. predictive-accuracy
Source sentence
Results demonstrate that the MoE framework achieves significantly lower prediction errors—up to 33% in MSE improvement for volatile stocks and 28% for stable stocks—compared to linear or recurrent models alone.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.