Paper: arXiv 2410.07234
Authors: Diego Vallarino
Abstract
This study evaluates the effectiveness of a Mixture of Experts (MoE) model for stock price prediction by comparing it to a Recurrent Neural Network (RNN) and a linear regression model. The MoE framework combines an RNN for volatile stocks and a linear model for stable stocks, dynamically adjusting the weight of each model through a gating network. Results indicate that the MoE approach significantly improves predictive accuracy across different volatility profiles. The RNN effectively captures non-linear patterns for volatile companies but tends to overfit stable data, whereas the linear model performs well for predictable trends. The MoE model’s adaptability allows it to outperform each individual model, reducing errors such as Mean Squared Error (MSE) and Mean Absolute Error (MAE). Future work should focus on enhancing the gating mechanism and validating the model with real-world datasets to optimize its practical applicability.
Complexity vs Empirical Score
- Math Complexity: 5.5/10
- Empirical Rigor: 4.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper introduces advanced ML concepts like Mixture of Experts and RNNs/LSTMs with explicit formula derivations, indicating moderate mathematical density. However, it relies on simulated data rather than real-world backtesting, lacks code/dataset references, and explicitly mentions future validation is needed, limiting its empirical readiness.
Research Flowchart
flowchart TD
A["Research Goal: Compare MoE, RNN, and Linear Regression<br>for Stock Price Prediction"] --> B["Input Data: Stocks with<br>Different Volatility Profiles"]
B --> C["Methodology: Dynamic MoE Framework<br>Gate + Expert Model Selection"]
C --> D{"Compute & Train Models"}
D --> E["Expert 1: RNN Model<br>For Volatile Stocks"]
D --> F["Expert 2: Linear Regression<br>For Stable Stocks"]
E --> G["Key Outcomes: MoE outperforms<br>individual models (Lower MSE/MAE)"]
F --> G
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) model that combines an RNN for volatile stocks and a linear model for stable stocks, using a gating network to dynamically weight each expert.
Source sentence
The main contribution of this paper lies in the application and evaluation of the MoE model for stock price prediction, which integrates both machine learning and statistical approaches.
Assumes 4
- Stock prices follow a random walk with drift and normally distributed error. data
Source sentence
The prices were generated using a random walk with drift, where the price at time t was determined as: 𝑃𝑡= 𝑃𝑡−1 + μ + ϵ𝑡, where
- Companies with volatility σ greater than 0.05 are classified as volatile; others are stable. data
Source sentence
(σthreshold= 0.05) were classified as volatile, while those with lower volatility values were classified as stable.
- Model performance is evaluated using walk‑forward validation with an 80‑day training window and 20‑day validation window. execution
Source sentence
training window consisted of 80 days, followed by a validation window of 20 days. After each validation, the training window was incremented by 20 days, and the model was retrained and tested on the subsequent window.
- Performance metrics are limited to RMSE and MAE. model
Source sentence
To provide a comprehensive assessment of the models' performance, we used two primary metrics: Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE).
Data
- Synthetic dataset of 100 companies over 100 trading days synthetic unknown
Source sentence
The dataset comprised 100 companies over a span of 100 trading days, designed to mimic real‑world price dynamics with both stable and highly volatile behaviors.
Costs
Unclear
Finds 3
- The RNN model on volatile companies yields MSE = 2.4265 and MAE = 1.4678, indicating high error on extreme variations. predictive-accuracy
Source sentence
The Mean Squared Error (MSE) was calculated to be 2.4265, while the Mean Absolute Error (MAE) was 1.4678.
- The MoE approach significantly improves prediction accuracy over individual models by adapting to volatility profiles. performance
Source sentence
The MoE approach significantly improves prediction accuracy over using individual models alone by effectively adapting to different volatility conditions.
- The linear model performs well for stable companies, avoiding over‑fitting and delivering lower error than the RNN on such data. predictive-accuracy
Source sentence
The linear model, in contrast, shows excellent performance for non‑volatile stocks, reinforcing the notion that simpler models are often better suited for stable datasets.
Checked by others
Supported by
- Volatility Forecasting in Global Financial Markets Using TimeMixer
Compare the findings
This paper: The RNN model on volatile companies yields MSE = 2.4265 and MAE = 1.4678, indicating high error on extreme variations.
That paper: Highly volatile assets such as DOGE/USD and SOL/USD exhibit notably higher errors, especially over longer horizons.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (11)
- A Survey on Mixture of Experts in Large Language Models (2024) — link
- Multi-Head Mixture-of-Experts (2024) — link
- Online Learner Categories Recognition Based on MoE-LSTM Model (2024) — link
- Optimal portfolio diversification with a multi-chain regime-switching spillover GARCH model (2023) — link
- Predicting stock market index using LSTM (2022) — link
- Research on financial assets transaction prediction model based on LSTM neural network (2020) — link
- Hierarchical Gating Networks for Sequential Recommendation (2019) — link
- Forecasting Economics and Financial Time Series: ARIMA vs. LSTM (2018) — link
and 3 more
0 of 11 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.