Paper: arXiv 2503.08696
Authors: Kasymkhan Khubiev, Mikhail Semenov
Abstract
Classical asset price forecasting methods primarily rely on numerical data, such as price time series, trading volumes, limit order book data, and technical analysis indicators. However, the news flow plays a significant role in price formation, making the development of multimodal approaches that combine textual and numerical data for improved prediction accuracy highly relevant. This paper addresses the problem of forecasting financial asset prices using the multimodal approach that combines candlestick time series and textual news flow data. A unique dataset was collected for the study, which includes time series for 176 Russian stocks traded on the Moscow Exchange and 79,555 financial news articles in Russian. For processing textual data, pre-trained models RuBERT and Vikhr-Qwen2.5-0.5b-Instruct (a large language model) were used, while time series and vectorized text data were processed using an LSTM recurrent neural network. The experiments compared models based on a single modality (time series only) and two modalities, as well as various methods for aggregating text vector representations. Prediction quality was estimated using two key metrics: Accuracy (direction of price movement prediction: up or down) and Mean Absolute Percentage Error (MAPE), which measures the deviation of the predicted price from the true price. The experiments showed that incorporating textual modality reduced the MAPE value by 55%. The resulting multimodal dataset holds value for the further adaptation of language models in the financial sector. Future research directions include optimizing textual modality parameters, such as the time window, sentiment, and chronological order of news messages.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 9.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced deep learning architectures (LSTMs, pre-trained LLMs like RuBERT/Qwen) for multimodal fusion, representing substantial mathematical modeling complexity, and demonstrates empirical rigor with a custom dataset of 176 stocks and 79,555 news articles, rigorous backtesting metrics (Accuracy and MAPE), and a clear 55% MAPE improvement claim.
Research Flowchart
flowchart TD A["Research Goal: Forecast Russian stock prices using multimodal data"] --> B["Data Collection"] B --> C["Modality Processing"] C --> D["Model Training & Prediction"] D --> E["Evaluation & Findings"] B --> B1["(176 Stocks: Candlestick Time Series)"] B --> B2["(79,555 News Articles: Textual Data)"] C --> C1["Text Embeddings: RuBERT & Vikhr-Qwen2.5 LLM"] C --> C2["Time Series: LSTM Processing"] D --> D1["Multimodal Fusion"] D1 --> D2["LSTM Model Training"] E --> E1["Metrics: Accuracy & MAPE"] E --> E2["Key Result: 55% MAPE Reduction"] E --> E3["Outcome: Validated Multimodal Dataset"]
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 new multimodal approach that integrates news text vectors with candlestick time series for asset price forecasting, and introduces a Russian-language multimodal dataset.
Source sentence
We propose a new multimodal approach for integrating news flow into time series numerical data.
Assumes 2
- Closing price predictions are made for each trading day at market opening using only news published before the start of that day. market
Source sentence
We assume that the closing price prediction for an asset is made for each trading day at the market opening.
- If no news articles are available for the previous day, a zero vector is used in the multimodal input. data
Source sentence
If no publications were available on the previous day or before the market opened on the current day, a zero vector was concatenated with the relative price return vector
Data
- Multimodal Russian stock and news dataset other proprietary
Source sentence
A unique dataset was collected for the study, which includes time series for 176 Russian stocks traded on the Moscow Exchange and 79,555 financial news articles in Russian.
Costs
Not applicable
Finds 2
- Incorporating textual news modality reduced the Mean Absolute Percentage Error (MAPE) by 55% compared to using only time series data. predictive-accuracy
Source sentence
The experiments showed that incorporating textual modality reduced the MAPE value by 55%.
- The multimodal approach with vector averaging outperformed the best single-modality results across all metrics except MAE on NFLX and R2 on TSLA. performance
Source sentence
Across all metrics, except for MAE on NFLX and R2 on TSLA, the proposed multimodal approach with vector averaging outperformed the best-performing results from the approach in [7].
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 10 related papers). Coverage is partial, so this is neither replication nor refutation.
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.