Paper: arXiv 2309.06559
Authors: Luke Sanborn, Matthew Sahagun
Abstract
The financial industry poses great challenges with risk modeling and profit generation. These entities are intricately tied to the sophisticated prediction of stock movements. A stock forecaster must untangle the randomness and ever-changing behaviors of the stock market. Stock movements are influenced by a myriad of factors, including company history, performance, and economic-industry connections. However, there are other factors that aren’t traditionally included, such as social media and correlations between stocks. Social platforms such as Reddit, Facebook, and X (Twitter) create opportunities for niche communities to share their sentiment on financial assets. By aggregating these opinions from social media in various mediums such as posts, interviews, and news updates, we propose a more holistic approach to include these “media moments” within stock market movement prediction. We introduce a method that combines financial data, social media, and correlated stock relationships via a graph neural network in a hierarchical temporal fashion. Through numerous trials on current S&P 500 index data, with results showing an improvement in cumulative returns by 28%, we provide empirical evidence of our tool’s applicability for use in investment decisions.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 6.0/10 (capped from 6.5: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 6.5)
Why this score: The paper employs advanced deep learning architectures (GAT, LSTM) with mathematical formulations for attention mechanisms, while also presenting a detailed empirical setup with specific datasets, train/val/test splits, and backtesting results (28% improvement in cumulative returns).
Research Flowchart
flowchart TD Start["Research Goal<br>Predict stock movements using<br>non-traditional media data"] --> Inputs["Data Inputs<br>Financial Data + Social Media<br>Posts & Stock Correlations"] Inputs --> GNN["Graph Neural Network<br>Model Relationships between Stocks"] GNN --> HTL["Hierarchical Temporal Learning<br>Analyze Time-based Media & Price Patterns"] HTL --> Sentiment["Sentiment Analysis<br>Quantify Social Media Sentiment"] Sentiment --> Prediction["Stock Movement Classification<br>Up vs Down Prediction"] Prediction --> Outcome["Key Outcome<br>28% Improvement in Cumulative Returns<br>S&P 500 Validation"]
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 ATOMIC-SM, a graph attention network model that integrates financial data, social media, and stock relationships for stock movement classification.
Source sentence
We propose ATOMIC-SM, a model for stock movement prediction that builds upon prior work and focuses on a holistic overview of media moments.
Assumes 6
- The efficient market hypothesis is assumed as a basis for the model. market
Source sentence
Continuing on the efficient market hypothesis, we present ATOMIC-SM
- Trading strategy assumes buying the top 4 most confident positive predictions and selling at next day market open. execution
Source sentence
We adopted a trading strategy where we would buy the top 4 most confident positive price movement predictions. Those stocks were then sold at the market opening the next day.
- Only mid‑cap and large‑cap Nasdaq stocks (market cap $2B–$1T) are included; penny stocks are excluded. data
Source sentence
Since our data was collected between medium-market capitalization and large-market capitalization stocks, any penny stock was disqualified from being evaluated.
- Company relationships are updated yearly (Jan 2019, Jan 2020, Jan 2021) and assumed to reflect true inter‑company connections. model
Source sentence
Company relationships were updated per year of training data, reflecting changes between January 2019, January 2020, and January 2021 for those stocks.
Show 2 more
- Social media sentiment scores from SocialSentiment.io are assumed to accurately capture media moments. data
Source sentence
We used SocialSentiment.io for the social media score and activity for each respective stock in the dataset.
- The model’s performance is evaluated on a real‑world test set from March to December 2021, assuming no data leakage. statistical
Source sentence
For our real-world test set, we took the same 87 stocks the model was trained on and fed it data from March to December 2021. This distinct time period avoids any leaks from training/validation/testing data into the real-world dataset
Data
S&P 500 stocks · equities
- ATOMIC‑SM stock dataset market-prices unknown
Source sentence
We collected data for stocks with a market capitalization between two billion and one trillion dollars within the Nasdaq. ATOMIC‑SM was trained on data from January 2019 to March 2021 with approximately 600 trading days...
- SocialSentiment.io scores alternative unknown
Source sentence
We used SocialSentiment.io for the social media score and activity for each respective stock in the dataset.
- S&P 500 ETF (SPY) benchmark benchmark public
Source sentence
To provide a benchmark for our model's performance, we compared its results to those of the S&P 500 ETF known as SPY.
Costs
Not modeled
Finds 5
total return 28.0%
- Cumulative returns improved by 28% compared to baseline. performance
Source sentence
Through numerous trials on current S&P 500 index data, with results showing an improvement in cumulative returns by 28%, we provide empirical evidence of our tool's applicability for use in investment decisions.
- ATOMIC‑SM outperformed the market benchmark in a 9‑month real‑world test while having a slightly reduced Sharpe coefficient. performance
Source sentence
In our 9-month real-world test, ATOMIC‑SM outperformed the market benchmark while having a slightly reduced Sharpe coefficient.
- ATOMIC‑SM achieved higher F1, MCC, and accuracy than prior machine‑learning baselines. predictive-accuracy
Source sentence
As Table 3 shows, ATOMIC‑SM achieved a greater performance than prior MLMs (Machine Learning Models) on the baselines of F1, MCC, and accuracy metrics.
- Including media moments from multiple social platforms boosted model performance over baselines. performance
Source sentence
In table 2, we observed that including these media moments boasted better performance than the baselines.
- The model’s MCC metric was used to handle class imbalance, reflecting balanced predictive quality. statistical
Source sentence
In our evaluation of the ATOMIC‑SM model, we used the Matthews Correlation Coefficient (MCC) in addition to the F1 score.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 32 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.