{"paper_id":"2309.06559","title":"Media Moments and Corporate Connections: A Deep Learning Approach to Stock Movement Classification","authors":["Luke Sanborn","Matthew Sahagun"],"paper_date":"2023-09-08","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2309.06559v1","page_url":"https://thequant.space/flowcharts/media-moments-and-corporate-connections-a-deep-learning-app/","scores":{"empirical_rigor":6,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes ATOMIC-SM, a graph attention network model that integrates financial data, social media, and stock relationships for stock movement classification.","quote":"We propose ATOMIC-SM, a model for stock movement prediction that builds upon prior work and focuses on a holistic overview of media moments.","summary":"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…","strategy":"graph neural network hierarchical temporal model"},"assumptions":{"status":"available","items":[{"assumption":"The efficient market hypothesis is assumed as a basis for the model.","type":"market","stated":"explicit","quote":"Continuing on the efficient market hypothesis, we present ATOMIC-SM"},{"assumption":"Trading strategy assumes buying the top 4 most confident positive predictions and selling at next day market open.","type":"execution","stated":"explicit","quote":"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."},{"assumption":"Only mid‑cap and large‑cap Nasdaq stocks (market cap $2B–$1T) are included; penny stocks are excluded.","type":"data","stated":"explicit","quote":"Since our data was collected between medium-market capitalization and large-market capitalization stocks, any penny stock was disqualified from being evaluated."},{"assumption":"Company relationships are updated yearly (Jan 2019, Jan 2020, Jan 2021) and assumed to reflect true inter‑company connections.","type":"model","stated":"explicit","quote":"Company relationships were updated per year of training data, reflecting changes between January 2019, January 2020, and January 2021 for those stocks."},{"assumption":"Social media sentiment scores from SocialSentiment.io are assumed to accurately capture media moments.","type":"data","stated":"explicit","quote":"We used SocialSentiment.io for the social media score and activity for each respective stock in the dataset."},{"assumption":"The model’s performance is evaluated on a real‑world test set from March to December 2021, assuming no data leakage.","type":"statistical","stated":"explicit","quote":"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"}]},"datasets":{"status":"available","items":[{"name":"ATOMIC‑SM stock dataset","provider":null,"kind":"market-prices","access":"unknown","period":"January 2019 to March 2021 (training) an","universe":"Nasdaq stocks with market cap $2B–$1T","quote":"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..."},{"name":"SocialSentiment.io scores","provider":"SocialSentiment.io","kind":"alternative","access":"unknown","period":null,"universe":"same stock universe as above","quote":"We used SocialSentiment.io for the social media score and activity for each respective stock in the dataset."},{"name":"S&P 500 ETF (SPY) benchmark","provider":"public index fund","kind":"benchmark","access":"public","period":null,"universe":"S&P 500 index","quote":"To provide a benchmark for our model's performance, we compared its results to those of the S&P 500 ETF known as SPY."}],"known":[],"sample":{"start":null,"end":null,"market":"S&P 500 stocks","asset_class":"equities","frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Cumulative returns improved by 28% compared to baseline.","kind":"performance","quote":"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."},{"finding":"ATOMIC‑SM outperformed the market benchmark in a 9‑month real‑world test while having a slightly reduced Sharpe coefficient.","kind":"performance","quote":"In our 9-month real-world test, ATOMIC‑SM outperformed the market benchmark while having a slightly reduced Sharpe coefficient."},{"finding":"ATOMIC‑SM achieved higher F1, MCC, and accuracy than prior machine‑learning baselines.","kind":"predictive-accuracy","quote":"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."},{"finding":"Including media moments from multiple social platforms boosted model performance over baselines.","kind":"performance","quote":"In table 2, we observed that including these media moments boasted better performance than the baselines."},{"finding":"The model’s MCC metric was used to handle class imbalance, reflecting balanced predictive quality.","kind":"statistical","quote":"In our evaluation of the ATOMIC‑SM model, we used the Matthews Correlation Coefficient (MCC) in addition to the F1 score."}],"reported_performance":{"strategy":"graph neural network hierarchical temporal model","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":28,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":32,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T18:22:27Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}