{"paper_id":"2309.16196","title":"Stock Volatility Prediction Based on Transformer Model Using Mixed-Frequency Data","authors":["Wenting Liu","Zhaozhong Gui","Guilin Jiang","Lihua Tang","Lichun Zhou","Wan Leng","Xulong Zhang","Yujiang Liu"],"paper_date":"2023-09-28","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2309.16196v1","page_url":"https://thequant.space/flowcharts/stock-volatility-prediction-based-on-transformer-model-using/","scores":{"empirical_rigor":6,"math_complexity":5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a Transformer model for stock volatility prediction that incorporates mixed‑frequency macroeconomic data and investor attention indicators via a GARCH‑MIDAS preprocessing step.","quote":"The outputs of the GARCH-MIDAS model are deployed to train the Transformer model, which is one of the innovations of this paper.","summary":"With the increasing volume of high-frequency data in the information age, both challenges and opportunities arise in the prediction of stock volatility. On one hand, the outcome of prediction using tradition method combining stock technical and macroeconomic indicators still leaves room for…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The conditional mathematical expectation of the return at the (i‑1)‑th moment is µ and the short‑term component follows a GARCH(1,1) process.","type":"model","stated":"explicit","quote":"Assuming that the conditional mathematical expectation of ri,t at the (i-1)-th moment is µ and that the short-term component of the returns follows a GARCH(1, 1) process"},{"assumption":"The lag period K of the GARCH‑MIDAS model is set to 12 after repeated trials.","type":"model","stated":"explicit","quote":"The optimal value of the lag period K of the GARCH-MIDAS model is decided to be 12 after repeated trials."},{"assumption":"Data are split into training and testing sets with a 9:1 ratio.","type":"data","stated":"explicit","quote":"To prove the validity of model construction, the prepared data is split into training and testing data with a 9:1 ratio."},{"assumption":"Ten macroeconomic indicators are selected via grey correlation analysis to avoid subjective selection and redundancy.","type":"data","stated":"explicit","quote":"ten representative indicators are finally selected through grey correlation analysis to eliminate the influence of subjective selection and information redundancy"},{"assumption":"Missing values are filled and data are normalized before model construction.","type":"data","stated":"explicit","quote":"pre-processing of all data is required before model construction for later use and analysis. The pre-processing mainly includes two aspects: missing value filling and data normalization."}]},"datasets":{"status":"available","items":[{"name":"Macro‑economic indicators","provider":null,"kind":"macro","access":"unknown","period":"Jan 2011 to Sep 2021","universe":null,"quote":"The data to be fed into the GARCH-MIDAS model contains macroeconomic indicators and returns of CSI300, which are collected from Jan 2011 to Sep 2021 on a monthly and daily basis, respectively."},{"name":"CSI300 daily returns","provider":null,"kind":"market-prices","access":"unknown","period":"Jan 2011 to Sep 2021","universe":"CSI300 index","quote":"The data to be fed into the GARCH-MIDAS model contains macroeconomic indicators and returns of CSI300, which are collected from Jan 2011 to Sep 2021 on a monthly and daily basis, respectively."},{"name":"Baidu search indices (CSI300, CSI500, SSE50)","provider":"Baidu","kind":"alternative","access":"unknown","period":null,"universe":null,"quote":"Subjective Factors Attention Indicators “CSI300” Baidu Search Index CSI300 “CSI500” Baidu Search Index CSI500 “SSE50” Baidu Search Index SSE50"}],"known":[{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"Their analysis encompassed an ex-tensive dataset of over 24 million news records sourced from Thomson Reuters, examining their correlations with trading behaviors within the S&P US Index’s prominent 206 stocks.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"Incorporating both macroeconomic and subjective attention indicators reduces the mean square error from 1.00 to 0.86.","kind":"predictive-accuracy","quote":"The adaption of both types of data under Transformer model significantly reduces the mean square error from 1.00 to 0.86."},{"finding":"The full indicator set (macro, attention, and technical) achieves the lowest MSE (0.8624) among tested groups.","kind":"performance","quote":"G4 0.8624 0.4460 0.5871 0.4787 1.3620 0.1710"},{"finding":"The Transformer model outperforms LSTM, CNN, XGBoost, and GRU on most loss functions, achieving the best MSE (0.8624).","kind":"performance","quote":"According to the results in Table 6, the Transformer model performs better than others commonly used for volatility prediction. It has the best results in terms of MSE, HMSE and MAPE loss functions"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":22,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":1},"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":"2026-11-14T18:21:30Z"},"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/"}