{"paper_id":"2506.02796","title":"Deep Learning Enhanced Multivariate GARCH","authors":["Haoyuan Wang","Chen Liu","Minh-Ngoc Tran","Chao Wang"],"paper_date":"2025-06-03","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2506.02796v1","page_url":"https://thequant.space/flowcharts/deep-learning-enhanced-multivariate-garch/","scores":{"empirical_rigor":7.5,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a novel multivariate volatility modeling framework called LSTM-BEKK that integrates Long Short-Term Memory networks with the BEKK GARCH model.","quote":"This paper introduces a novel multivariate volatility modeling framework, named Long Short-Term Memory enhanced BEKK (LSTM-BEKK), that integrates deep learning into multivariate GARCH processes.","summary":"This paper introduces a novel multivariate volatility modeling framework, named Long Short-Term Memory enhanced BEKK (LSTM-BEKK), that integrates deep learning into multivariate GARCH processes. By combining the flexibility of recurrent neural networks with the econometric structure of BEKK models…","strategy":"LSTM-BEKK"},"assumptions":{"status":"available","items":[{"assumption":"Parameters a and b are non‑negative and satisfy a + b < 1.","type":"model","stated":"explicit","quote":"To impose these conditions in practice, apart from the condition a, b ≥0, a + b < 1, it suffices to bound the maximum eigenvalue of CtC′t."},{"assumption":"The matrix norm of CtC′t is bounded almost surely for all t.","type":"model","stated":"explicit","quote":"assume that ∥CtC′t∥is bounded almost surely for all t."},{"assumption":"Daily log‑return series are scaled by 100 and treated as percentage terms.","type":"data","stated":"explicit","quote":"All returns are computed as daily log returns and then scaled by 100 to express them in percentage terms."},{"assumption":"The sample is split into 70% training, 15% validation, and 15% testing.","type":"data","stated":"explicit","quote":"each return dataset is partitioned into 70% for training, 15% for validation, and 15% for testing."},{"assumption":"The universe consists of the top 250 publicly traded equities by market capitalization in each market.","type":"market","stated":"explicit","quote":"daily log returns for the top 250 publicly traded equities from the United States (U.S.), the United Kingdom (U.K.), and Japan, selected based on market capitalization."}]},"datasets":{"status":"available","items":[{"name":"Daily log returns of top 250 equities (U.S., U.K., Japan)","provider":"Refinitiv","kind":"market-prices","access":"commercial","period":"U.S.: Mar 2014–Dec 2023; U.K.: Jul 2014–","universe":"Top 250 equities by market cap in each country","quote":"The data employed in this study comprises daily log returns for the top 250 publicly traded equities from the United States (U.S.), the United Kingdom (U.K.), and Japan, selected based on market capitalization. The data is sourced from Refinitiv data platform","known_id":"refinitiv"}],"known":[{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"The data is sourced from Refinitiv data platform: https://www.refinitiv.com, which aggregates information from key exchanges includ-ing NASDAQ OMX – NASDAQ BASIC, the New York Stock Exchange (NYSE), the London Stock Exchange (LSE), and the Tokyo Stock Exchange (TSE), thereby ensuring comprehensive and high-quality market coverage across major global financial centers.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":"U.S., U.K., Japan equity markets","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":"unclear","findings":[{"finding":"LSTM‑BEKK outperforms traditional BEKK and DCC models in predictive accuracy, measured by out‑of‑sample negative log‑likelihood and annualized volatility of the global minimum variance portfolio.","kind":"predictive-accuracy","quote":"Substantial empirical results demonstrate that the LSTM-BEKK model outperforms traditional BEKK and DCC models in terms of predictive accuracy, as measured by standard evaluation metrics such as out-of-sample negative log-likelihood and annualized"},{"finding":"Across U.S., U.K., and Japan equity markets, LSTM‑BEKK achieves superior out‑of‑sample portfolio risk forecasts compared to benchmark models.","kind":"performance","quote":"empirical results across multiple equity markets confirm that the LSTM-BEKK model achieves superior performance in terms of out-of-sample portfolio risk forecast"},{"finding":"The model remains robust and generalizes well across different market environments, as shown by consistent results on high‑dimensional portfolios.","kind":"performance","quote":"These findings are robust across datasets covering portfolios constructed from the top companies by market capitalization in Japan, the U.S., and the U.K., reflecting the model’s strong generalization capability across different market environments."}],"reported_performance":{"strategy":"LSTM-BEKK","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":37,"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":"2026-11-14T17:36:18Z"},"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/"}