{"paper_id":"2402.06642","title":"From GARCH to Neural Network for Volatility Forecast","authors":["Pengfei Zhao","Haoren Zhu","Wilfred Siu Hung NG","Dik Lun Lee"],"paper_date":"2024-01-29","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2402.06642v1","page_url":"https://thequant.space/flowcharts/from-garch-to-neural-network-for-volatility-forecast/","scores":{"empirical_rigor":7,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a GARCH-NN approach that derives neural network equivalents of GARCH models and integrates them into NN architectures, exemplified by a GARCH‑LSTM model.","quote":"we introduce an innovative approach, named GARCH-NN, for constructing NN-based volatility models.","summary":"Volatility, as a measure of uncertainty, plays a crucial role in numerous financial activities such as risk management. The Econometrics and Machine Learning communities have developed two distinct approaches for financial volatility forecasting: the stochastic approach and the neural network (NN)…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Stationarity of the GARCH process is ensured by 0 < α + β < 1.","type":"model","stated":"explicit","quote":"0 < α + β < 1 guarantees the stationarity of the GARCH process."},{"assumption":"The error distribution D is typically assumed to be normal or leptokurtic.","type":"statistical","stated":"explicit","quote":"(typically assumed to be a normal or a leptokurtic one)"},{"assumption":"Log returns are computed as the logarithmic difference of daily close prices.","type":"data","stated":"explicit","quote":"rt = log pt pt−1 denotes the log return at t"},{"assumption":"The time series are stationary, as indicated by near‑zero p‑values in ADF tests and KPSS tests.","type":"statistical","stated":"explicit","quote":"The near-zero p-value rejects the null hypothesis that a unit root is present in the time series data, indicating the stationarity of the time series."},{"assumption":"Training, validation, and testing splits are roughly in an 8:1:1 ratio.","type":"execution","stated":"explicit","quote":"We split the complete dataset into training, validation, and testing parts and the split ratio is roughly 8:1:1."},{"assumption":"Returns and realized volatility are multiplied by a factor of 100 to avoid numeric underflow.","type":"data","stated":"explicit","quote":"To avoid numeric underflow, we multiply the ϵt and σ2 t+h by a factor 100."}]},"datasets":{"status":"available","items":[{"name":"S&P 500","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"S&P 500 index","quote":"Data were downloaded from https://finance.yahoo.com/","known_id":"yahoo-finance"},{"name":"DJI","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"Dow Jones Industrial Average","quote":"Data were downloaded from https://finance.yahoo.com/","known_id":"yahoo-finance"},{"name":"NASDAQ","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"NASDAQ Composite","quote":"Data were downloaded from https://finance.yahoo.com/","known_id":"yahoo-finance"},{"name":"EUR-USD","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"Euro to US Dollar exchange rate","quote":"Data were downloaded from https://finance.yahoo.com/","known_id":"yahoo-finance"},{"name":"Gold","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":null,"universe":"Gold price","quote":"Data were downloaded from https://finance.yahoo.com/","known_id":"yahoo-finance"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"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":"Amalgamating NN counterparts of GARCH models into NN architectures improves volatility forecasting compared to using stochastic or NN models alone.","kind":"performance","quote":"Experiment results validate that amalgamating the NN counterparts of the GARCH family models into established NN models leads to enhanced outcomes compared to employing the stochastic and NN models in isolation."},{"finding":"T‑loss with freedom (v=5) provides the best performance for training GARCH‑LSTM across most datasets.","kind":"performance","quote":"the T‑loss with freedom (v=5) has the best performance in most cases, thus we select it as the loss function for training GARCH‑LSTM"},{"finding":"GARCH‑LSTM outperforms other deep learning models (Autoformer, Informer, Reformer, Transformer) on 1‑day horizon for S&P 500 (MAE 0.205 vs 0.636‑0.494).","kind":"performance","quote":"S&P 500 1D 0.205 0.070 ... 0.636 0.546 ... 0.494 0.387"},{"finding":"Scaling returns and volatility by 100 avoids numeric underflow without degrading forecast quality.","kind":"other","quote":"To avoid numeric underflow, we multiply the ϵt and σ2 t+h by a factor 100."}],"reported_performance":null,"out_of_sample":{"method":"walk-forward","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[],"contradicts":[],"qualifies":[],"compared_with":25,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":1,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[{"paper":{"paper_id":"2504.09380","title":"Unified GARCH-Recurrent Neural Network in Financial Volatility Forecasting","paper_date":"2025-04-13","authors":["Jingyi Wei","Steve Yang","Zhenyu Cui"],"empirical_rigor":8.5,"page_url":"https://thequant.space/flowcharts/unified-garch-recurrent-neural-network-in-financial-volatili/"},"how":"uses_method_or_data","quote":null}],"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":"2026-11-14T18:15:45Z"},"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/"}