{"paper_id":"2311.04727","title":"Forecasting Volatility with Machine Learning and Rough Volatility: Example from the Crypto-Winter","authors":["Siu Hin Tang","Mathieu Rosenbaum","Chao Zhou"],"paper_date":"2023-11-08","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2311.04727v2","page_url":"https://thequant.space/flowcharts/forecasting-volatility-with-machine-learning-and-rough-volat/","scores":{"empirical_rigor":6.5,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a universal LSTM volatility forecaster trained on pooled cryptocurrency data and a parsimonious rough‑volatility parametric model with a Zumbach effect.","quote":"We first show that to forecast volatility, a universal LSTM approach trained on a pool of assets outperforms traditional models. We then consider a parsimonious parametric model based on rough volatility and Zumbach effect.","summary":"We extend the application and test the performance of a recently introduced volatility prediction framework encompassing LSTM and rough volatility. Our asset class of interest is cryptocurrencies, at the beginning of the “crypto-winter” in 2022. We first show that to forecast volatility, a…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Cryptocurrencies trade every day, so HAR windows are set to 7 and 30 days.","type":"market","stated":"explicit","quote":"Due to the fact that cryptocurrencies are traded every day, for the HAR, the windows are chosen to be 7 and 30 instead of 5 and 22 in the literature on traditional markets."},{"assumption":"Only coins that are not stablecoins, leveraged, low‑liquidity or have insufficient data are kept.","type":"data","stated":"explicit","quote":"We select 213 coins after filtering out coins that are stablecoins, leveraged, or have low liquidity or insufficient data."},{"assumption":"Training data start from 2020; earlier data are excluded.","type":"data","stated":"explicit","quote":"We choose not to use training data from before 2020."},{"assumption":"Training period is 2020‑01‑01 to 2021‑12‑31; test period is 2022‑01‑01 to 2022‑06‑30.","type":"data","stated":"explicit","quote":"The training set is chosen to be the two-year period from 2020-01-01 to 2021-12-31, and the test set is a six-month period from 2022-01-01 to 2022-06-30."},{"assumption":"Volatility forecasts are evaluated with mean‑squared‑error (MSE).","type":"statistical","stated":"explicit","quote":"For model evaluation, we choose to use the mean-squared-error or MSE, calculated as MSE(σ, ˆσ) = 1/T"},{"assumption":"Volatilities and returns are scaled and normalized by each coin’s training‑set statistics.","type":"data","stated":"explicit","quote":"The volatilities and returns are scaled and normalized respectively by the individual coin’s statistics calculated on the training set."}]},"datasets":{"status":"available","items":[{"name":"Binance 5‑minute intraday price data","provider":"Binance.com","kind":"market-prices","access":"public","period":"2020‑01‑01 to 2022‑06‑30","universe":"213 selected cryptocurrencies (excluding stablecoins, leveraged, low‑liquidity)","quote":"Our dataset contains 5-minute intraday prices of cryptocurrencies traded on the Binance exchange, with USDT, a popularly used stablecoin, as the base currency.","known_id":"binance"}],"known":[{"id":"binance","name":"Binance","provider":"Binance","kind":"market-prices","access":"public","quote":"2 Data and evaluation metrics Our dataset contains 5-minute intraday prices of cryptocurrencies traded on the Binance exchange, with USDT, a popularly used stablecoin, as the base currency.","mentions":3,"used_in":2}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"intraday-bars"},"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":"Universal LSTM models outperform the HAR baseline for most coins.","kind":"performance","quote":"LSTMs outperform HAR in general."},{"finding":"The rough‑volatility RFSV model outperforms HAR and is comparable to LSTM variants.","kind":"performance","quote":"RFSV outperforms HAR, is slightly better than LSTM7,var and is close to LSTM30,var, but underperforms LSTM7,ret and LSTM30,ret."},{"finding":"Adding past returns via the QRH component improves RFSV, matching LSTM30,ret at λ≈0.15.","kind":"performance","quote":"Incorporating past returns in the RFSV through the QRH component is a clear improvement. At λ = 0.15, the model performs best and is very similar to LSTM30,ret."},{"finding":"Both non‑parametric LSTM and parsimonious RFSV outperform all traditional industry models.","kind":"performance","quote":"The non‑parametric LSTM and parsimonious RFSV, like for US stocks, outperform all other traditional models used in industry and past literature."},{"finding":"The volatility formation process appears universal across a wide selection of cryptocurrencies, similar to stocks.","kind":"theoretical","quote":"We see that LSTMs and RFSV, which are calibrated with the pooled dataset, outperform linear models calibrated specifically to each coin. This not only supports the use of the two non‑linear model types, but also the suspected universality in volatility which is captured by these models."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2403.12653","title":"To be or not to be: Roughness or long memory in volatility?","paper_date":"2024-03-19","authors":["Mikkel Bennedsen","Kim Christensen","Peter Christensen"],"empirical_rigor":7.5,"page_url":"https://thequant.space/flowcharts/to-be-or-not-to-be-roughness-or-long-memory-in-volatility/"},"question":"Do cryptocurrencies exhibit a universal, rough volatility formation process?","this_paper_finding":"The volatility formation process appears universal across a wide selection of cryptocurrencies, similar to stocks.","other_paper_finding":"The log‑variance of large‑cap cryptocurrency spot rates exhibits roughness, aligning with prior findings on volatility formation.","other_paper_quote":"In our empirical implementation of the fOU process, we discover that the log-variance of the large-cap segment of the cryptocurrency spot exchange rate market exhibits roughness.","note":"Both findings report rough volatility in crypto; B confirms A’s universal formation claim for large‑cap spot rates, showing consistent roughness across samples.","confidence":0.763}],"contradicts":[],"qualifies":[],"compared_with":21,"citations":{"cited_by_in_corpus":1,"by_independent_authors":0,"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":"2026-11-14T18:20:04Z"},"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/"}