{"paper_id":"2412.03668","title":"Hidden Markov graphical models with state-dependent generalized hyperbolic distributions","authors":["Beatrice Foroni","Luca Merlo","Lea Petrella"],"paper_date":"2024-12-04","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2412.03668v1","page_url":"https://thequant.space/flowcharts/hidden-markov-graphical-models-with-state-dependent-generali/","scores":{"empirical_rigor":7,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a sparse hidden Markov graphical model with state‑dependent symmetric generalized hyperbolic emission distributions, estimated via a penalized EM/ECME algorithm with an L1 penalty on state‑specific precision matrices.","quote":"In this paper, we contribute to the existing literature by introducing a sparse hidden Markov graphical model to investigate time-varying conditional correlation structures in multivariate time series data, without assuming normally distributed returns. To build our network model, we consider multivariate symmetric generalized hyperbolic (GH, McNeil et al. 2015) distributions with time-dependent parameters evolving","summary":"In this paper we develop a novel hidden Markov graphical model to investigate time-varying interconnectedness between different financial markets. To identify conditional correlation structures under varying market conditions and accommodate stylized facts embedded in financial time series, we rely…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Emissions follow a symmetric multivariate generalized hyperbolic distribution with time‑dependent parameters.","type":"model","stated":"explicit","quote":"To build our network model, we consider multivariate symmetric generalized hyperbolic (GH, McNeil et al. 2015) distributions with time-dependent parameters evolving according to a discrete, homogeneous latent Markov chain.","support":"full"},{"assumption":"The latent Markov chain is discrete, homogeneous and has a fixed number of states K.","type":"model","stated":"explicit","quote":"To build our network model, we consider multivariate symmetric generalized hyperbolic (GH, McNeil et al. 2015) distributions with time-dependent parameters evolving according to a discrete, homogeneous latent Markov chain.","support":"full"},{"assumption":"Precision matrices are sparse and can be recovered using an L1 (Lasso) penalty.","type":"statistical","stated":"explicit","quote":"We exploit its location-scale mixture representation to build a penalized EM algorithm for estimating the state-specific sparse precision matrices by means of an $L_1$ penalty. Since we are interested in detecting only the most important connections, we also propose a penalized version of the EM by considering the Lasso-type regularization of Tibshirani (1996) for sparse matrix estimation in high-dimensional","support":"full"},{"assumption":"Daily returns are stationary (ADF test at 1% level).","type":"data","stated":"explicit","quote":"In concluding, the Augmented Dickey-Fuller (ADF) test (Dickey & Fuller 1979) shows that all daily returns are stationary at the 1% level of signiﬁcance.","support":"full"},{"assumption":"Returns are non‑Gaussian and exhibit heavy tails, justifying the use of GH family instead of normality.","type":"model","stated":"explicit","quote":"Nevertheless, “non-standard” features of returns cannot be ac-commodated by standard models such as those based on normality assumptions. In this paper, we contribute to the existing literature by introducing a sparse hidden Markov graphical model to investigate time-varying conditional correlation structures in multivariate time series data, without assuming normally distributed returns.","support":"full"},{"assumption":"The mixture representation of GH allows use of Gaussian graphical tools (glasso) within the M‑step.","type":"execution","stated":"explicit","quote":"To induce sparsity and recover only the most prominent relations, we develop a Penalized ECME algorithm exploiting the Gaussian location-scale mixture representation of the GH distribution with a Lasso L1 penalty on the elements of the state-speciﬁc precision matrices. Our procedure is computationally eﬃcient and can be implemented by making use of the glasso of Friedman et al. (2008) within the conditional M-step","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Daily returns of 29 financial assets (stock indices, commodity futures, cryptocurrencies)","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":"Nov 10 2017 to Sep 28 2023","universe":"10 largest cryptocurrencies, 10 most traded metal/energy commodities, 9 major world stock ","quote":"In this paper, we are interested in investigating how the degree of network connectivity across diﬀerent asset classes changes in severe market turbulence, which can possibly threaten the integrity of the ﬁnancial system and represent a potential source of ﬁnancial instability. For this purpose, we apply the proposed methodology to daily returns of d = 29 ﬁnancial assets comprising stock market indices, commodity","support":"partial","known_id":"yahoo-finance"},{"name":"Synthetic data for simulation study","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"4 Simulation study We conduct a simulation study to validate the performance of our model under diﬀerent scenarios in terms of: 1. recovering the true values of the parameters for a varying number of hidden states, 2. clustering performance, 3. the ability to correctly retrieve the edge sets associated to each hidden state. Scenario 1: In the ﬁrst scenario we consider a 2-dimensional vector Y t with sample size of T","support":"full"}],"known":[{"id":"yahoo-finance","name":"Yahoo Finance","provider":"Yahoo","kind":"market-prices","access":"public","quote":"The sample dataset is collected from Yahoo Finance and the study period starts on November 10, 2017 and ends on September 28, 2023, for a total of T = 1250 observations after removing missing data.","mentions":1,"used_in":1}],"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":"unclear","findings":[{"finding":"Simulation results show that the estimated location, scale, and GH parameters are very close to the true values across all six data‑generating models.","kind":"statistical","quote":"For every model, the estimated location and scale parameters are very close to the true values. This is also true for the parameters (λk, χk, ψk) in every state.","support":"full"},{"finding":"When dealing with non‑Gaussian heavy‑tailed data, the proposed dynamic graphical model matches the performance of the graphical lasso, offering a robust alternative for static and dynamic settings.","kind":"performance","quote":"Overall, these results suggest that when we deal with non-Gaussian heavy tailed data, our static and dynamic graphical models have the same performance of the graphical lasso, representing a robust alternative to existing static approaches and a novelty for dynamic graphical models in non-Gaussian settings.","support":"full"},{"finding":"Cryptocurrencies exhibit the highest volatility among the three asset classes, with the largest standard deviations and strong deviations from normality.","kind":"statistical","quote":"As one can see, each asset displays the typical stylized facts that characterize the ﬁnancial markets, and cryptocurrencies are generally much more volatile than commodity futures and stock market indices, having the highest standard deviations. Further, the Jarque-Bera test strongly rejects the normality assumption of daily returns for all series considered.","support":"full"},{"finding":"The model successfully captures regime‑specific conditional correlation structures, revealing varying degrees of network connectivity over time.","kind":"statistical","quote":"The proposed approach leads to regime-specific conditional correlation graphs that allow us to identify different degrees of network connectivity of returns over time. The proposed methodology allows us to capture regime-speciﬁc conditional correlation structures for probability distributions within the GH family.","support":"full"},{"finding":"Clustering performance depends on the underlying data‑generating distribution, with slightly worse results for Cauchy‑generated data and better performance for fewer hidden states.","kind":"performance","quote":"Firstly, we observe that the data generating model plays a fundamental role in estimating the true states partition, as we obtain slightly worst clustering performance for the model with the Cauchy distribution. Secondly, the goodness of the clustering obtained depends, as expected, on the number of states of the Markov chain considered, being the values slightly higher with K = 2 than with K = 3.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2507.23414","title":"Complexity of Financial Time Series: Multifractal and Multiscale Entropy Analyses","paper_date":"2025-07-31","authors":["Oday Masoudi","Farhad Shahbazi","Mohammad Sharifi"],"empirical_rigor":3,"page_url":"https://thequant.space/flowcharts/complexity-of-financial-time-series-multifractal-and-multis/"},"question":"Which assets show the highest volatility as indicated by standard deviation or probability density width?","this_paper_finding":"Cryptocurrencies exhibit the highest volatility among the three asset classes, with the largest standard deviations and strong deviations from normality.","other_paper_finding":"Bitcoin and natural gas show wider probability density functions, reflecting higher volatility than GBP/USD and gold.","other_paper_quote":"The wider distribution of Bitcoin and natural gas indicates higher volatility in these assets. The overall ﬁndings of this research can be summarized as follows: Both the Bitcoin and Gas log-return time series exhibit higher volatility compared to GBP/USD and Gold, as reﬂected in their probability density function plots.","note":"Both findings rank Bitcoin (cryptocurrencies) as having the largest volatility; B also notes natural gas as high, while A compares crypto to two other classes (three total).","confidence":0.844}],"contradicts":[],"qualifies":[],"compared_with":23,"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":2,"extracted_at":"2026-11-14T19:53:52Z"},"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/"}