We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Finance > Risk Management

Title: Estimating value at risk: LSTM vs. GARCH

Abstract: Estimating value-at-risk on time series data with possibly heteroscedastic dynamics is a highly challenging task. Typically, we face a small data problem in combination with a high degree of non-linearity, causing difficulties for both classical and machine-learning estimation algorithms. In this paper, we propose a novel value-at-risk estimator using a long short-term memory (LSTM) neural network and compare its performance to benchmark GARCH estimators.
Our results indicate that even for a relatively short time series, the LSTM could be used to refine or monitor risk estimation processes and correctly identify the underlying risk dynamics in a non-parametric fashion. We evaluate the estimator on both simulated and market data with a focus on heteroscedasticity, finding that LSTM exhibits a similar performance to GARCH estimators on simulated data, whereas on real market data it is more sensitive towards increasing or decreasing volatility and outperforms all existing estimators of value-at-risk in terms of exception rate and mean quantile score.
Subjects: Risk Management (q-fin.RM); Statistical Finance (q-fin.ST); Machine Learning (stat.ML)
Cite as: arXiv:2207.10539 [q-fin.RM]
  (or arXiv:2207.10539v1 [q-fin.RM] for this version)

Submission history

From: Thorsten Schmidt [view email]
[v1] Thu, 21 Jul 2022 15:26:07 GMT (5492kb,D)

Link back to: arXiv, form interface, contact.