Nystrup Kolm Lindström 2021
Published in Expert Systems with Applications by Peter Nystrup, Petter N. Kolm and Erik Lindström, this paper introduces the sparse statistical jump model — a framework for joint feature selection, parameter and state-sequence estimation in jump models. It uses a coordinate-descent algorithm that alternates between selecting the features that distinguish states and estimating the model, scaling to high-dimensional, noisy feature sets where large-scale feature selection has historically been infeasible for standard HMMs. Benchmarked against K-means, sparse K-means, the HMM and other methods on simulated data, financial returns, protein sequences and text, the sparse JM outperforms all and is remarkably robust to noise. It appears in this vault as the paper defining the sparse JM variant; it is a methodology contribution, not a costed trading backtest, hence an inconclusive profitability grade.
Connections
- Statistical Jump Model — proposes_model, 2021, source: https://doi.org/10.1016/j.eswa.2021.115558
- Nystrup Lindström Madsen 2020 — supports, source: https://doi.org/10.1016/j.eswa.2021.115558
- Peter Nystrup — proposes_model, source: https://doi.org/10.1016/j.eswa.2021.115558
- Petter Kolm — proposes_model, source: https://doi.org/10.1016/j.eswa.2021.115558