Nystrup Lindström Madsen 2020
The originating methodology paper for the Statistical Jump Model, published in Expert Systems with Applications by Peter Nystrup, Erik Lindström and Henrik Madsen. It proposes learning HMM-style persistent states by clustering temporal features while penalising jumps between states, giving direct control over the transition rate and addressing the unrealistically rapid switching of misspecified or misestimated HMMs. It appears in this vault as the source that defines the jump model; its advantages over Baum-Welch/EM estimation — faster joint estimation, robustness to misspecification, less initialisation sensitivity — are testable and have been picked up by later work. Its trading content is a single illustration showing better persistence estimates reduce transaction costs, not a graded out-of-sample profitability test, so its profitability grade is inconclusive.
Connections
- Statistical Jump Model — proposes_model, 2020, source: https://doi.org/10.1016/j.eswa.2020.113307
- Hidden Markov Model Regime Detection — compares_benchmark, source: https://doi.org/10.1016/j.eswa.2020.113307
- Peter Nystrup — proposes_model, source: https://doi.org/10.1016/j.eswa.2020.113307
- Transaction Costs and Slippage — includes_costs, source: https://doi.org/10.1016/j.eswa.2020.113307