Robust statistical jump models

Statistical Jump Models are a class of regime identification models,similar to Hidden Markov Models, with the added advantage of facilitatinglarge-scale integration of exogenous variables that influence thelatent regime-switching process. Recent advancements in SJMs includethe introduction of automatic feature selection techniques, which eliminateirrelevant exogenous information and enhance the model’s ability toidentify regimes more effectively.We generalize the regular Statistical Jump Model by considering robustdistances measures, while preserving the feature selection. This isimportant for financial data, as these are often heavy-tailed.The simulation study confirm that the proposed modified algorithmis outperforming the regular statistical jump model on heavy-tailed databut not on light tailed data, as predicted by statistical theory. We alsofind that the corresponding latent regime process is more persistent whenestimated on index returns than comparable estimates derived using HiddenMarkov Models or the regular statistical jump model, an obviousadvantage for any trading strategy based on the latent regime process.