Differential entropy; nonparametric differential entropy estimator ; kernel density estimator; heavy-tailed kernel; cluster analysis; stochastic regime
In financial risk, the conventional approach has typically linked risk to the variance of a variable, such as the return of a stock or portfolio. By recognizing the constraints of this conventional method and the need for various risk metrics, alternative measures have been developed to address downside risk or extreme outcomes specifically. One such complementary metric is the uncertainty measure, which enables us to capture and describe different aspects of risk, going beyond traditional notions of variability alone. Obtaining a robust estimator with desirable properties for entropy is crucial for its practical application. In particular, our study aims to conduct a comprehensive review of non-parametric differential entropy estimators and then propose adjustments regarding the choice and optimization of their use to find an estimator with convenient properties for application in financial data, which are often characterized by distributions with heavy tails. We also conducted real-data applications to illustrate the use of the proposed measures.