Nonparametric model validations for hidden Markov models with applications in financial econometrics.

Area of Science:

• Statistics
• Econometrics
• Time Series Analysis

Background:

• Hidden Markov Models (HMMs) are widely used but validating their nonparametric forms, especially with partially observable data, remains challenging.
• Existing validation methods often struggle with the complexity of hidden states and observable variable transitions.

Purpose of the Study:

• To develop a robust nonparametric model validation technique for HMMs with partially observable variables and hidden states.
• To provide a method for assessing the goodness-of-fit for complex statistical models.

Main Methods:

• Constructing a nonparametric simultaneous confidence envelope for the transition density function of observable variables.
• Developing a specification test by checking if parametric density estimates fall within the derived confidence envelope.
• Leveraging the functional link between observable variable transitions and the hidden state Markov transition kernel.

Main Results:

• The proposed method provides a valid nonparametric confidence envelope for transition densities.
• The specification test effectively validates HMMs by assessing the inclusion of parametric estimates within the envelope.
• Demonstrated applicability to various advanced models including diffusion, stochastic volatility, and nonlinear time series.

Conclusions:

• The developed confidence envelope approach offers a reliable tool for nonparametric validation of HMMs.
• This methodology enhances the trustworthiness of HMMs in applications involving partially observed data and complex dynamics.
• The approach is versatile and applicable across diverse fields utilizing advanced statistical modeling.