Robust autoregressive modeling and its diagnostic analytics with a COVID-19 related application

Area of Science:

• Statistics
• Econometrics
• Time Series Analysis

Background:

• Autoregressive models are fundamental tools in time series analysis across various scientific domains.
• Existing models may not adequately capture asymmetry or heavy tails often present in financial data.
• Robust methods are needed to identify influential observations that can distort model estimation and forecasting.

Purpose of the Study:

• To propose a novel skew-t autoregressive model for enhanced time series modeling.
• To develop and validate a methodology for identifying influential observations using local perturbation analysis.
• To apply the proposed model and methodology to analyze the impact of the COVID-19 pandemic on Brent crude oil futures.

Main Methods:

• Parameter estimation for the skew-t autoregressive model using the Expectation-Maximization (EM) algorithm.
• Development of an influence methodology based on local perturbations and calculation of normal curvatures.
• Monte Carlo simulations to assess the performance of the proposed influence diagnostics.
• Application to daily log-returns of Brent crude futures data.

Main Results:

• The skew-t autoregressive model provides a flexible framework for modeling asymmetric and heavy-tailed time series data.
• The local perturbation methodology effectively identifies influential observations in the time series.
• Analysis of Brent crude oil futures reveals potential impacts of the COVID-19 pandemic on daily log-returns, identified through influential points.

Conclusions:

• The proposed skew-t autoregressive model and influence methodology offer valuable tools for time series analysis, particularly in finance.
• The approach enhances the robustness of time series modeling by identifying and understanding the impact of outliers or influential data points.
• The study demonstrates the practical utility of the model in analyzing real-world financial data and significant events like the COVID-19 pandemic.