Statistical inference on specifying regression models for detecting dependence in autocorrelated series

Area of Science:

• Statistics
• Time Series Analysis
• Econometrics

Background:

• Traditional regression models for dependence detection often fail to account for autocorrelation in time series data.
• Existing research on specifying models with autocorrelated noise (ACN) or lagged dependent variables (LDV) is limited.
• Autocorrelation can lead to inaccurate estimations of dependence in time series.

Purpose of the Study:

• To propose a robust statistical analysis framework for model specification and estimation in the presence of autocorrelation.
• To address the limitations of traditional regression models in detecting dependence within autocorrelated time series.
• To provide a reliable method for analyzing the dependence between variables in time series data.

Main Methods:

• Development of a statistical framework for model specification and estimation.
• Validation of the proposed framework through simulation studies.
• Application to real-world case studies: precipitation dependence on temperature and temperature dependence on CO2.

Main Results:

• The proposed framework effectively handles autocorrelation, providing more accurate dependence estimates.
• Simulations validated the framework's effectiveness in model specification and estimation.
• Traditional and misspecified models were shown to overestimate dependence by up to three times in real-life cases.

Conclusions:

• The study provides a solid statistical basis for detecting dependence in autocorrelated time series.
• The proposed framework offers a significant improvement over traditional methods, reducing overestimation bias.
• Accurate modeling of autocorrelation is crucial for reliable dependence detection in time series analysis.