Long memory and changepoint models: a spectral classification procedure.

Area of Science:

• Econometrics
• Time Series Analysis
• Signal Processing

Background:

• Time series in finance and economics are often modeled using long memory processes.
• Alternative studies suggest the presence of changepoints, where the data generating process shifts.
• Ambiguity arises between these models due to spectral similarities, complicating model selection.

Purpose of the Study:

• To resolve the ambiguity between long memory and changepoint models in time series analysis.
• To develop a method for accurately classifying time series data into either long memory or changepoint processes.
• To evaluate the performance of the proposed classification method against existing approaches.

Main Methods:

• Utilizing a time-varying spectrum to analyze time series data.
• Employing the wavelet spectrum for detailed time-frequency analysis.
• Developing a classification approach based on the wavelet spectrum to distinguish between models.
• Conducting simulations across various models and applying the method to real-world financial data (stock correlations, US inflation).

Main Results:

• The time-varying spectrum effectively removes ambiguity between long memory and changepoint models.
• The proposed wavelet spectrum classification approach accurately identifies the underlying data generating process.
• Simulation results demonstrate the superiority of the classification method over existing hypothesis testing approaches in several scenarios.

Conclusions:

• The wavelet spectrum-based classification provides a robust solution for distinguishing between long memory and changepoint time series models.
• This method enhances the accuracy of time series modeling in finance and economics.
• The approach offers a valuable tool for researchers and practitioners dealing with complex time series data.