Regime-switching empirical similarity model: a comparison with baseline models

Area of Science:

• Econometrics
• Time Series Analysis
• Machine Learning

Background:

• The standard empirical similarity (ES) model by Gilboa et al. (2006) has limitations in capturing parameter changes.
• Existing models like autoregressions and Markov-switching autoregressions serve as baselines for comparison.

Purpose of the Study:

• To extend the standard empirical similarity (ES) model to accommodate parameter variations.
• To evaluate the predictive performance of the modified ES model against standard and baseline models.

Main Methods:

• The modified ES model is implemented by combining component ES models, inspired by Gaussian mixture models.
• Predictive power is assessed through a simulation exercise and an empirical application using US real GDP growth data.

Main Results:

• The modified ES model demonstrates superior empirical fit in scenarios with complex regime-switching behavior and high autocorrelation.
• The enhanced model shows improved predictive accuracy for extreme economic observations.

Conclusions:

• The modified empirical similarity (ES) model offers a more robust approach for time series analysis, particularly in dynamic economic environments.
• This extension provides a valuable tool for forecasting and understanding economic fluctuations with regime shifts.