Accurately forecasting time series data is a critical task in many fields, from finance and economics to engineering and the natural sciences. However, many of these time series are subject to regime changes. Regime changes are changes in the dynamics exhibited by the time series, often driven by external factors. These changes lead to shifts in the series’ statistical properties, which can be difficult to detect and can impact forecasting models’ predictive power.

This talk will discuss different strategies for identifying and adapting to regime changes in time series forecasting, from machine learning algorithms, to identify shifts in regimes, to the incorporation of external domain knowledge indicators as additional input features. This talk will also examine the effectiveness of ensemble techniques for combining multiple models, each trained on a different regime. Case studies from the financial data domain will demonstrate the importance of considering regime changes in time series forecasting and the potential for these techniques to improve accuracy.