Time Series Embedding
Yigit Aydede · 2023
Forecasting models use either direct or recursive forecasting, or their combinations. The difference between these two methods is related to discussion on prediction accuracy and forecasting variance. This chapter delves into the technique of time series embedding for direct forecasting, focusing on how to reorganize data effectively to estimate multiple models simultaneously. Recursive forecasting requires a parametric model and would face increasing forecasting error when the underlying model is not linear. Direct forecasting, can be achieved by a non-parametric predictive algorithm, while it may have a higher variance as the forecast horizon gets longer. Multi-period recursive forecasting uses a single time series model, like autoregressive.