Exploring the sources of uncertainty: Why does bagging for time series forecasting work?

•Bootstrapping with aggregation (bagging) performs well for time series forecasting.•Bagging handles three sources of uncertainty: data, model and parameter uncertainty.•We decompose the performance benefits of bagging for each source of uncertainty.•Tackling model uncertainty brings the most benefits, even outperforming bagging.•We propose a strategy for weighted combinations of the suitable models.