I and my team @uber just open-sourced a Bayesian forecast package – Orbit. Right now, I just include two Exponential Smoothing models:

  1. Damped Local Trend (DLT) - a rework on the classical ETS(A,A,A) model with damped trend
  2. Local Global Trend (LGT) suggested by Slawek Syml.

It is benchmarked against other popular models with decent performance. It also provides Bayesian regression for different modeling purpose. More information can be found:

  1. White paper
  2. Github page
  3. Comments from Sean Taylor, the author of forecasting package prophet and a data scientist I highly respect.

Last words, thank you all the eng work Steve has done and the suggestions Slawek and Fritz have made.