Faster auto.arima estimation

[studsttat]

No description provided.

[robjhyndman]

What makes you think this is possible? Any ideas how?

[studsttat]

It made me think of it since there is a python package named statsforecast, where it has a function named auto_arima(). The difference between auto.arima() and auto_arima() in statsforecast is that the auto_arima() function in statsforecast is 1.5x faster than auto.arima() (according to the GitHub page of statsforecast).

And so, I thought it can be done with the use of parallel computing.

[robjhyndman]

statsforecast achieves its speed using JIT computing through Numba, a Python compiler. There is no such facility available for R. This has nothing to do with parallel computing, which is already available in both the Python and R implementations.

[studsttat]

Nevermind, the auto.arima() function can achieve such speed. I tried fitting ARIMA in 50.000 obs. using auto.arima(), and the ARIMA was fitted in just an average of 3 seconds.