Tha major reason is because it has sound mathematical assumptions and highly interpretable. It ideas strong stochastic and probabilistic assumptions that majority of other methods do not have those properties, stationarity orders, regressions on input and error.

Given this highly interpretable mod, it produces accurate but not necessarily the best results.

It is highly practical and unlike popular methods in Kaggle, it can be utilized in an adaptive manner, e.g updating coefficients when novel observations come in.

Also, ARIMA is used as a benchmark as it gives you a deep insight and understanding about the stochastic process under the study.