I have a time series model that has seasonality (and therefore it has autocorrelated errors fitting an OLS model) - how can I account for autocorrelation without a complicated autoregressive model like an ARIMA? And any book recommendations?

This passage from Richard McElreath's Statistical Rethinking v2 struck a chord with me

"There are several famous problems with autoregressive models, despite how often they are used. They are surely generalized linear madness. First, nothing that happened two time periods ago causes the present, except through its influence on the state of the system one time period ago. So no lag beyond one period makes any causal sense. It’s pure predictive epicycle. Of course some causal influences act slower than others. But that means you need another variable, not that the distance past can influence the present."

  • $\begingroup$ ARIMA are a class of relatively simple models. What is ARIMA not doing that you want some other time series type model to do? Can you provide plots of your data or any context at all? $\endgroup$ – jcken Jul 3 at 13:57
  • 1
    $\begingroup$ Immediately following the quoted passage, the author suggests using state space models instead. He seems to be ignorant of the fact that ARIMA has a state-space representation. This makes me skeptical of his rant. By the way, how is the bayesian tag relevant to the question? The fact that the book is on Bayesian statistics does not justify it, IMHO. $\endgroup$ – Richard Hardy Jul 8 at 19:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.