Ols vs arima?

[Davi Lucena]

If we know that ARIMA effectively models time series with temporal dependence, why would anyone insist on using OLS, which completely ignores autocorrelation, to analyze time-dependent data? Wouldn't that be disregarding the very essence of the data by using such a simplistic model?

[Andrew Musau]

Serial correlation does not cause bias in the estimated coefficients (i.e., for the OLS estimator in the presence of serial correlation, \(\mathbb{E}(\hat{\beta}) = \beta\). Therefore, an approach that adjusts the standard errors for serial correlation without affecting the coefficients would be acceptable and would allow for valid inference (such as estimating linear regression using OLS with Newey and West standard errors).