OLS time series regression, importance of autocorrelation?


I'm just a uni student suffering from covid-struct regression lessons (getting help for questions is not easy). I hope I could get some answers from here..

I'm doing OLS regression with to goal to see effects of certain cryptocurrency on-chain factors on the price differences of a cryptocurrency.

E.g. model is the sort of

Price of crypto = B0 + B on-chain factor + e

All the variables are daily data in logarithmic first difference for stationarity and do not contain unit roots.

So what I'm wondering here, is should I address autocorrelation in some sense in such a regression? I don't seem to find answer to this and do not have anyone who could help me. Any insight or help on this one would be greatly appreciated.

Kind regards


Fortran must die
How do you know they don't have a unit root, that is what test told you this. You differenced so obviously they did have a unit root once upon a time. Did you analyze the differenced data to be sure the data is now stationary?

There are many answers to the issue of autocorrelation and probably lots of disagreement. Autoregression distorts the standard errors and thus the t test. One solution is to do a time series method like ARIMA or regression with ARIMA error which specifically models autocorrelation (well the first does, I have not worked with the later). Another is to use HAC standard errors, although again there is disagreement on that topic.

I am a data analyst rather than a statistician so take what I say with a grain of salt. Unfortunately even most statisticians (including the ones here) rarely work with time series so you might be stuck with my imperfect suggestions. :p