1
$\begingroup$

I want to run ols regression for time series data in R, but my data is short that is annual from 2000-2009. There are only 9 variables(2000-2009) and i collected data for inflation and exchange rate that how these variables between 2000-2009 affect gdp growth between 2000-2018. Is it ok for regression?

$\endgroup$
1
  • $\begingroup$ if you have time series ( as you do ) be very wary of ols regression as @Fr1 reflected although it might be possible to tweak out a useful model . This piece that I authored 20 plus years ago might help you . autobox.com/pdfs/regvsbox-old.pdf $\endgroup$ – IrishStat Aug 12 '19 at 15:00
0
$\begingroup$

Well, in my regression I had also only 10 years and it was OK. So yes you can do it. It's not a lot of data though, so keep the regression as simple as possible (the two covariates are the max I would say).

$\endgroup$
3
  • $\begingroup$ Simple regression is mostly simply wrong when time series is in play . See the URL in my comment to the OP $\endgroup$ – IrishStat Aug 12 '19 at 15:03
  • $\begingroup$ @IrishStat I'm talking simplicity covariate-wise, I'm not advising to use normal linear regression! I assumed, after reading the question, that OP knows which statistic to use. I am just advising to use very few covariates. $\endgroup$ – Innate Imunity is The Way Aug 12 '19 at 19:37
  • 1
    $\begingroup$ totally agree with you ! sorry if it came across as something else .... $\endgroup$ – IrishStat Aug 12 '19 at 20:42
0
$\begingroup$

Theoretically you can still define the coefficients as the ones that minimize the sum of squared errors even with 9 data points. The problem is however that once you have estimated you will do practically nothing reliably like hypothesis testing, etc because the sample size is too short. Not to mention that if you work with time series data then your Error term is likely to be divergent from the OLS assumptions, so you should not use OLS because likely it will not be BLUE.. to give you a very very simple (albeit only partially correct) improvement GLS would likely be better. But as I said, it is only partially correct and for time series you should fit other kind of time series models estimated via MLE and taking into account the Heteroskedastic of the Error and the possible non-stationarity of data.. but I do not want to overcomplicate here because it seems that you are not working on time series

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.