Hi ! I would like to run a linear regression with time series variables. My goal is to explain the determinants of industrialization in Switzerland between 1890 and 1913. My dependent variable is a proxy for industrialization and I have 5 explanatory variables (like education, tarifs etc.). All variables are time-series. I also add a time trend taking the value 1 in 1890, 2 in 1891 and so on.
I would like to know if I have to stationnarize my variables before running OLS estimation?
Thanks !
You have to do stationarity test before running OLS. As stated by Granger and Newbold (1974) thatt regression models for non-stationary variables give spurious results
The requirement when using ols regression is that the residuals , the response conditional on the predictors, shows no remaining temporal pattern. So I would start with ols and regress the response on the predictors but not include time. Then plot the residuals against time to see if there is a pattern, the plot may even suggest an important missing variable. more formally you could do a runs test or a durbin-Watson test, but your series is not very long.
you may find it difficult to disentangle due to the lack of variation in the response variable and collinearity in the predictors, and you do not have much statistical power with 23 observations and 5 predictors. I am presuming that you are treating the country as a whole and do not have regional data.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series