Thank you for your help in advance !

I'm working on a research project at undergraduate level on the effect of human capital on growth in Nigeria from 1980 to 2018.

My current model specification is : ln(RGDP)=ln(K)+ln(L)+ln(secondary)+ln(primary)+ln(TFP)+epsilon

where K : total capital formation

  • L : working age labor force
  • secondary : enrollment rate in secondary education
  • primary : enrollment rate in primary education
  • TFP: total factor productivity -> wanted to use this as a proxy for technology

Also - augmented dicker fuller test :

  • log(K) is not stationnary (p value=0,54)
  • log(L) is not stationnary (p value=0,67)
  • log(secondary) is not stationnary (p value=0,25)
  • log(primary) is not stationnary (p value=0,15)
  • log(TFP) is not stationnary (p value=0,55)

The results of the regression are okay

  • 3 stars : log(K), log(TFP)
  • 2 stars : log(primary), log(secondary)
  • 1 star : intercept
  • Other variables : log(labor), p value of 0,46

BUT the VIF is quite high for log(K) : 13 and log(L) : 20

I also tested the residuals, which "are" (we can't reject) normally distributed
And there seems to be no autocorrelation between the residual (D-W statistics=1,86 ; p value=0,25)

Thus my questions are :

  • Is it a good specification ? How can I improve it, if not ?
  • How to reduce the VIF ?
  • Should I differentiate the variables (and which variables) to have stationnary time series ?
  • Should I lag the variables ?
  • Why is the coefficient of "primary" negative, when we can expect a positive one ?

Thank you very much for your future answers, I wish you a great day !



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.