1

I have few seemingly simple questions.

  1. I am working on time series data and applying vector error correction model. I find different results when I transform the data into LOG and LN. Which one is better?

  2. I have variables in % and in levels (numbers). I have for example one which is measured in million USD and others in %. My question: shall I list the whole number in eviews like 47,000,000,000 or just 47 and then bring in the billion when interpreting the results? Does it make any difference when running Johansen's test?

kjetil b halvorsen
  • 77,844
user48050
  • 21
  • 3
    That is at least two questions. For (1): $\ln(x) \approx 2.302585\times \log_{10}(x)$ so the only impact should be that of a scaling factor – Henry Jun 09 '14 at 22:52

1 Answers1

2

1) We have that $\log_a y = \frac{\ln y}{\ln a}$ so the only effect of the difference is that of a scale factor.

2) It is in most cases best to enter the data as 47, as you say (and not as 47 000 000 000), and take the billions in the units! That will not make any difference on Johansen's (or most others) tests.

kjetil b halvorsen
  • 77,844
  • Thank you so much!!! I really appreciate that. I have one more question if you don't mind. Which values should I use for Johansen's test and estimating the VECM? level or the differenced values? Thanks again! – user48050 Jun 10 '14 at 08:59
  • You would use level values since you are testing for cointegrating relationships. With cointegration analysis you are modeling two or more (usually) I(1) or I(2) variables (or fractionally integrated variables) which are non-stationary variables but which have a stationary linear relationship between them. By differencing you make your variables stationary in case they are I(1) in the first place. So test for cointegration on your level variables! – Plissken Jul 13 '14 at 11:03