I have annual time-series data from 1972-2014. Is the number of observation sufficient to correctly estimate the Johansen cointegration test? One professor told me that minimum number of observation should be at least 200! Is it true? Please let me know. If it is possible, please refer a book where the requirement of the minimum number of observation for the cointegration test is given.
Hi sir!
I have an objection on the figure stated as it is too large observation. We can see the published works for evidences. Most of the published works were based on observations between 40 and 100. Even, Johansen (1995) and Juselius (2006) used a sample size of less than 100 while dealing with co-integration issues.
I think it is better first to try Engle-Granger Two Step procedure, you then use Johansen Approach. If your observations are more than 40, it is then OK to use Johansen. The only difference between the two techniques is that if you have more than two series better to use Johansen rather than Engle-Granger procedure.
How many time series do you have? I think the 'minimum' length required is dependent on the number of time series (cf. random matrix theory results).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics