At least 60 observations. Good results in 75 observations and more.

At least 30 observations.

You can read more on the link below

https://www.researchgate.net/post/How_many_observation_for_VAR_Model

Article Growth Trajectories and Detrended Intraindividual Variabilit...

Hox (2002) suggested that 100 cases and at least 10 measures per case are required for models interested in variance. Salthouse, Nesselroade, and Berish (2006) also reported that more than 50 measures were required to achieve an asymptotic level of competency.

The following articles dealing with time series in various field of studies may be helpful in comparing what researchers had used among various disciplines.

LINKS:

http://www.bishophill.com/admin/sidebar_images/1741759940_test.pdf

https://otexts.org/fpp2/long-short-ts.html

https://pdfs.semanticscholar.org/447b/57101a06401e00b97c4eeaeaa3331ae62ff1.pdf

https://peerj.com/preprints/3168.pdf

50 observations and above, preferably in quarterly form. 30 observations is very small for VECM, especially when there are more than 4 variables, due to the issue of over-parameterization and the dangers of micro-numerousity, as well as the problem of losing too many degrees of freedom.

It may seem a simple question but it does not have a simple answer. The longer your VECM lag structure, the more observations you need. The weaker the ECM effect, the longer the time span that may be required in order for it to be evident. Also, for inference, you need enough observations to justify using the asymptotically valid critical values. Readers are likely to challenge your conclusions if you base them on much less than 30 observations.

At least 60 observations. Good results in 75 observations and more.

Muntasir Murshed If you have 3-4 variables in you model, 40 or more than that is good. In lag selection criterion, if you have annual data, restrict maximum order of lag to 2-3. Addition of any variable to the model requires inclusion of at least 10 years data and more for better results.

Try to find quarterly data.

A long series would, definitely, help you. However, if your data no more than 40 observations, you may use FMOLS instead of VECM.

I would be grateful for more explanation from Huthaifa. FMOLS is an estimator, VECM is a model structure so I do not understand how they can be compared.

I agreed with Prof. Vince Daly. The FMOLS estimator and VEC model cannot be compared. Regarding the minimum data observations for time series analysis, you should need at least 30 observations for annual data, preferably more than 50 years if you need to analyze for forecasting purposes.

I also agreed with Chekwube Madichie. 30 observations are very small for analyzing the VEC model. Preferably you should use quarterly or monthly data.

The larger the sample, the better. It must not be less than 30 in theory.

it should be minimum 50, unless you want to monitor fewer obs through central limit thorem

Noted. Thanks.