Our data is yearly time series on macroeconomic variables. Observations = 20 years. Dependent variable is GDP growth. Independent variables are governement terms (dummies) Consumptions, net trade, FDI, F. reserve, population growth, and agri. prod. The main objective is to examine the impact of government terms on economic development.
I think this paper can be of interest to you:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.133.6097&rep=rep1&type=pdf
Mr. Samantar@ Thanks for your cooperation, but my data is time series...
20 years of data and you have so many independent variables may run into problems when using GMM. GMM is an instrumental approach to estimation and most effective to address heterogeneity issues in cross sectional dynamic panel models. For your problem there are other estimators you can use which you can try and compared. Time series data normally have serial correlaton problems compared to heteroskedasticity in cross sectional data.
GMM approaches (both first-differenced and system versions) to panel data model is in essence for dynamic panel data models and hence appropriate for wide and short data sets, that is N (number of observations) is should be much, much larger than T (number of time periods). In your case, T is 20 but we do not know N. Assuming your panel data model is dynamic, that is the lag of dependent variable as an explanatory variable is included in your regression, I strongly suggest the paper written by Judson and Owen. (1999).
The second issue I would like to mention that the GMM estimators suffers from instrument proliferation and this can cause test for over-identifying restrictions (such as the Sargan test or the Hansen test) to be weak (Roodman (2009). This problem is particularly more important for the system GMM because it uses more instruments than the first-differenced GMM. In this respect you may refer to the paper by Roodman (2009).
Best regards.
References:
Judson, R., and Owen, A. (1999). “Estimating dynamic panel data models: A guide for macroeconomists”. Economics Letters 65, 9–15.
Roodman, D. (2009) “A note on the theme of too many instruments”, Oxford Bulletin of Economics and Statistics, 71, 135–58.
Thanks dr. Bulent for your kind suggestions. But, my data type is time series. If you have any suggestion regarding this, please do....
Sorry for my misunderstanding, I supposed that you are studying on panel data.
Yes. You have too few observations. I suggest you extend your problem by using a panel data set of a group of countries. The World Bank has an excellent data set available. Good luck!
Dr. Raja@ Thanks sir for your kind suggestion. But, my objective will not be surved if i take panel data....
Unfortunately, you cannot use time series data for only one country if you have such a short time series. There are several research papers that are concerned with the impact of government expenditure on the economy that use a panel data set of several countries (either developed, developing, or all). You will have to extend your analysis.
A dataset of 20 observations is simply too small to run a regression with any validity, much less one with multiple independent variables.
This can be seen in many sources. The absolute minimum standard I commonly hold to is no less than 30 observations for any regression and that's a great stretch. I would prefer significantly over 100.
One example of minimum observations that I got from a different thread here) is:
Advice from Tabachnick & Fidell, Using Multivariate Statistics (k is the number of predictors)
a minimum N > 50 + (8*k) for tests of multiple R
a minimum N > 104 + k for tests of individual predictors
In general, for publishable results, N should be over 100.
Unfortunately, you will need to increase your sample size. Could you maybe move to quarterly data rather than annual? I think most of the data you mentioned should be available in quarterly and that would give you 80 observations. Still low, but better than what you have right now.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series