i have a vector of variable comprises of X1:X5 and yearly basis totally 11. when i want to develop my VAR model in Eviews i receive No sufficient observation error constantly. data does not have missing value. is there any criteria for minimum observation. in other word why I receive this notice?
hi, at least 30 observation should be. This can be 30 year or quarterly data.
Out of interest, what VaR model is it for? FX, IR, EQ, COM, etc? Is it for academic purposes or for business? If its for Basel II/III compliant Market VaR models, you need at least 60 trading days of data for to be able to do a Historical Simulation.
HI,
In order to get reliable results, you need to have more observation. Although you can run a VAR model, say 30 obs,it does not guarantee that your results will be reliable. Try to obtain data as much as possible. If it is secondary data. normally it is the case in VAR modeling, getting a big sample is not a big issue.
Kind regards
Thushara
Zehra Dogan Caliskan is right if you fit a full VAR(1) model with a constant vector because you will have 5 parameters for the constant vector and 5² = 25 parameters for the VAR(1) coefficient. What you can do is to fit a model with only a diagonal matrix (so 5 + 5 = 10 parameters) but there will be only one degree of freedom so the results will not be reliable as S. S. Thushara says. Victor Gumbo speaks of VaR, not VAR, something not related at all.
Each equation in a VAR system is estimated by OLS. If you are estimating a VAR with a single lag with 5 variables and a constant each equation will have 6 coefficients, with 2 lags 11 coefficients. In either case 11 observations is not sufficient. With the 5 equations there are 30 coefficients with one lag and 55 with two lags. With a VAR with 5 variables I would thin that 30 observations may be a little small. The power of any tests is likely to be very small but it might be worth trying.
generally in time series models the minimum of data should be about 90 data. but in Iran we suffer from poor data in some cases. Therefore you should at first determine the optimum lag for your model and if for example the optimum lag is one there for you can apply your VAR model without any problem but if optimum lag is two or three there for because of order of lag and lack of time series data Eviews send syntax error
is monthly data usable for VAR model. I guess if you have quarterly data then convert them into monthly so that you can reach the needing observation. But i am not sure if monthly data usable.
One of the most important things for using monthly or quarterly models is checking seasonal unit root for HEGY approach. And I think convrting quarterly data into monthly data lead to losing some information in the original time series.It seems that if we haven' t adequate data or we face with poor data problem we must use bayesian methods such as bayesian VAR.
One of the most important things for using monthly or quarterly models is checking seasonal unit root for HEGY approach. And I think convrting quarterly data into monthly data lead to losing some information in the original time series.It seems that if we haven' t adequate data or we face with poor data problem we must use bayesian methods such as bayesian VAR.
Econometrics is the application of statistical (and mathematical) analysis to economic questions. Any econometric analysis should start with some kind of economic question which may vary from forecasting to a full structural model. If you give some idea of your economic question you might get better advice.
A basic understanding of VAR methodology would indicate that 11 annual observations is not sufficient for any VAR analysis involving 5 variables. If you do not under stand this point and had corresponding quarterly or monthly data I wold be very concerned concerned about how you would interpret your results. Why do you think think that VAR is an appropriate methodology for your problem? Perhaps you should discuss this problem with a local econometrician.
The idea of 30 observations possibly comes from notions of convergent to a normal distribution. It is often close to the number of annual observations available in many macro analyzes. With that number of observations one can try the analysis but you may find that important coefficients are not statistically significant. Larger samples are likely to give more reliable results. Using quarterly or monthly data may not give better answers if an adjustment to equilibrium is slow.
If your data is so limited, in your analysis, you should concentrate on your economic explanation and some graphical analysis. The kind of measurement derived from econometric analysis is not possible and theory is more important
Hamed Jabarian For any multivariate model like VAR, we need a minimum of 60 observations
All those saying we need 'n' minimum number of observations... provide reference.:)
Hi, can i apply var model for time series data of 20 years, and would the results be reliable?
@Sarah Chawsheen Please see my earlier answer. If you have 240 monthly or 80 quarterly observations of stationary data series you should be able to use the VAR methodology. 20 annual observations may not be enough. The reliability of your results depends on the validity of your underlying assumptions and the volatility of your data.
- Badges
- Science topic
- Similar topics
- Academic Journals