I am going to work on disaggregated exports for Pakistan but the data of some export categories are available for short period. i have collected the required data but the total observations are hardly in the range of 38 to 42 observations. In this case what i do or it is enough observations for time series analysis as i am going to estimated the ARDL, or var and ECM models for final estimation
https://www.researchgate.net/topics.QuestionsAsked.html
Dear Husaain
You need 30 observation at least for time series analysis. Normal Distribution !!!
You did not mention how many variables and parameters you have? Your observations are monthly, yearly or quarterly? The most important point, there must be enough variations within sample period? As an example for the number of observations and parameters in small samples you can refer to these articles?
ENVIRONMENTAL KUZNETS CURVE AND PAKISTAN: AN EMPIRICAL ANALYSIS
THE MONETARY TRANSMISSION MECHANISM IN THE EURO AREA: MORE EVIDENCE FROM VAR ANALYSIS
THE ROLE OF MONETARY POLICY IN MACROECONOMIC VOLATILITY OF ASEAN-4 COUNTRIES AGAINST OIL PRICE SHOCK OVER TIME
Dear Fatemeh Razmi
thanks alot
The variables are 5 including dependent variables. The data are quarterly then the total observation are hardly 38 available. in the case of yearly the number of observations are 42 available. what i am do in this case.
Dear hussain
If you want to creat a multipel regression equation y=f(x1,x2,x3,...,xn) with a series it must to test the stationsry firstly, then you estmat the regression equation. But about the numbre of the observation it mast be more than 30 and it must use the same unit for exampl use year or month for all variables
Most important (but often not recognized) is a careful specitication of the equation you want to estimate. I assume that your quarterly data are for the last 8-10 years and the annual ones reach back around 40 years, but they relate to the same variables. In this case, I think it is best to do two separate regressions, one with the annual and one with the quarterly data (for this regression it seems necessary to include quarterly dummies to cover seasonal fluctuations). You can then look whether the estimated coefficients fit together. Before running the regressions you should, of course, analyse the relations between the variables one by one. I think that the number of your data is enough.
Hello, minimum observations most 20 observations for time-series analysis
Use sample size calculator formula to decide based on the confidence interval (margin of error) confidence level and size of population.
Please can my professional colleagues help me out. A reviewer has rejected my paper on the grounds that in the ARDL methodology I used, there were 32 observations ( quarterly data ranging from 2009 to 2017). He is insisting that it must be annual data of not less than 30 years. But my argument is that though the minimum number of observations must be 30, it must not be annual data. He thinks otherwise.
Please I need clarification on this.
Secondly, does ARDL not admit of monthly, quarterly, or other high frequency data in as long as the minimum sample size or observations are met?
ARDL can, of course, be applied to quarterly data, too. The problem is, that these data, in general, show a seasonal pattern. That means, that the degrees of freedom are, in fact, much lower than the number of observations suggest. Additionally, your data cover a period after the financial crisis, where the economies started from a very low level. Therefore, your data refer to a boom period only. By the way, ARDL - like other more sophisticated methods - has the disadvantage (which, in general, seen as an advantage of the econometric program packages), that it is easy to apply on any kind of data, which tempts people to less carefully analyse (theoretically) the problem they are working on.
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology
- Sectioning