My variable of interest is Trade in services(As % of GDP)...i need continuous time series from 1970-2010...but in between data for year 1994,95,96,97 is missing ? how to deal with it? I want to apply ARDL technique so need at least 30 yrs continuous data....Please guide...
First, I suggest to apply some easy interpolation techniques (see for example: spline interpolation). The quality of the result strongly depends on the characteristics of your time series. For your kind of values it could be interesting to apply some kind of fractal interpolation (see for example: self affine FIF).
In traditional econometric analysis, researchers used to just omit those year. For example when the time series include war year, those years are dropped. But you seem wanting to have a replacement for those years because of the sample size. I remember old softwares used to accommodate that by asking if you would like to include the median or some average for those missing years. Such average or other interpolation will do the job, but may affect statistical significant values that are based on the sample size.
Thanks Elisa and Lall Sir for taking time and answering my question. Sir, I used tsfill in stata to fill those missing values but i'm not sure how this kind of replacement will affect the statistical significance of my results....
Dear Rashmi,
I googled tsfill and noticed that stata interpolate values for the missing years. In that case, I will do the procedure twice. Once with the tsfill values and another without those missing years observation. If the statistics are the same then you can report that the tsfill procedure is neutral. Otherwise you can just state the difference. One way to view the results is that the required sample size (of 30) is only a semi guarantee that your results will be statistically significant. If you can achieve significance in the neighborhood of that required sample size of 30 by the missing data approach, then you may not draw criticism to the effect that your significance is attained by a fluke.
okies..Thank you so much sir for your help...I'll do it as you suggested....
I have similar problem. My time series data is recorded with 10 seconds intervals, and data for two months that are not consecutive ( July and November) out of 14 months are missing,(around 518000 records (15%) are missing). what is the best way to deal with this issue? how should I choose my training and test sets?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series