Dear all,
My thesis involves five economic indicators, but two of my key variables have 10 years missing values, that is from 1986-1996. I will apply many pre-tests such as Unit root test, structural breaks tests, cointegration and so on. I am worried how to fill in the gaps. My actual data is from 1960-2015...but in between data for year 1986,87,88,89,90,91,92,93,94,95,96 is missing ? How to deal with it? I want to apply many methods/models technique. I need guidance please.... Please help with me with both statistical or econmetrics theory, and also R, Eviews and STATA commands to handle this issue. Thank you!
Dear Joe
Don't be so worry about it. According to your question you have already collected most the the data only some of years are missing. For that purpose you just find a reasonable proxy or just take the mean values.
For different estimation techniques there are lots of stuff available online. I have found some studies regarding your missing data. Hope this ll help.
http://ieeexplore.ieee.org/document/7858189/?reload=true
https://gking.harvard.edu/files/pr.pdf
https://stats.stackexchange.com/questions/121414/how-do-i-handle-nonexistent-or-missing-data
https://liberalarts.utexas.edu/prc/_files/cs/Missing-Data.pdf
Regards
Saqib
Thank you very much Mr. Shimelis Kebede and Mr. Muhammad Saqib Irshad. I will read them very well before doing anything.
Dear Joe,
Mr. Saqib has given a good answer to your question. In order to fill the missing data, first make sure the methods you want to use are not working or not giving desired results. If you are getting the desired results, no need to go for filling the missing data. However, if there are problems in the data, you may interpolate the missing data. Interpolation can be done both in STATA and EViews. Filling the missing observations with series mean or mean of nearby values and multiple imputation are other ways of doing it.
Warm Regards,
Atif
Joe,
When looking to supplement the missing data, the first thing is to see if the amount of data you are supplementing is large in comparison to the entire same and to the series. There are no strict rules (same textbook even say it is ok as long as you stay below 30% - which in my opinion is too much), so you have to use your judgment here. In my case, I try not to go above 5% for the given series.
As for the methods of supplementing the data:
1) Use data from another source - here be sure that both sources define the variable you are interested in in the same way. If not, then you cannot use this method, as at times the values of the same variable can differ significantly across sources.
2) Use the average of neighboring values - this is easy, but it works only if you have one missing value. You can also try to spread the difference between the neighboring values evenly across the periods, for which you do not have data.
3) Estimate a simple trend equation - this works, but under the assumption that there were no significant events during the period, for which the data is missing; especially if you use a linear trend.
Obviously, the best way is to find a substitute variable, which has the data available for the entire period, but I understand that this is not always possible.
All best,
TMN
Thank you Mr. Muhammad Atif Nawaz and Mr. Tomasz Marcin Napiórkowski for your kind suggestions. I will try to read extensively before applying the proposed methods.
YES I AGREE WITH TOMASZ BUT BE CAREFUL. WE USE LINEAR INTERPOLATION ONLY IF DATA DONT HAVE A HIGH VOLATILITY
- Badges
- Science topic