Dear all,

I am currently working with unbalanced panel data which I need to prepare for regressions. Since only yearly data is available, I have to interpolate the values for some variables, such as loans, to get quaterly data. The observations are aggregated to groups with a unique group-ID (newid) and the data is formated as a time series (tsset).

Since I have to deal with unbalanced panel data, some years in the time series are missing. I worry about missing years somewhere in the middle of the sequence, e.g. if 2004 and 2006 are available, but 2005 not. Therefore, my questions are:

1. After using tsspell, I have seen that some groups have several spells (up to 3). Is it possible to incorporate this information in the ipolate command, e.g. by creating new subgroup IDs? I thought that they could refer to one sequence before a spell occurs. How should I implement it technically?

2. Is there a common approach in such a case?


So far, I have been using:

by newid: ipolate loans dateq, generate (loans2)


Data example with one (three) spells within different groups:

Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input float(newid dateq) byte(_spell _seq _end) float loans2

395 176 1 1 0     .
395 177 1 2 0     .
395 178 1 3 0     .
395 179 1 4 0 15576
395 180 1 5 0     .
.
.
.
396 204 3  1 0   .
396 205 3  2 0   .
396 206 3  3 0   .
396 207 3  4 0 605
396 208 3  5 0   .
396 209 3  6 0   .
396 210 3  7 0   .
396 211 3  8 0 511 .

end
format %tq dateq



I am using Stata 15.0

Thank you a lot in advance!
Eva